back to indexPeter Wang: Python and the Source Code of Humans, Computers, and Reality | Lex Fridman Podcast #250
link |
The following is a conversation with Peter Wang,
link |
one of the most impactful leaders and developers
link |
in the Python community.
link |
Former physicist, current philosopher,
link |
and someone who many people told me about
link |
and praised as a truly special mind
link |
that I absolutely should talk to.
link |
Recommendations ranging from Travis Hallifont
link |
to Eric Weinstein.
link |
This is the Lex Friedman podcast.
link |
To support it, please check out our sponsors
link |
in the description.
link |
And now, here's my conversation with Peter Wang.
link |
You're one of the most impactful humans
link |
in the Python ecosystem.
link |
So, you're an engineer, leader of engineers,
link |
but you're also a philosopher.
link |
So, let's talk both in this conversation
link |
about programming and philosophy.
link |
First, programming.
link |
What to you is the best
link |
or maybe the most beautiful feature of Python?
link |
Or maybe the thing that made you fall in love
link |
or stay in love with Python?
link |
Well, those are three different things.
link |
What I think is the most beautiful,
link |
what made me fall in love, what made me stay in love.
link |
When I first started using it
link |
was when I was a C++ computer graphics performance nerd.
link |
Yeah, in the late 90s.
link |
And that was my first job out of college.
link |
And we kept trying to do more and more abstract
link |
and higher order programming in C++,
link |
which at the time was quite difficult.
link |
With templates, the compiler support wasn't great, et cetera.
link |
So, when I started playing around with Python,
link |
that was my first time encountering
link |
really first class support for types, for functions,
link |
and things like that.
link |
And it felt so incredibly expressive.
link |
So, that was what kind of made me fall in love
link |
with it a little bit.
link |
And also, once you spend a lot of time
link |
in a C++ dev environment,
link |
the ability to just whip something together
link |
that basically runs and works the first time is amazing.
link |
So, really productive scripting language.
link |
I mean, I knew Perl, I knew Bash, I was decent at both.
link |
But Python just made everything,
link |
it made the whole world accessible.
link |
I could script this and that and the other,
link |
network things, little hard drive utilities.
link |
I could write all of these things
link |
in the space of an afternoon.
link |
And that was really, really cool.
link |
So, that's what made me fall in love.
link |
Is there something specific you could put your finger on
link |
that you're not programming in Perl today?
link |
Like, why Python for scripting?
link |
I think there's not a specific thing
link |
as much as the design motif of both the creator
link |
of the language and the core group of people
link |
that built the standard library around him.
link |
There was definitely, there was a taste to it.
link |
I mean, Steve Jobs used that term
link |
in somewhat of an arrogant way,
link |
but I think it's a real thing,
link |
that it was designed to fit.
link |
A friend of mine actually expressed this really well.
link |
He said, Python just fits in my head.
link |
And there's nothing better to say than that.
link |
Now, people might argue modern Python,
link |
there's a lot more complexity,
link |
but certainly as version 1.5.2,
link |
I think was my first version,
link |
that fit in my head very easily.
link |
So, that's what made me fall in love with it.
link |
Okay, so the most beautiful feature of Python
link |
that made you stay in love.
link |
It's like over the years, what has like,
link |
you do a double take and you return too often
link |
as a thing that just brings you a smile.
link |
I really still like the ability to play with meta classes
link |
and express higher order of things.
link |
When I have to create some new object model
link |
to model something, right?
link |
cause I'm pretty expert as a Python programmer.
link |
I can easily put all sorts of lovely things together
link |
and use properties and decorators and other kinds of things
link |
and create something that feels very nice.
link |
So, that to me, I would say that's tied
link |
with the NumPy and vectorization capabilities.
link |
I love thinking in terms of the matrices and the vectors
link |
and these kind of data structures.
link |
So, I would say those two are kind of tied for me.
link |
So, the elegance of the NumPy data structure,
link |
like slicing through the different multi dimensional.
link |
Yeah, there's just enough things there.
link |
It's like a very, it's a very simple, comfortable tool.
link |
Just, it's easy to reason about what it does
link |
when you don't stray too far afield.
link |
Can you put your finger on how to design a language
link |
such that it fits in your head?
link |
Certain things like the colon
link |
or the certain notation aspects of Python
link |
that just kind of work.
link |
Is it something you have to kind of write out on paper,
link |
look and say, it's just right?
link |
Is it a taste thing or is there a systematic process?
link |
What's your sense?
link |
I think it's more of a taste thing.
link |
But one thing that should be said
link |
is that you have to pick your audience, right?
link |
So, the better defined the user audience is
link |
or the users are, the easier it is to build something
link |
that fits in their minds because their needs
link |
will be more compact and coherent.
link |
It is possible to find a projection, right?
link |
A compact projection for their needs.
link |
The more diverse the user base, the harder that is.
link |
And so, as Python has grown in popularity,
link |
that's also naturally created more complexity
link |
as people try to design any given thing.
link |
There'll be multiple valid opinions
link |
about a particular design approach.
link |
And so, I do think that's the downside of popularity.
link |
It's almost an intrinsic aspect
link |
of the complexity of the problem.
link |
Well, at the very beginning,
link |
aren't you an audience of one, isn't ultimately,
link |
aren't all the greatest projects in history
link |
were just solving a problem that you yourself had?
link |
Well, so Clay Shirky in his book on crowdsourcing
link |
or his kind of thoughts on crowdsourcing,
link |
he identifies the first step of crowdsourcing
link |
is me first collaboration.
link |
You first have to make something
link |
that works well for yourself.
link |
It's very telling that when you look at all of the impactful
link |
big project, well, they're fundamental projects now
link |
in the SciPy and Pydata ecosystem.
link |
They all started with the people in the domain
link |
trying to scratch their own itch.
link |
And the whole idea of scratching your own itch
link |
is something that the open source
link |
or the free software world has known for a long time.
link |
But in the scientific computing areas,
link |
these are assistant professors
link |
or electrical engineering grad students.
link |
They didn't have really a lot of programming skill
link |
necessarily, but Python was just good enough
link |
for them to put something together
link |
that fit in their domain, right?
link |
So it's almost like a,
link |
it's a necessity is the mother of invention aspect.
link |
And also it was a really harsh filter
link |
for utility and compactness and expressiveness.
link |
Like it was too hard to use,
link |
then they wouldn't have built it
link |
because that was just too much trouble, right?
link |
It was a side project for them.
link |
And also necessity creates a kind of deadline.
link |
It seems like a lot of these projects
link |
are quickly thrown together in the first step.
link |
And that, even though it's flawed,
link |
that just seems to work well for software projects.
link |
Well, it does work well for software projects in general.
link |
And in this particular space,
link |
one of my colleagues, Stan Siebert identified this,
link |
that all the projects in the SciPy ecosystem,
link |
if we just rattle them off,
link |
there's NumPy, there's SciPy
link |
built by different collaborations of people.
link |
Although Travis is the heart of both of them.
link |
But NumPy coming from numeric and numery,
link |
these are different people.
link |
And then you've got Pandas,
link |
you've got Jupyter or IPython,
link |
there's Matplotlib,
link |
there's just so many others, I'm not gonna do justice
link |
if I try to name them all.
link |
But all of them are actually different people.
link |
And as they rolled out their projects,
link |
the fact that they had limited resources
link |
meant that they were humble about scope.
link |
A great famous hacker, Jamie Zawisky,
link |
once said that every geek's dream
link |
is to build the ultimate middleware, right?
link |
And the thing is with these scientists turned programmers,
link |
they had no such dream.
link |
They were just trying to write something
link |
that was a little bit better for what they needed,
link |
and they were gonna leverage what everyone else had built.
link |
So naturally, almost in kind of this annealing process
link |
or whatever, we built a very modular cover
link |
of the basic needs of a scientific computing library.
link |
If you look at the whole human story,
link |
how much of a leap is it?
link |
We've developed all kinds of languages,
link |
all kinds of methodologies for communication.
link |
It just kind of like grew this collective intelligence,
link |
civilization grew, it expanded, wrote a bunch of books,
link |
and now we tweet how big of a leap is programming
link |
if programming is yet another language?
link |
Is it just a nice little trick
link |
that's temporary in our human history,
link |
or is it like a big leap in the,
link |
almost us becoming another organism
link |
at a higher level of abstraction, something else?
link |
I think the act of programming
link |
or using grammatical constructions
link |
of some underlying primitives,
link |
that is something that humans do learn,
link |
but every human learns this.
link |
Anyone who can speak learns how to do this.
link |
What makes programming different
link |
has been that up to this point,
link |
when we try to give instructions to computing systems,
link |
all of our computers, well, actually this is not quite true,
link |
but I'll first say it,
link |
and then I'll tell you why it's not true.
link |
But for the most part,
link |
we can think of computers as being these iterated systems.
link |
So when we program,
link |
we're giving very precise instructions to iterated systems
link |
that then run at incomprehensible speed
link |
and run those instructions.
link |
some people are just better equipped
link |
to model systematic iterated systems,
link |
well, whatever, iterated systems in their head.
link |
Some people are really good at that,
link |
and other people are not.
link |
And so when you have like, for instance,
link |
sometimes people have tried to build systems
link |
that make programming easier by making it visual,
link |
And the issue is you can have a drag and drop thing,
link |
but once you start having to iterate the system
link |
with conditional logic,
link |
handling case statements and branch statements
link |
and all these other things,
link |
the visual drag and drop part doesn't save you anything.
link |
You still have to reason about this giant iterated system
link |
with all these different conditions around it.
link |
That's the hard part, right?
link |
So handling iterated logic, that's the hard part.
link |
The languages we use then emerge
link |
to give us the ability and capability over these things.
link |
Now, the one exception to this rule, of course,
link |
is the most popular programming system in the world,
link |
which is Excel, which is a data flow
link |
and a data driven, immediate mode,
link |
data transformation oriented programming system.
link |
And this actually not an accident
link |
that that system is the most popular programming system
link |
because it's so accessible
link |
to a much broader group of people.
link |
I do think as we build future computing systems,
link |
you're actually already seeing this a little bit,
link |
it's much more about composition of modular blocks.
link |
They themselves actually maintain all their internal state
link |
and the interfaces between them
link |
are well defined data schemas.
link |
And so to stitch these things together using like IFTTT
link |
or Zapier or any of these kind of,
link |
I would say compositional scripting kinds of things,
link |
I mean, HyperCard was also a little bit in this vein.
link |
That's much more accessible to most people.
link |
It's really that implicit state
link |
that's so hard for people to track.
link |
Yeah, okay, so that's modular stuff,
link |
but there's also an aspect
link |
where you're standing on the shoulders of giants.
link |
So you're building like higher and higher levels
link |
of abstraction, but you do that a little bit with language.
link |
So with language, you develop sort of ideas,
link |
philosophies from Plato and so on.
link |
And then you kind of leverage those philosophies
link |
as you try to have deeper and deeper conversations.
link |
But with programming,
link |
it seems like you can build much more complicated systems.
link |
Like without knowing how everything works,
link |
you can build on top of the work of others.
link |
And it seems like you're developing
link |
more and more sophisticated expressions,
link |
ability to express ideas in a computational space.
link |
I think it's worth pondering the difference here
link |
between complexity and complication.
link |
Okay, right. Back to Excel.
link |
Well, not quite back to Excel,
link |
but the idea is when we have a human conversation,
link |
all languages for humans emerged
link |
to support human relational communications,
link |
which is that the person we're communicating with
link |
is a person and they would communicate back to us.
link |
And so we sort of hit a resonance point, right?
link |
When we actually agree on some concepts.
link |
So there's a messiness to it and there's a fluidity to it.
link |
With computing systems,
link |
when we express something to the computer and it's wrong,
link |
we just try again.
link |
So we can basically live many virtual worlds
link |
of having failed at expressing ourselves to the computer
link |
until the one time we expressed ourselves right.
link |
Then we kind of put in production
link |
and then discover that it's still wrong
link |
a few days down the road.
link |
So I think the sophistication of things
link |
that we build with computing,
link |
one has to really pay attention to the difference
link |
between when an end user is expressing something
link |
onto a system that exists
link |
versus when they're extending the system
link |
to increase the system's capability
link |
for someone else to then interface with.
link |
We happen to use the same language for both of those things
link |
in most cases, but it doesn't have to be that.
link |
And Excel is actually a great example of this,
link |
of kind of a counterpoint to that.
link |
Okay, so what about the idea of, you said messiness.
link |
Wouldn't you put the software 2.0 idea,
link |
this idea of machine learning
link |
into the further and further steps
link |
into the world of messiness.
link |
The same kind of beautiful messiness of human communication.
link |
Isn't that what machine learning is?
link |
Is building on levels of abstraction
link |
that don't have messiness in them,
link |
that at the operating system level,
link |
then there's Python, the programming languages
link |
that have more and more power.
link |
But then finally, there's neural networks
link |
that ultimately work with data.
link |
And so the programming is almost in the space of data
link |
and the data is allowed to be messy.
link |
Isn't that a kind of program?
link |
So the idea of software 2.0 is a lot of the programming
link |
happens in the space of data, so back to Excel,
link |
all roads lead back to Excel, in the space of data
link |
and also the hyperparameters of the neural networks.
link |
And all of those allow the same kind of messiness
link |
that human communication allows.
link |
It does, but my background is in physics.
link |
I took like two CS courses in college.
link |
So I don't have, now I did cram a bunch of CS in prep
link |
when I applied for grad school,
link |
but still I don't have a formal background
link |
in computer science.
link |
But what I have observed in studying programming languages
link |
and programming systems and things like that
link |
is that there seems to be this triangle.
link |
It's one of these beautiful little iron triangles
link |
that you find in life sometimes.
link |
And it's the connection between the code correctness
link |
and kind of expressiveness of code,
link |
the semantics of the data,
link |
and then the kind of correctness or parameters
link |
of the underlying hardware compute system.
link |
So there's the algorithms that you wanna apply,
link |
there's what the bits that are stored on whatever media
link |
actually represent, so the semantics of the data
link |
within the representation,
link |
and then there's what the computer can actually do.
link |
And every programming system, every information system
link |
ultimately finds some spot in the middle
link |
of this little triangle.
link |
Sometimes some systems collapse them into just one edge.
link |
Are we including humans as a system in this?
link |
No, no, I'm just thinking about computing systems here.
link |
And the reason I bring this up is because
link |
I believe there's no free lunch around this stuff.
link |
So if we build machine learning systems
link |
to sort of write the correct code
link |
that is at a certain level of performance,
link |
so it'll sort of select with hyperparameters
link |
we can tune kind of how we want the performance boundary
link |
in SLA to look like for transforming some set of inputs
link |
into certain kinds of outputs.
link |
That training process itself is intrinsically sensitive
link |
to the kinds of inputs we put into it.
link |
It's quite sensitive to the boundary conditions
link |
we put around the performance.
link |
So I think even as we move to using automated systems
link |
to build this transformation,
link |
as opposed to humans explicitly
link |
from a top down perspective, figuring out,
link |
well, this schema and this database and these columns
link |
get selected for this algorithm,
link |
and here we put a Fibonacci heap for some other thing.
link |
Human design or computer design,
link |
ultimately what we hit,
link |
the boundaries that we hit with these information systems
link |
is when the representation of the data hits the real world
link |
is where there's a lot of slop and a lot of interpretation.
link |
And that's where actually I think
link |
a lot of the work will go in the future
link |
is actually understanding kind of how to better
link |
in the view of these live data systems,
link |
how to better encode the semantics of the world
link |
There'll be less of the details
link |
of how we write a particular SQL query.
link |
Okay, but given the semantics of the real world
link |
and the messiness of that,
link |
what does the word correctness mean
link |
when you're talking about code?
link |
There's a lot of dimensions to correctness.
link |
Historically, and this is one of the reasons I say
link |
that we're coming to the end of the era of software,
link |
because for the last 40 years or so,
link |
software correctness was really defined
link |
about functional correctness.
link |
I write a function, it's got some inputs,
link |
does it produce the right outputs?
link |
If so, then I can turn it on,
link |
hook it up to the live database and it goes.
link |
And more and more now we have,
link |
I mean, in fact, I think the bright line in the sand
link |
between machine learning systems
link |
or modern data driven systems
link |
versus classical software systems
link |
is that the values of the input
link |
actually have to be considered with the function together
link |
to say this whole thing is correct or not.
link |
And usually there's a performance SLA as well.
link |
Like did it actually finish making this?
link |
Sorry, service level agreement.
link |
So it has to return within some time.
link |
You have a 10 millisecond time budget
link |
to return a prediction of this level of accuracy, right?
link |
So these are things that were not traditionally
link |
in most business computing systems for the last 20 years
link |
at all, people didn't think about it.
link |
But now we have value dependence on functional correctness.
link |
So that question of correctness
link |
is becoming a bigger and bigger question.
link |
What does that map to the end of software?
link |
We've thought about software as just this thing
link |
that you can do in isolation with some test trial inputs
link |
and in a very sort of sandboxed environment.
link |
And we can quantify how does it scale?
link |
How does it perform?
link |
How many nodes do we need to allocate
link |
if we wanna scale this many inputs?
link |
When we start turning this stuff into prediction systems,
link |
real cybernetic systems,
link |
you're going to find scenarios where you get inputs
link |
that you're gonna wanna spend
link |
a little more time thinking about.
link |
You're gonna find inputs that are not,
link |
it's not clear what you should do, right?
link |
So then the software has a varying amount of runtime
link |
and correctness with regard to input.
link |
And that is a different kind of system altogether.
link |
Now it's a full on cybernetic system.
link |
It's a next generation information system
link |
that is not like traditional software systems.
link |
Can you maybe describe what is a cybernetic system?
link |
Do you include humans in that picture?
link |
So is a human in the loop kind of complex mess
link |
of the whole kind of interactivity of software
link |
with the real world or is it something more concrete?
link |
Well, when I say cybernetic,
link |
I really do mean that the software itself
link |
is closing the observe, orient, decide, act loop by itself.
link |
So humans being out of the loop is the fact
link |
what for me makes it a cybernetic system.
link |
And humans are out of that loop.
link |
When humans are out of the loop,
link |
when the machine is actually sort of deciding on its own
link |
what it should do next to get more information,
link |
that makes it a cybernetic system.
link |
So we're just at the dawn of this, right?
link |
I think everyone talking about MLAI, it's great.
link |
But really the thing we should be talking about
link |
is when we really enter the cybernetic era
link |
and all of the questions of ethics and governance
link |
and all correctness and all these things,
link |
they really are the most important questions.
link |
Okay, can we just linger on this?
link |
What does it mean for the human to be out of the loop
link |
in a cybernetic system, because isn't the cybernetic system
link |
that's ultimately accomplishing some kind of purpose
link |
that at the bottom, the turtles all the way down,
link |
at the bottom turtle is a human.
link |
Well, the human may have set some criteria,
link |
but the human wasn't precise.
link |
So for instance, I just read the other day
link |
that earlier this year,
link |
or maybe it was last year at some point,
link |
the Libyan army, I think,
link |
sent out some automated killer drones with explosives.
link |
And there was no human in the loop at that point.
link |
They basically put them in a geofenced area,
link |
said find any moving target, like a truck or vehicle
link |
that looks like this, and boom.
link |
That's not a human in the loop, right?
link |
So increasingly, the less human there is in the loop,
link |
the more concerned you are about these kinds of systems,
link |
because there's unintended consequences,
link |
like less the original designer and engineer of the system
link |
is able to predict, even one with good intent
link |
is able to predict the consequences of such a system.
link |
Is that it? That's right.
link |
There are some software systems, right,
link |
that run without humans in the loop
link |
that are quite complex.
link |
And that's like the electronic markets.
link |
And we get flash crashes all the time.
link |
We get in the heyday of high frequency trading,
link |
there's a lot of market microstructure,
link |
people doing all sorts of weird stuff
link |
that the market designers had never really thought about,
link |
contemplated or intended.
link |
So when we run these full on systems
link |
with these automated trading bots,
link |
now they become automated killer drones
link |
and then all sorts of other stuff.
link |
We are, that's what I mean by we're at the dawn
link |
of the cybernetic era and the end of the era
link |
of just pure software.
link |
Are you more concerned,
link |
if you're thinking about cybernetic systems
link |
or even like self replicating systems,
link |
so systems that aren't just doing a particular task,
link |
but are able to sort of multiply and scale
link |
in some dimension in the digital
link |
or even the physical world.
link |
Are you more concerned about like the lobster being boiled?
link |
So a gradual with us not noticing,
link |
collapse of civilization or a big explosion,
link |
like oops, kind of a big thing where everyone notices,
link |
but it's too late.
link |
I think that it will be a different experience
link |
for different people.
link |
I do share a common point of view
link |
with some of the climate,
link |
people who are concerned about climate change
link |
and just the big existential risks that we have.
link |
But unlike a lot of people who share my level of concern,
link |
I think the collapse will not be quite so dramatic
link |
as some of them think.
link |
And what I mean is that,
link |
I think that for certain tiers of let's say economic class
link |
or certain locations in the world,
link |
people will experience dramatic collapse scenarios.
link |
But for a lot of people, especially in the developed world,
link |
the realities of collapse will be managed.
link |
There'll be narrative management around it
link |
so that they essentially insulate,
link |
the middle class will be used to insulate the upper class
link |
from the pitchforks and the flaming torches and everything.
link |
It's interesting because,
link |
so my specific question wasn't as general.
link |
My question was more about cybernetic systems or software.
link |
but it would nevertheless perhaps be about class.
link |
So the effect of algorithms
link |
might affect certain classes more than others.
link |
I was more thinking about
link |
whether it's social media algorithms or actual robots,
link |
is there going to be a gradual effect on us
link |
where we wake up one day
link |
and don't recognize the humans we are,
link |
or is it something truly dramatic
link |
where there's like a meltdown of a nuclear reactor
link |
kind of thing, Chernobyl, like catastrophic events
link |
that are almost bugs in a program that scaled itself
link |
Yeah, I'm not as concerned about the visible stuff.
link |
And the reason is because the big visible explosions,
link |
I mean, this is something I said about social media
link |
is that at least with nuclear weapons,
link |
when a nuke goes off, you can see it
link |
and you're like, well, that's really,
link |
wow, that's kind of bad, right?
link |
I mean, Oppenheimer was reciting the Bhagavad Gita, right?
link |
When he saw one of those things go off.
link |
So we can see nukes are really bad.
link |
He's not reciting anything about Twitter.
link |
Well, but right, but then when you have social media,
link |
when you have all these different things that conspire
link |
to create a layer of virtual experience for people
link |
that alienates them from reality and from each other,
link |
that's very pernicious, that's impossible to see, right?
link |
And it kind of slowly gets in there, so.
link |
You've written about this idea of virtuality
link |
on this topic, which you define as the subjective phenomenon
link |
of knowingly engaging with virtual sensation and perception
link |
and suspending or forgetting the context
link |
that it's simulacrum.
link |
So let me ask, what is real?
link |
Is there a hard line between reality and virtuality?
link |
Like perception drifts from some kind of physical reality.
link |
We have to kind of have a sense of what is the line
link |
that's too, we've gone too far.
link |
For me, it's not about any hard line about physical reality
link |
as much as a simple question of,
link |
does the particular technology help people connect
link |
in a more integral way with other people,
link |
with their environment,
link |
with all of the full spectrum of things around them?
link |
So it's less about, oh, this is a virtual thing
link |
and this is a hard real thing,
link |
more about when we create virtual representations
link |
of the real things, always some things
link |
are lost in translation.
link |
Usually many, many dimensions are lost in translation.
link |
We're now coming to almost two years of COVID,
link |
people on Zoom all the time.
link |
You know it's different when you meet somebody in person
link |
than when you see them on,
link |
I've seen you on YouTube lots, right?
link |
But then seeing a person is very different.
link |
And so I think when we engage in virtual experiences
link |
all the time, and we only do that,
link |
there is absolutely a level of embodiment.
link |
There's a level of embodied experience
link |
and participatory interaction that is lost.
link |
And it's very hard to put your finger on exactly what it is.
link |
It's hard to say, oh, we're gonna spend $100 million
link |
building a new system that captures this 5% better,
link |
higher fidelity human expression.
link |
No one's gonna pay for that, right?
link |
So when we rush madly into a world of simulacrum
link |
and virtuality, the things that are lost are,
link |
Once everyone moves there, it can be hard to look back
link |
and see what we've lost.
link |
So is it irrecoverably lost?
link |
Or rather, when you put it all on the table,
link |
is it possible for more to be gained than is lost?
link |
If you look at video games,
link |
they create virtual experiences that are surreal
link |
and can bring joy to a lot of people,
link |
can connect a lot of people,
link |
and can get people to talk a lot of trash.
link |
So they can bring out the best and the worst in people.
link |
So is it possible to have a future world
link |
where the pros outweigh the cons?
link |
I mean, it's possible to have that in the current world.
link |
But when literally trillions of dollars of capital
link |
are tied to using those things
link |
to groom the worst of our inclinations
link |
and to attack our weaknesses in the limbic system
link |
to create these things into id machines
link |
versus connection machines,
link |
then those good things don't stand a chance.
link |
Can you make a lot of money by building connection machines?
link |
Is it possible, do you think,
link |
to bring out the best in human nature
link |
to create fulfilling connections and relationships
link |
in the digital world and make a shit ton of money?
link |
If I figure it out, I'll let you know.
link |
But what's your intuition
link |
without concretely knowing what's the solution?
link |
My intuition is that a lot of our digital technologies
link |
give us the ability to have synthetic connections
link |
or to experience virtuality.
link |
They have co evolved with sort of the human expectations.
link |
It's sort of like sugary drinks.
link |
As people have more sugary drinks,
link |
they need more sugary drinks to get that same hit, right?
link |
So with these virtual things and with TV and fast cuts
link |
and TikToks and all these different kinds of things,
link |
we're co creating essentially humanity
link |
that sort of asks and needs those things.
link |
And now it becomes very difficult
link |
to get people to slow down.
link |
It gets difficult for people to hold their attention
link |
on slow things and actually feel that embodied experience.
link |
So mindfulness now more than ever is so important in schools
link |
and as a therapy technique for people
link |
because our environment has been accelerated.
link |
And McLuhan actually talks about this
link |
in the electric environment of the television.
link |
And that was before TikTok and before front facing cameras.
link |
So I think for me, the concern is that
link |
it's not like we can ever switch to doing something better,
link |
but more of the humans and technology,
link |
they're not independent of each other.
link |
The technology that we use kind of molds what we need
link |
for the next generation of technology.
link |
Yeah, but humans are intelligent and they're introspective
link |
and they can reflect on the experiences of their life.
link |
So for example, there's been many years in my life
link |
where I ate an excessive amount of sugar.
link |
And then a certain moment I woke up and said,
link |
why do I keep doing this?
link |
This doesn't feel good.
link |
And I think, so going through the TikTok process
link |
of realizing, okay, when I shorten my attention span,
link |
actually that does not make me feel good longer term.
link |
And realizing that and then going to platforms,
link |
going to places that are away from the sugar.
link |
So in so doing, you can create platforms
link |
that can make a lot of money to help people wake up
link |
to what actually makes them feel good longterm.
link |
Develop, grow as human beings.
link |
And it just feels like humans are more intelligent
link |
than mice looking for cheese.
link |
They're able to sort of think, I mean,
link |
we can contemplate our own mortality.
link |
We can contemplate things like longterm love
link |
and we can have a longterm fear
link |
of certain things like mortality.
link |
We can contemplate whether the experiences,
link |
the sort of the drugs of daily life
link |
that we've been partaking in is making us happier,
link |
And then once we contemplate that,
link |
we can make financial decisions in using services
link |
and paying for services that are making us better people.
link |
So it just seems that we're in the very first stages
link |
of social networks that just were able to make a lot of money
link |
really quickly, but in bringing out sometimes
link |
the bad parts of human nature, they didn't destroy humans.
link |
They just fed everybody a lot of sugar.
link |
And now everyone's gonna wake up and say,
link |
hey, we're gonna start having like sugar free social media.
link |
Well, there's a lot to unpack there.
link |
I think some people certainly have the capacity for that.
link |
And I certainly think, I mean, it's very interesting
link |
even the way you said it, you woke up one day
link |
and you thought, well, this doesn't feel very good.
link |
Well, it's still your limbic system saying
link |
this doesn't feel very good, right?
link |
You have a cat brain's worth of neurons around your gut,
link |
And so maybe that saturated and that was telling you,
link |
hey, this isn't good.
link |
Humans are more than just mice looking for cheese
link |
or monkeys looking for sex and power, right?
link |
Now a lot of people would argue with you on that one,
link |
Well, we're more than just that, but we're at least that.
link |
And we're very, very seldom not that.
link |
So I don't actually disagree with you
link |
that we could be better and that better platforms exist.
link |
And people are voluntarily noping out of things
link |
like Facebook and noping out.
link |
That's an awesome verb.
link |
It's a great term.
link |
I use it all the time.
link |
You're welcome, Mike.
link |
I'm gonna have to nope out of that.
link |
I'm gonna have to nope out of that, right?
link |
It's gonna be a hard pass and that's great.
link |
But that's again, to your point,
link |
that's the first generation of front facing cameras
link |
of social pressures.
link |
And you as a self starter, self aware adult
link |
have the capacity to say, yeah, I'm not gonna do that.
link |
I'm gonna go and spend time on long form reads.
link |
I'm gonna spend time managing my attention.
link |
I'm gonna do some yoga.
link |
If you're a 15 year old in high school
link |
and your entire social environment
link |
is everyone doing these things,
link |
guess what you're gonna do?
link |
You're gonna kind of have to do that
link |
because your limbic system says,
link |
hey, I need to get the guy or the girl or the whatever.
link |
And that's what I'm gonna do.
link |
And so one of the things that we have to reason about here
link |
is the social media systems or social media,
link |
I think is our first encounter with a technological system
link |
that runs a bit of a loop around our own cognition
link |
It's not the last, it's far from the last.
link |
And it gets to the heart of some of the philosophical
link |
Achilles heel of the Western philosophical system,
link |
which is each person gets to make their own determination.
link |
Each person is an individual that's sacrosanct
link |
in their agency and their sovereignty and all these things.
link |
The problem with these systems is they come down
link |
and they are able to make their own decisions.
link |
They come down and they are able to manage everyone on mass.
link |
And so every person is making their own decision,
link |
but together the bigger system is causing them to act
link |
with a group dynamic that's very profitable for people.
link |
So this is the issue that we have is that our philosophies
link |
are actually not geared to understand
link |
what is it for a person to have a high trust connection
link |
as part of a collective and for that collective
link |
to have its right to coherency and agency.
link |
That's something like when a social media app
link |
causes a family to break apart,
link |
it's done harm to more than just individuals, right?
link |
So that concept is not something we really talk about
link |
or think about very much, but that's actually the problem
link |
is that we're vaporizing molecules into atomic units
link |
and then we're hitting all the atoms with certain things.
link |
That's like, yeah, well, that person chose to look at my app.
link |
So our understanding of human nature
link |
at the individual level, it emphasizes the individual
link |
too much because ultimately society operates
link |
at the collective level.
link |
And these apps do as well.
link |
And the apps do as well.
link |
So for us to understand the progression and the development
link |
of this organism we call human civilization,
link |
we have to think at the collective level too.
link |
I would say multi tiered.
link |
So individual as well.
link |
Individuals, family units, social collectives
link |
and all the way up.
link |
Okay, so you've said that individual humans
link |
are multi layered susceptible to signals and waves
link |
and multiple strata, the physical, the biological,
link |
social, cultural, intellectual.
link |
So sort of going along these lines,
link |
can you describe the layers of the cake
link |
that is a human being and maybe the human collective,
link |
So I'm just stealing wholesale here from Robert Persig,
link |
who is the author of Zen and the Art of Motorcycle
link |
Maintenance and his follow on book has a sequel to it
link |
He goes into this in a little more detail.
link |
But it's a crude approach to thinking about people.
link |
But I think it's still an advancement
link |
over traditional subject object metaphysics,
link |
where we look at people as a dualist would say,
link |
well, is your mind, your consciousness,
link |
is that just merely the matter that's in your brain
link |
or is there something kind of more beyond that?
link |
And they would say, yes, there's a soul,
link |
sort of ineffable soul beyond just merely the physical body.
link |
And I'm not one of those people.
link |
I think that we don't have to draw a line between are things
link |
only this or only that.
link |
Collectives of things can emerge structures and patterns
link |
that are just as real as the underlying pieces.
link |
But they're transcendent, but they're still
link |
of the underlying pieces.
link |
So your body is this way.
link |
I mean, we just know physically you consist of atoms
link |
And then the atoms are arranged into molecules
link |
which then arrange into certain kinds of structures
link |
that seem to have a homeostasis to them.
link |
We call them cells.
link |
And those cells form sort of biological structures.
link |
Those biological structures give your body
link |
its physical ability and the biological ability
link |
to consume energy and to maintain homeostasis.
link |
But humans are social animals.
link |
I mean, human by themselves is not very long for the world.
link |
So part of our biology is why are two connect to other people?
link |
From the mirror neurons to our language centers
link |
and all these other things.
link |
So we are intrinsically, there's a layer,
link |
there's a part of us that wants to be part of a thing.
link |
If we're around other people, not saying a word,
link |
but they're just up and down jumping and dancing, laughing,
link |
we're going to feel better.
link |
And there was no exchange of physical anything.
link |
They didn't give us like five atoms of happiness.
link |
But there's an induction in our own sense of self
link |
that is at that social level.
link |
And then beyond that, Persick puts the intellectual level
link |
kind of one level higher than social.
link |
I think they're actually more intertwined than that.
link |
But the intellectual level is the level of pure ideas.
link |
That you are a vessel for memes.
link |
You're a vessel for philosophies.
link |
You will conduct yourself in a particular way.
link |
I mean, I think part of this is if we think about it
link |
from a physics perspective, you're not,
link |
there's the joke that physicists like to approximate things.
link |
And we'll say, well, approximate a spherical cow, right?
link |
You're not a spherical cow, you're not a spherical human.
link |
You're a messy human.
link |
And we can't even say what the dynamics of your emotion
link |
will be unless we analyze all four of these layers, right?
link |
If you're Muslim at a certain time of day, guess what?
link |
You're going to be on the ground kneeling and praying, right?
link |
And that has nothing to do with your biological need
link |
to get on the ground or physics of gravity.
link |
It is an intellectual drive that you have.
link |
It's a cultural phenomenon
link |
and an intellectual belief that you carry.
link |
So that's what the four layered stack is all about.
link |
It's that a person is not only one of these things,
link |
they're all of these things at the same time.
link |
It's a superposition of dynamics that run through us
link |
that make us who we are.
link |
So no layer is special.
link |
Not so much no layer is special,
link |
each layer is just different.
link |
Each layer gets the participation trophy.
link |
Yeah, each layer is a part of what you are.
link |
You are a layer cake, right, of all these things.
link |
And if we try to deny, right,
link |
so many philosophies do try to deny
link |
the reality of some of these things, right?
link |
Some people will say, well, we're only atoms.
link |
Well, we're not only atoms
link |
because there's a lot of other things that are only atoms.
link |
I can reduce a human being to a bunch of soup
link |
and they're not the same thing,
link |
even though it's the same atoms.
link |
So I think the order and the patterns
link |
that emerge within humans to understand,
link |
to really think about what a next generation philosophy
link |
would look like, that would allow us to reason
link |
about extending humans into the digital realm
link |
or to interact with autonomous intelligences
link |
that are not biological in nature.
link |
We really need to appreciate these,
link |
that human, what human beings actually are
link |
is the superposition of these different layers.
link |
You mentioned consciousness.
link |
Are each of these layers of cake conscious?
link |
Is consciousness a particular quality of one of the layers?
link |
Is there like a spike if you have a consciousness detector
link |
at these layers or is something that just permeates
link |
all of these layers and just takes different form?
link |
I believe what humans experience as consciousness
link |
is something that sits on a gradient scale
link |
of a general principle in the universe
link |
that seems to look for order and reach for order
link |
when there's an excess of energy.
link |
You know, it would be odd to say a proton is alive, right?
link |
It'd be odd to say like this particular atom
link |
or molecule of hydrogen gas is alive,
link |
but there's certainly something we can make assemblages
link |
of these things that have autopoetic aspects to them
link |
that will create structures that will, you know,
link |
crystalline solids will form very interesting
link |
and beautiful structures.
link |
This gets kind of into weird mathematical territories.
link |
You start thinking about Penrose and Game of Life stuff
link |
about the generativity of math itself,
link |
like the hyperreal numbers, things like that.
link |
But without going down that rabbit hole,
link |
I would say that there seems to be a tendency
link |
in the world that when there is excess energy,
link |
things will structure and pattern themselves.
link |
And they will then actually furthermore try to create
link |
an environment that furthers their continued stability.
link |
It's the concept of externalized extended phenotype
link |
or niche construction.
link |
So this is ultimately what leads to certain kinds
link |
of amino acids forming certain kinds of structures
link |
and so on and so forth until you get the ladder of life.
link |
So what we experience as consciousness,
link |
no, I don't think cells are conscious at that level,
link |
but is there something beyond mere equilibrium state biology
link |
and chemistry and biochemistry
link |
that drives what makes things work?
link |
So Adrian Bajan has his ConstructoLaw.
link |
There's other things you can look at.
link |
When you look at the life sciences
link |
and you look at any kind of statistical physics
link |
and statistical mechanics,
link |
when you look at things far out of equilibrium,
link |
when you have excess energy, what happens then?
link |
Life doesn't just make a hotter soup.
link |
It starts making structure.
link |
There's something there.
link |
The poetry of reaches for order
link |
when there's an excess of energy.
link |
Because you brought up game of life.
link |
You did it, not me.
link |
I love cellular automata,
link |
so I have to sort of linger on that for a little bit.
link |
So cellular automata, I guess, or game of life
link |
is a very simple example of reaching for order
link |
when there's an excess of energy.
link |
Or reaching for order and somehow creating complexity.
link |
Within this explosion of just turmoil,
link |
somehow trying to construct structures.
link |
And in so doing, create very elaborate
link |
organism looking type things.
link |
What intuition do you draw from this simple mechanism?
link |
Well, I like to turn that around its head.
link |
And look at it as what if every single one of the patterns
link |
created life, or created, not life,
link |
but created interesting patterns?
link |
Because some of them don't.
link |
And sometimes you make cool gliders.
link |
And other times, you start with certain things
link |
and you make gliders and other things
link |
that then construct like AND gates and NOT gates, right?
link |
And you build computers on them.
link |
All of these rules that create these patterns
link |
that we can see, those are just the patterns we can see.
link |
What if our subjectivity is actually limiting
link |
our ability to perceive the order in all of it?
link |
What if some of the things that we think are random
link |
are actually not that random?
link |
We're simply not integrating at a final level
link |
across a broad enough time horizon.
link |
And this is again, I said, we go down the rabbit holes
link |
and the Penrose stuff or like Wolfram's explorations
link |
There is something deep and beautiful
link |
in the mathematics of all this.
link |
That is hopefully one day I'll have enough money
link |
to work and retire and just ponder those questions.
link |
But there's something there.
link |
But you're saying there's a ceiling to,
link |
when you have enough money and you retire and you ponder,
link |
there's a ceiling to how much you can truly ponder
link |
because there's cognitive limitations
link |
in what you're able to perceive as a pattern.
link |
And maybe mathematics extends your perception capabilities,
link |
but it's still finite.
link |
Yeah, the mathematics we use is the mathematics
link |
that can fit in our head.
link |
Did God really create the integers?
link |
Or did God create all of it?
link |
And we just happen at this point in time
link |
to be able to perceive integers.
link |
Well, he just did the positive in it.
link |
She, I just said, did she create all of it?
link |
She just created the natural numbers
link |
and then we screwed it all up with zero and then I guess.
link |
But we did, we created mathematical operations
link |
so that we can have iterated steps
link |
to approach bigger problems, right?
link |
I mean, the entire point of the Arabic Neural System
link |
and it's a rubric for mapping a certain set of operations,
link |
folding them into a simple little expression,
link |
but that's just the operations that we can fit in our heads.
link |
There are many other operations besides, right?
link |
The thing that worries me the most about aliens and humans
link |
is that the aliens are all around us and we're too dumb.
link |
Oh, certainly, yeah.
link |
Or life, let's say just life,
link |
life of all kinds of forms or organisms.
link |
You know what, just even the intelligence of organisms
link |
is imperceptible to us
link |
because we're too dumb and self centered.
link |
Well, we're looking for a particular kind of thing.
link |
When I was at Cornell,
link |
I had a lovely professor of Asian religions,
link |
and she would tell this story about a musical,
link |
a musician, a Western musician who went to Japan
link |
and he taught classical music
link |
and could play all sorts of instruments.
link |
He went to Japan and he would ask people,
link |
he would basically be looking for things in the style of
link |
a Western chromatic scale and these kinds of things.
link |
And then finding none of it,
link |
he would say, well, there's really no music in Japan,
link |
but they're using a different scale.
link |
They're playing different kinds of instruments, right?
link |
The same thing she was using as a sort of a metaphor
link |
for religion as well.
link |
In the West, we center a lot of religion,
link |
certainly the religions of Abraham,
link |
we center them around belief.
link |
And in the East, it's more about practice, right?
link |
Spirituality and practice rather than belief.
link |
So anyway, the point is here to your point,
link |
life, we, I think so many people are so fixated
link |
on certain aspects of self replication
link |
or homeostasis or whatever.
link |
But if we kind of broaden and generalize this thing
link |
of things reaching for order,
link |
under which conditions can they then create an environment
link |
that sustains that order, that allows them,
link |
the invention of death is an interesting thing.
link |
There are some organisms on earth
link |
that are thousands of years old.
link |
And it's not like they're incredibly complex,
link |
they're actually simpler than the cells that comprise us,
link |
but they never die.
link |
So at some point, death was invented,
link |
somewhere along the eukaryotic scale,
link |
I mean, even the protists, right?
link |
And why is that along with the sexual reproduction, right?
link |
There is something about the renewal process,
link |
something about the ability to respond
link |
to a changing environment,
link |
where it just becomes,
link |
just killing off the old generation
link |
and letting new generations try,
link |
seems to be the best way to fit into the niche.
link |
Human historians seems to write about wheels and fires,
link |
the greatest inventions,
link |
but it seems like death and sex are pretty good.
link |
And they're kind of essential inventions
link |
at the very beginning.
link |
At the very beginning, yeah.
link |
Well, we didn't invent them, right?
link |
Well, Broad, you didn't invent them.
link |
you particular Homo sapiens did not invent them,
link |
but we together, it's a team project,
link |
just like you're saying.
link |
I think the greatest Homo sapiens invention
link |
So when you say collaboration,
link |
Peter, where do ideas come from
link |
and how do they take hold in society?
link |
Is that the nature of collaboration?
link |
Is that the basic atom of collaboration is ideas?
link |
It's not not ideas, but it's not only ideas.
link |
There's a book I just started reading
link |
called Death From A Distance.
link |
Have you heard of this?
link |
It's a really fascinating thesis,
link |
which is that humans are the only conspecific,
link |
the only species that can kill other members
link |
of the species from range.
link |
And maybe there's a few exceptions,
link |
but if you look in the animal world,
link |
you see like pronghorns butting heads, right?
link |
You see the alpha lion and the beta lion
link |
and they take each other down.
link |
Humans, we developed the ability
link |
to chuck rocks at each other,
link |
well, at prey, but also at each other.
link |
And that means the beta male can chunk a rock
link |
at the alpha male and take them down.
link |
And he can throw a lot of rocks actually,
link |
miss a bunch of times, but just hit once and be good.
link |
So this ability to actually kill members
link |
of our own species from range
link |
without a threat of harm to ourselves
link |
created essentially mutually assured destruction
link |
where we had to evolve cooperation.
link |
If we didn't, then if we just continue to try to do,
link |
like I'm the biggest monkey in the tribe
link |
and I'm gonna own this tribe and you have to go,
link |
if we do it that way, then those tribes basically failed.
link |
And the tribes that persisted
link |
and that have now given rise to the modern Homo sapiens
link |
are the ones where respecting the fact
link |
that we can kill each other from a range
link |
without harm, like there's an asymmetric ability
link |
to snipe the leader from range.
link |
That meant that we sort of had to learn
link |
how to cooperate with each other, right?
link |
Come back here, don't throw that rock at me.
link |
Let's talk our differences out.
link |
So violence is also part of collaboration.
link |
The threat of violence, let's say.
link |
Well, the recognition, maybe the better way to put it
link |
is the recognition that we have more to gain
link |
by working together than the prisoner's dilemma
link |
of both of us defecting.
link |
So mutually assured destruction in all its forms
link |
is part of this idea of collaboration.
link |
Well, and Eric Weinstein talks about our nuclear peace,
link |
right, I mean, it kind of sucks
link |
with thousands of warheads aimed at each other,
link |
we mean Russia and the US, but it's like,
link |
on the other hand, we only fought proxy wars, right?
link |
We did not have another World War III
link |
of like hundreds of millions of people dying
link |
to like machine gun fire and giant guided missiles.
link |
So the original nuclear weapon is a rock
link |
that we learned how to throw, essentially?
link |
The original, yeah, well, the original scope of the world
link |
for any human being was their little tribe.
link |
I would say it still is for the most part.
link |
Eric Weinstein speaks very highly of you,
link |
which is very surprising to me at first
link |
because I didn't know there's this depth to you
link |
because I knew you as an amazing leader of engineers
link |
and an engineer yourself and so on, so it's fascinating.
link |
Maybe just as a comment, a side tangent that we can take,
link |
what's your nature of your friendship with Eric Weinstein?
link |
How did the two, how did such two interesting paths cross?
link |
Is it your origins in physics?
link |
Is it your interest in philosophy
link |
and the ideas of how the world works?
link |
It's very random, Eric found me.
link |
He actually found Travis and I.
link |
Oliphant, yeah, we were both working
link |
at a company called Enthought back in the mid 2000s
link |
and we were doing a lot of consulting
link |
around scientific Python and we'd made some tools
link |
and Eric was trying to use some of these Python tools
link |
to visualize, he had a fiber bundle approach
link |
to modeling certain aspects of economics.
link |
He was doing this and that's how he kind of got in touch
link |
This was in the early.
link |
This was mid 2000s, oh seven timeframe, oh six, oh seven.
link |
Eric Weinstein trying to use Python.
link |
Right, to visualize fiber bundles.
link |
Using some of the tools that we had built
link |
in the open source.
link |
That's somehow entertaining to me, the thought of that.
link |
It's very funny but then we met with him a couple times,
link |
a really interesting guy and then in the wake
link |
of the oh seven, oh eight kind of financial collapse,
link |
he helped organize with Lee Smolin a symposium
link |
at the Perimeter Institute about okay, well clearly,
link |
big finance can't be trusted, government's in its pockets
link |
with regulatory capture, what the F do we do?
link |
And all sorts of people, Nassim Tlaib was there
link |
and Andy Lowe from MIT was there and Bill Janeway,
link |
I mean just a lot of top billing people were there
link |
and he invited me and Travis and another one
link |
of our coworkers, Robert Kern, who is anyone
link |
in the SciPy, NumPy community knows Robert.
link |
So the three of us also got invited to go to this thing
link |
and that's where I met Brett Weinstein
link |
for the first time as well.
link |
Yeah, I knew him before he got all famous
link |
for unfortunate reasons, I guess.
link |
But anyway, so we met then and kind of had a friendship
link |
throughout since then.
link |
You have a depth of thinking that kind of runs
link |
with Eric in terms of just thinking about the world deeply
link |
and thinking philosophically and then there's Eric's
link |
interest in programming.
link |
I actually have never, you know, he'll bring up programming
link |
to me quite a bit as a metaphor for stuff.
link |
But I never kind of pushed the point of like,
link |
what's the nature of your interest in programming?
link |
I think he saw it probably as a tool.
link |
That you visualize, to explore mathematics
link |
and explore physics and I was wondering like,
link |
what's his depth of interest and also his vision
link |
for what programming would look like in the future.
link |
Have you had interaction with him, like discussion
link |
in the space of Python, programming?
link |
Well, in the sense of sometimes he asks me,
link |
why is this stuff still so hard?
link |
Yeah, you know, everybody's a critic.
link |
But actually, no, Eric.
link |
Programming, you mean, like in general?
link |
Yes, yes, well, not programming in general,
link |
but certain things in the Python ecosystem.
link |
But he actually, I think what I find in listening
link |
to some of his stuff is that he does use
link |
programming metaphors a lot, right?
link |
He'll talk about APIs or object oriented
link |
and things like that.
link |
So I think that's a useful set of frames
link |
for him to draw upon for discourse.
link |
I haven't pair programmed with him in a very long time.
link |
You've previously pair coded with Eric.
link |
Well, I mean, I look at his code trying to help
link |
like put together some of the visualizations
link |
around these things.
link |
But it's been a very, not really pair programmed,
link |
but like even looked at his code, right?
link |
How legendary would be is that like Git repo
link |
with Peter Wang and Eric Weinstein?
link |
Well, honestly, Robert Kern did all the heavy lifting.
link |
So I have to give credit where credit is due.
link |
Robert is the silent but incredibly deep, quiet,
link |
not silent, but quiet, but incredibly deep individual
link |
at the heart of a lot of those things
link |
that Eric was trying to do.
link |
But we did have, you know, as Travis and I
link |
were starting our company in 2012 timeframe,
link |
we went to New York.
link |
Eric was still in New York at the time.
link |
He hadn't moved to, this is before he joined Teal Capital.
link |
We just had like a steak dinner somewhere.
link |
Maybe it was Keynes, I don't know, somewhere in New York.
link |
So it was me, Travis, Eric, and then Wes McKinney,
link |
the creative pandas, and then Wes's then business partner,
link |
Adam, the five of us sat around having this,
link |
just a hilarious time, amazing dinner.
link |
I forget what all we talked about,
link |
but it was one of those conversations,
link |
which I wish as soon as COVID is over,
link |
maybe Eric and I can sit down.
link |
Recreate it somewhere in LA, or maybe he comes here,
link |
because a lot of cool people are here in Austin, right?
link |
Yeah, we're all here.
link |
He should come here.
link |
So he uses the metaphor source code sometimes
link |
to talk about physics.
link |
We figure out our own source code.
link |
So you with a physics background
link |
and somebody who's quite a bit of an expert in source code,
link |
do you think we'll ever figure out our own source code
link |
in the way that Eric means?
link |
Do you think we'll figure out the nature of reality?
link |
Well, I think we're constantly working on that problem.
link |
I mean, I think we'll make more and more progress.
link |
For me, there's some things I don't really doubt too much.
link |
Like, I don't really doubt that one day
link |
we will create a synthetic, maybe not fully in silicon,
link |
but a synthetic approach to
link |
cognition that rivals the biological
link |
20 watt computers in our heads.
link |
What's cognition here?
link |
Perception, attention, memory, recall,
link |
asking better questions.
link |
That for me is a measure of intelligence.
link |
Doesn't Roomba vacuum cleaner already do that?
link |
Or do you mean, oh, it doesn't ask questions.
link |
I mean, no, it's, I mean, I have a Roomba,
link |
but it's not even as smart as my cat, right?
link |
Yeah, but it asks questions about what is this wall?
link |
It now, new feature asks, is this poop or not, apparently.
link |
Yes, a lot of our current cybernetic system,
link |
it's a cybernetic system.
link |
It will go and it will happily vacuum up some poop, right?
link |
The older generations would.
link |
The new one, just released, does not vacuum up the poop.
link |
This is a commercial for.
link |
I wonder if it still gets stuck
link |
under my first rung of my stair.
link |
In any case, these cybernetic systems we have,
link |
they are mold, they're designed to be sent off
link |
into a relatively static environment.
link |
And whatever dynamic things happen in the environment,
link |
they have a very limited capacity to respond to.
link |
A human baby, a human toddler of 18 months of age
link |
has more capacity to manage its own attention
link |
and its own capacity to make better sense of the world
link |
than the most advanced robots today.
link |
So again, my cat, I think can do a better job of my two
link |
and they're both pretty clever.
link |
So I do think though, back to my kind of original point,
link |
I think that it's not, for me, it's not question at all
link |
that we will be able to create synthetic systems
link |
that are able to do this better than the human,
link |
at an equal level or better than the human mind.
link |
It's also for me, not a question that we will be able
link |
to put them alongside humans
link |
so that they capture the full broad spectrum
link |
of what we are seeing as well.
link |
And also looking at our responses,
link |
listening to our responses,
link |
even maybe measuring certain vital signs about us.
link |
So in this kind of sidecar mode,
link |
a greater intelligence could use us
link |
and our whatever 80 years of life to train itself up
link |
and then be a very good simulacrum of us moving forward.
link |
So who is in the sidecar
link |
in that picture of the future exactly?
link |
The baby version of our immortal selves.
link |
Okay, so once the baby grows up,
link |
is there any use for humans?
link |
I think that out of epistemic humility,
link |
we need to keep humans around for a long time.
link |
And I would hope that anyone making those systems
link |
would believe that to be true.
link |
Out of epistemic humility,
link |
what's the nature of the humility that?
link |
That we don't know what we don't know.
link |
First we have to build systems
link |
that help us do the things that we do know about
link |
that can then probe the unknowns that we know about.
link |
But the unknown unknowns, we don't know.
link |
We could always know.
link |
Nature is the one thing
link |
that is infinitely able to surprise us.
link |
So we should keep biological humans around
link |
for a very, very, very long time.
link |
Even after our immortal selves have transcended
link |
and have gone off to explore other worlds,
link |
gone to go communicate with the lifeforms living in the sun
link |
So I think that's,
link |
for me, these seem like things that are going to happen.
link |
Like I don't really question that,
link |
that they're gonna happen.
link |
Assuming we don't completely destroy ourselves.
link |
Is it possible to create an AI system
link |
that you fall in love with and it falls in love with you
link |
and you have a romantic relationship with it?
link |
Or a deep friendship, let's say.
link |
I would hope that that is the design criteria
link |
for any of these systems.
link |
If we cannot have a meaningful relationship with it,
link |
then it's still just a chunk of silicon.
link |
So then what is meaningful?
link |
Because back to sugar.
link |
Well, sugar doesn't love you back, right?
link |
So the computer has to love you back.
link |
And what does love mean?
link |
Well, in this context, for me, love,
link |
I'm gonna take a page from Alain de Botton.
link |
Love means that it wants to help us
link |
become the best version of ourselves.
link |
Yes, that's beautiful.
link |
That's a beautiful definition of love.
link |
So what role does love play in the human condition
link |
at the individual level and at the group level?
link |
Because you were kind of saying that humans,
link |
we should really consider humans
link |
both at the individual and the group and the societal level.
link |
What's the role of love in this whole thing?
link |
We talked about sex, we talked about death,
link |
thanks to the bacteria that invented it.
link |
At which point did we invent love, by the way?
link |
I mean, is that also?
link |
No, I think love is the start of it all.
link |
And the feelings of, and this gets sort of beyond
link |
just romantic, sensual, whatever kind of things,
link |
but actually genuine love as we have for another person.
link |
Love as it would be used in a religious text, right?
link |
I think that capacity to feel love
link |
more than consciousness, that is the universal thing.
link |
Our feeling of love is actually a sense
link |
of that generativity.
link |
When we can look at another person
link |
and see that they can be something more than they are,
link |
and more than just a pigeonhole we might stick them in.
link |
I mean, I think there's, in any religious text,
link |
you'll find voiced some concept of this,
link |
that you should see the grace of God in the other person.
link |
They're made in the spirit of the love
link |
that God feels for his creation or her creation.
link |
And so I think this thing is actually the root of it.
link |
So I would say, I don't think molecules of water
link |
feel consciousness, have consciousness,
link |
but there is some proto micro quantum thing of love.
link |
That's the generativity when there's more energy
link |
than what they need to maintain equilibrium.
link |
And that when you sum it all up is something that leads to,
link |
I mean, I had my mind blown one day as an undergrad
link |
at the physics computer lab.
link |
I logged in and when you log into bash for a long time,
link |
there was a little fortune that would come out.
link |
And it said, man was created by water
link |
to carry itself uphill.
link |
And I was logging into work on some problem set
link |
and I logged in and I saw that and I just said,
link |
son of a bitch, I just, I logged out
link |
and I went to the coffee shop and I got a coffee
link |
and I sat there on the quad and I'm like,
link |
you know, it's not wrong and yet WTF, right?
link |
So when you look at it that way,
link |
it's like, yeah, okay, non equilibrium physics is a thing.
link |
And so when we think about love,
link |
when we think about these kinds of things, I would say
link |
that in the modern day human condition,
link |
there's a lot of talk about freedom and individual liberty
link |
and rights and all these things,
link |
but that's very Hegelian, it's very kind of following
link |
from the Western philosophy of the individual as sacrosanct,
link |
but it's not really couched I think the right way
link |
because it should be how do we maximize people's ability
link |
to love each other, to love themselves first,
link |
to love each other, their responsibilities
link |
to the previous generation, to the future generations.
link |
Those are the kinds of things
link |
that should be our design criteria, right?
link |
Those should be what we start with to then come up
link |
with the philosophies of self and of rights
link |
and responsibilities, but that love being at the center
link |
of it, I think when we design systems for cognition,
link |
it should absolutely be built that way.
link |
I think if we simply focus on efficiency and productivity,
link |
these kind of very industrial era,
link |
all the things that Marx had issues with, right?
link |
Those, that's a way to go and really I think go off
link |
the deep end in the wrong way.
link |
So one of the interesting consequences of thinking of life
link |
in this hierarchical way of an individual human
link |
and then there's groups and there's societies
link |
is I believe that you believe that corporations are people.
link |
So this is a kind of a politically dense idea,
link |
all those kinds of things.
link |
If we just throw politics aside,
link |
if we throw all of that aside,
link |
in which sense do you believe that corporations are people?
link |
And how does love connect to that?
link |
Right, so the belief is that groups of people
link |
have some kind of higher level, I would say mesoscopic
link |
So where do I, let's start with this.
link |
Most people would say, okay, individuals have claims
link |
to agency and sovereignty.
link |
Nations, we certainly act as if nations,
link |
so at a very large, large scale,
link |
nations have rights to sovereignty and agency.
link |
Like everyone plays the game of modernity
link |
as if that's true, right?
link |
We believe France is a thing,
link |
we believe the United States is a thing.
link |
But to say that groups of people at a smaller level
link |
than that, like a family unit is the thing.
link |
Well, in our laws, we actually do encode this concept.
link |
I believe that in a relationship and a marriage, right,
link |
one partner can sue for loss of consortium, right?
link |
If someone breaks up the marriage or whatever.
link |
So these are concepts that even in law,
link |
we do respect that there is something about the union
link |
and about the family.
link |
So for me, I don't think it's so weird to think
link |
that groups of people have a right to,
link |
a claim to rights and sovereignty of some degree.
link |
I mean, we look at our clubs, we look at churches.
link |
These are, we talk about these collectives of people
link |
as if they have a real agency to them, and they do.
link |
But I think if we take that one step further and say,
link |
okay, they can accrue resources.
link |
Well, yes, check, you know, and by law they can.
link |
They can own land, they can engage in contracts,
link |
they can do all these different kinds of things.
link |
So we in legal terms support this idea
link |
that groups of people have rights.
link |
Where we go wrong on this stuff
link |
is that the most popular version of this
link |
is the for profit absentee owner corporation
link |
that then is able to amass larger resources
link |
than anyone else in the landscape, anything else,
link |
any other entity of equivalent size.
link |
And they're able to essentially bully around individuals,
link |
whether it's laborers, whether it's people
link |
whose resources they want to capture.
link |
They're also able to bully around
link |
our system of representation,
link |
which is still tied to individuals, right?
link |
So I don't believe that's correct.
link |
I don't think it's good that they, you know,
link |
they're people, but they're assholes.
link |
I don't think that corporations as people
link |
acting like assholes is a good thing.
link |
But the idea that collectives and collections of people
link |
that we should treat them philosophically
link |
as having some agency and some mass,
link |
at a mesoscopic level, I think that's an important thing
link |
because one thing I do think we underappreciate sometimes
link |
is the fact that relationships have relationships.
link |
So it's not just individuals
link |
having relationships with each other.
link |
But if you have eight people seated around a table, right?
link |
Each person has a relationship with each of the others
link |
and that's obvious.
link |
But then if it's four couples,
link |
each couple also has a relationship
link |
with each of the other couples, right?
link |
And if it's couples, but one is the, you know,
link |
father and mother older, and then, you know,
link |
one of their children and their spouse,
link |
that family unit of four has a relationship
link |
with the other family unit of four.
link |
So the idea that relationships have relationships
link |
is something that we intuitively know
link |
in navigating the social landscape,
link |
but it's not something I hear expressed like that.
link |
It's certainly not something that is,
link |
I think, taken into account very well
link |
when we design these kinds of things.
link |
So I think the reason why I care a lot about this
link |
is because I think the future of humanity
link |
requires us to form better sense make,
link |
collective sense making units at something, you know,
link |
around Dunbar number, you know, half to five X Dunbar.
link |
And that's very different than right now
link |
where we defer sense making
link |
to massive aging zombie institutions.
link |
Or we just do it ourselves.
link |
Go to the dark force of the internet by ourselves.
link |
So that's really interesting.
link |
So you've talked about agency,
link |
I think maybe calling it a convenient fiction
link |
at all these different levels.
link |
So even at the human individual level,
link |
it's kind of a fiction.
link |
We all believe, because we are, like you said,
link |
made of cells and cells are made of atoms.
link |
So that's a useful fiction.
link |
And then there's nations that seems to be a useful fiction,
link |
but it seems like some fictions are better than others.
link |
You know, there's a lot of people that argue
link |
the fiction of nation is a bad idea.
link |
One of them lives two doors down from me,
link |
Michael Malice, he's an anarchist.
link |
You know, I'm sure there's a lot of people
link |
who are into meditation that believe the idea,
link |
this useful fiction of agency of an individual
link |
is a troublesome as well.
link |
We need to let go of that in order to truly,
link |
like to transcend, I don't know.
link |
I don't know what words you want to use,
link |
but suffering or to elevate the experience of life.
link |
So you're kind of arguing that,
link |
okay, so we have some of these useful fictions of agency.
link |
We should add a stronger fiction that we tell ourselves
link |
about the agency of groups in the hundreds
link |
of the half a Dunbar's number, 5X Dunbar's number.
link |
Yeah, something on that order.
link |
And we call them fictions,
link |
but really they're rules of the game, right?
link |
Rules that we feel are fair or rules that we consent to.
link |
Yeah, I always question the rules
link |
when I lose like a monopoly.
link |
That's when I usually question the rules.
link |
When I'm winning, I don't question the rules.
link |
We should play a game Monopoly someday.
link |
There's a trippy version of it that we could do.
link |
Contract Monopoly is introduced by a friend of mine to me
link |
where you can write contracts on future earnings
link |
or landing on various things.
link |
And you can hand out like, you know,
link |
you can land the first three times you land
link |
in a park place, it's free or whatever.
link |
And then you can start trading those contracts for money.
link |
And then you create a human civilization
link |
and somehow Bitcoin comes into it.
link |
Okay, but some of these.
link |
Actually, I bet if me and you and Eric sat down
link |
to play a game Monopoly and we were to make NFTs
link |
out of the contracts we wrote, we could make a lot of money.
link |
Now it's a terrible idea.
link |
I would never do it,
link |
but I bet we could actually sell the NFTs around.
link |
I have other ideas to make money that I could tell you
link |
and they're all terrible ideas.
link |
Yeah, including cat videos on the internet.
link |
Okay, but some of these rules of the game,
link |
some of these fictions are,
link |
it seems like they're better than others.
link |
They have worked this far to cohere human,
link |
to organize human collective action.
link |
But you're saying something about,
link |
especially this technological age
link |
requires modified fictions, stories of agency.
link |
Why the Dunbar number?
link |
And also, you know, how do you select the group of people?
link |
You know, Dunbar numbers, I think I have the sense
link |
that it's overused as a kind of law
link |
that somehow we can have deep human connection at this scale.
link |
Like some of it feels like an interface problem too.
link |
It feels like if I have the right tools,
link |
I can deeply connect with a larger number of people.
link |
It just feels like there's a huge value
link |
to interacting just in person, getting to share
link |
traumatic experiences together,
link |
beautiful experiences together.
link |
There's other experiences like that in the digital space
link |
that you can share.
link |
It just feels like Dunbar's number
link |
could be expanded significantly,
link |
perhaps not to the level of millions and billions,
link |
but it feels like it could be expanded.
link |
So how do we find the right interface, you think,
link |
for having a little bit of a collective here
link |
You're right that there's many different ways
link |
that we can build trust with each other.
link |
My friend Joe Edelman talks about a few different ways
link |
that, you know, mutual appreciation, trustful conflict,
link |
just experiencing something like, you know,
link |
there's a variety of different things that we can do,
link |
but all those things take time and you have to be present.
link |
The less present you are, I mean, there's just, again,
link |
a no free lunch principle here.
link |
The less present you are, the more of them you can do,
link |
but then the less connection you build.
link |
So I think there is sort of a human capacity issue
link |
around some of these things.
link |
Now, that being said, if we can use certain technologies,
link |
so for instance, if I write a little monograph
link |
on my view of the world,
link |
you read it asynchronously at some point,
link |
and you're like, wow, Peter, this is great.
link |
I'm like, wow, Lex, this is awesome.
link |
We can be friends without having to spend 10 years,
link |
you know, figuring all this stuff out together.
link |
We just read each other's thing and be like,
link |
oh yeah, this guy's exactly in my wheelhouse
link |
And we can then, you know, connect just a few times a year
link |
and maintain a high trust relationship.
link |
It can be expanded a little bit,
link |
but it also requires,
link |
these things are not all technological in nature.
link |
It requires the individual themselves
link |
to have a certain level of capacity,
link |
to have a certain lack of neuroticism, right?
link |
If you want to use like the ocean big five sort of model,
link |
people have to be pretty centered.
link |
The less centered you are,
link |
the fewer authentic connections you can really build
link |
for a particular unit of time.
link |
It just takes more time.
link |
Other people have to put up with your crap.
link |
Like there's just a lot of the stuff
link |
that you have to deal with
link |
if you are not so well balanced, right?
link |
So yes, we can help people get better
link |
to where they can develop more relationships faster,
link |
and then you can maybe expand Dunbar number by quite a bit,
link |
but you're not going to do it.
link |
I think it's going to be hard to get it beyond 10X,
link |
kind of the rough swag of what it is, you know?
link |
Well, don't you think that AI systems could be an addition
link |
to the Dunbar's number?
link |
Do you count as one system or multiple AI systems?
link |
Multiple AI systems.
link |
So I do believe that AI systems,
link |
for them to integrate into human society as it is now,
link |
have to have a sense of agency.
link |
So there has to be a individual
link |
because otherwise we wouldn't relate to them.
link |
We could engage certain kinds of individuals
link |
to make sense of them for us and be almost like,
link |
did you ever watch Star Trek?
link |
Like Voyager, like there's the Volta,
link |
who are like the interfaces,
link |
the ambassadors for the Dominion.
link |
We may have ambassadors that speak
link |
on behalf of these systems.
link |
They're like the Mentats of Dune, maybe,
link |
or something like this.
link |
I mean, we already have this to some extent.
link |
If you look at the biggest sort of,
link |
I wouldn't say AI system,
link |
but the biggest cybernetic system in the world
link |
is the financial markets.
link |
It runs outside of any individual's control,
link |
and you have an entire stack of people on Wall Street,
link |
Wall Street analysts to CNBC reporters, whatever.
link |
They're all helping to communicate what does this mean?
link |
You know, like Jim Cramer,
link |
like coming around and yelling and stuff.
link |
Like all of these people are part of that lowering
link |
of the complexity there to meet sense,
link |
you know, to help do sense making for people
link |
at whatever capacity they're at.
link |
And I don't see this changing with AI systems.
link |
I think you would have ringside commentators
link |
talking about all this stuff
link |
that this AI system is trying to do over here, over here,
link |
because it's actually a super intelligence.
link |
So if you want to talk about humans interfacing,
link |
making first contact with the super intelligence,
link |
we're already there.
link |
We do it pretty poorly.
link |
And if you look at the gradient of power and money,
link |
what happens is the people closest to it
link |
will absolutely exploit their distance
link |
for personal financial gain.
link |
So we should look at that and be like,
link |
oh, well, that's probably what the future
link |
will look like as well.
link |
But nonetheless, I mean,
link |
we're already doing this kind of thing.
link |
So in the future, we can have AI systems,
link |
but you're still gonna have to trust people
link |
to bridge the sense making gap to them.
link |
See, I just feel like there could be
link |
like millions of AI systems that have,
link |
have agencies, you have,
link |
when you say one super intelligence,
link |
super intelligence in that context means
link |
it's able to solve particular problems extremely well.
link |
But there's some aspect of human like intelligence
link |
that's necessary to be integrated into human society.
link |
So not financial markets,
link |
not sort of weather prediction systems,
link |
or I don't know, logistics optimization.
link |
I'm more referring to things that you interact with
link |
on the intellectual level.
link |
And that I think requires,
link |
there has to be a backstory.
link |
There has to be a personality.
link |
I believe it has to fear its own mortality in a genuine way.
link |
Like there has to be all,
link |
many of the elements that we humans experience
link |
that are fundamental to the human condition,
link |
because otherwise we would not have
link |
a deep connection with it.
link |
But I don't think having a deep connection with it
link |
is necessarily going to stop us from building a thing
link |
that has quite an alien intelligence aspect here.
link |
So the other kind of alien intelligence on this planet
link |
is the octopuses or octopodes
link |
or whatever you wanna call them.
link |
Octopi. Octopi, yeah.
link |
There's a little controversy
link |
as to what the plural is, I guess.
link |
But an octopus. I look forward to your letters.
link |
it really acts as a collective intelligence
link |
of eight intelligent arms, right?
link |
Its arms have a tremendous amount of neural density to them.
link |
And I see if we can build,
link |
I mean, just let's go with what you're saying.
link |
If we build a singular intelligence
link |
that interfaces with humans that has a sense of agency
link |
so it can run the cybernetic loop
link |
and develop its own theory of mind
link |
as well as its theory of action,
link |
all these things, I agree with you
link |
that that's the necessary components
link |
to build a real intelligence, right?
link |
There's gotta be something at stake.
link |
It's gotta make a decision.
link |
It's gotta then run the OODA loop.
link |
Okay, so we build one of those.
link |
Well, if we can build one of those,
link |
we can probably build 5 million of them.
link |
So we build 5 million of them.
link |
And if their cognitive systems are already digitized
link |
and already kind of there,
link |
we stick an antenna on each of them,
link |
bring it all back to a hive mind
link |
that maybe doesn't make all the individual decisions
link |
for them, but treats each one
link |
as almost like a neuronal input
link |
of a much higher bandwidth and fidelity,
link |
going back to a central system
link |
that is then able to perceive much broader dynamics
link |
that we can't see.
link |
In the same way that a phased array radar, right?
link |
You think about how phased array radar works.
link |
It's just sensitivity.
link |
It's just radars, and then it's hypersensitivity
link |
and really great timing between all of them.
link |
And with a flat array,
link |
it's as good as a curved radar dish, right?
link |
So with these things,
link |
it's a phased array of cybernetic systems
link |
that'll give the centralized intelligence
link |
much, much better, a much higher fidelity understanding
link |
of what's actually happening in the environment.
link |
But the more power,
link |
the more understanding the central super intelligence has,
link |
the dumber the individual like fingers
link |
of this intelligence are, I think.
link |
I don't see what has to be.
link |
This argument, there has to be,
link |
the experience of the individual agent
link |
has to have the full richness of the human like experience.
link |
You have to be able to be driving the car in the rain,
link |
listening to Bruce Springsteen,
link |
and all of a sudden break out in tears
link |
because remembering something that happened to you
link |
We can implant those memories
link |
if that's really needed.
link |
No, but the central agency,
link |
like I guess I'm saying for, in my view,
link |
for intelligence to be born,
link |
you have to have a decentralization.
link |
Like each one has to struggle and reach.
link |
So each one in excess of energy has to reach for order
link |
as opposed to a central place doing so.
link |
Have you ever read like some sci fi
link |
where there's like hive minds?
link |
Like the Wernher Vinge, I think has one of these.
link |
And then some of the stuff from the Commonwealth Saga,
link |
the idea that you're an individual,
link |
but you're connected with like a few other individuals
link |
telepathically as well.
link |
And together you form a swarm.
link |
So if you are, I ask you,
link |
what do you think is the experience of if you are like,
link |
well, a Borg, right?
link |
If you are one, if you're part of this hive mind,
link |
outside of all the aesthetics, forget the aesthetics,
link |
internally, what is your experience like?
link |
Because I have a theory as to what that looks like.
link |
The one question I have for you about that experience is
link |
how much is there a feeling of freedom, of free will?
link |
Because I obviously as a human, very unbiased,
link |
but also somebody who values freedom and biased,
link |
it feels like the experience of freedom is essential for
link |
trying stuff out, to being creative
link |
and doing something truly novel, which is at the core of.
link |
Yeah, well, I don't think you have to lose any freedom
link |
when you're in that mode.
link |
Because I think what happens is we think,
link |
we still think, I mean, you're still thinking about this
link |
in a sense of a top down command and control hierarchy,
link |
which is not what it has to be at all.
link |
I think the experience, so I'll just show by cards here.
link |
I think the experience of being a robot in that robot swarm,
link |
a robot who has agency over their own local environment
link |
that's doing sense making
link |
and reporting it back to the hive mind,
link |
I think that robot's experience would be one,
link |
when the hive mind is working well,
link |
it would be an experience of like talking to God, right?
link |
That you essentially are reporting to,
link |
you're sort of saying, here's what I see.
link |
I think this is what's gonna happen over here.
link |
I'm gonna go do this thing.
link |
Because I think if I'm gonna do this,
link |
this will make this change happen in the environment.
link |
And then God, she may tell you, that's great.
link |
And in fact, your brothers and sisters will join you
link |
to help make this go better, right?
link |
And then she can let your brothers and sisters know,
link |
hey, Peter's gonna go do this thing.
link |
Would you like to help him?
link |
Because we think that this will make this thing go better.
link |
And they'll say, yes, we'll help him.
link |
So the whole thing could be actually very emergent.
link |
The sense of, what does it feel like to be a cell
link |
and a network that is alive, that is generative.
link |
And I think actually the feeling is serendipity.
link |
That there's random order, not random disorder or chaos,
link |
but random order, just when you need it to hear Bruce Springsteen,
link |
you turn on the radio and bam, it's Bruce Springsteen, right?
link |
That feeling of serendipity, I feel like,
link |
this is a bit of a flight of fancy,
link |
but every cell in your body must have,
link |
what does it feel like to be a cell in your body?
link |
When it needs sugar, there's sugar.
link |
When it needs oxygen, there's just oxygen.
link |
Now, when it needs to go and do its work
link |
and pull like as one of your muscle fibers, right?
link |
It does its work and it's great.
link |
It contributes to the cause, right?
link |
So this is all, again, a flight of fancy,
link |
but I think as we extrapolate up,
link |
what does it feel like to be an independent individual
link |
with some bounded sense of freedom?
link |
All sense of freedom is actually bounded,
link |
but it was a bounded sense of freedom
link |
that still lives within a network that has order to it.
link |
And I feel like it has to be a feeling of serendipity.
link |
So the cell, there's a feeling of serendipity, even though.
link |
It has no way of explaining why it's getting oxygen
link |
and sugar when it gets it.
link |
So you have to, each individual component has to be too dumb
link |
to understand the big picture.
link |
No, the big picture is bigger than what it can understand.
link |
But isn't that an essential characteristic
link |
of the individual is to be too dumb
link |
to understand the bigger picture.
link |
Like not dumb necessarily,
link |
but limited in its capacity to understand.
link |
Because the moment you understand,
link |
I feel like that leads to, if you tell me now
link |
that there are some bigger intelligence
link |
controlling everything I do,
link |
intelligence broadly defined, meaning like,
link |
you know, even the Sam Harris thing, there's no free will.
link |
If I'm smart enough to truly understand that that's the case,
link |
that's gonna, I don't know if I.
link |
We have philosophical breakdown, right?
link |
Because we're in the West and we're pumped full of this stuff
link |
of like, you are a golden, fully free individual
link |
with all your freedoms and all your liberties
link |
and go grab a gun and shoot whatever you want to.
link |
No, it's actually, you don't actually have a lot of these,
link |
you're not unconstrained,
link |
but the areas where you can manifest agency,
link |
you're free to do those things.
link |
You can say whatever you want on this podcast.
link |
You can create a podcast, right?
link |
You're not, I mean, you have a lot of this kind of freedom,
link |
but even as you're doing this, you are actually,
link |
I guess where the denouement of this is that
link |
we are already intelligent agents in such a system, right?
link |
In that one of these like robots
link |
of one of 5 million little swarm robots
link |
or one of the Borg,
link |
they're just posting on internal bulletin board.
link |
I mean, maybe the Borg cube
link |
is just a giant Facebook machine floating in space
link |
and everyone's just posting on there.
link |
They're just posting really fast and like, oh yeah.
link |
It's called the metaverse now.
link |
That's called the metaverse, that's right.
link |
Here's the enterprise.
link |
Maybe we should all go shoot it.
link |
Yeah, everyone upvotes and they're gonna go shoot it, right?
link |
But we already are part of a human online
link |
collaborative environment
link |
and collaborative sensemaking system.
link |
It's not very good yet.
link |
It's got the overhangs of zombie sensemaking institutions
link |
all over it, but as that washes away
link |
and as we get better at this,
link |
we are going to see humanity improving
link |
at speeds that are unthinkable in the past.
link |
And it's not because anyone's freedoms were limited.
link |
In fact, the open source,
link |
and we started this with open source software, right?
link |
The collaboration, what the internet surfaced
link |
was the ability for people all over the world
link |
to collaborate and produce some of the most
link |
foundational software that's in use today, right?
link |
That entire ecosystem was created
link |
by collaborators all over the place.
link |
So these online kind of swarm kind of things
link |
It's just, I'm just suggesting that future AI systems,
link |
if you can build one smart system,
link |
you have no reason not to build multiple.
link |
If you build multiple,
link |
there's no reason not to integrate them all
link |
into a collective sensemaking substrate.
link |
And that thing will certainly have immersion intelligence
link |
that none of the individuals
link |
and probably not any of the human designers
link |
will be able to really put a bow around and explain.
link |
But in some sense, would that AI system
link |
still be able to go like rural Texas,
link |
buy a ranch, go off the grid, go full survivalist?
link |
Like, can you disconnect from the hive mind?
link |
You may not want to.
link |
So to be ineffective, to be intelligent.
link |
You have access to way more intelligence capability
link |
if you're plugged into five million other
link |
really, really smart cyborgs.
link |
Why would you leave?
link |
So like there's a word control that comes to mind.
link |
So it doesn't feel like control,
link |
like overbearing control.
link |
It's just knowledge.
link |
I think systems, well, this is to your point.
link |
I mean, look at how much,
link |
how uncomfortable you are with this concept, right?
link |
I think systems that feel like overbearing control
link |
will not evolutionarily win out.
link |
I think systems that give their individual elements
link |
the feeling of serendipity and the feeling of agency
link |
that that will, those systems will win.
link |
But that's not to say that there will not be
link |
emergent higher level order on top of it.
link |
And that's the thing, that's the philosophical breakdown
link |
that we're staring right at,
link |
which is in the Western mind,
link |
I think there's a very sharp delineation
link |
between explicit control,
link |
Cartesian, like what is the vector?
link |
Where is the position?
link |
Where is it going?
link |
It's completely deterministic.
link |
And kind of this idea that things emerge.
link |
Everything we see is the emergent patterns
link |
And there is agency when there's extra energy.
link |
So you have spoken about a kind of meaning crisis
link |
that we're going through.
link |
But it feels like since we invented sex and death,
link |
we broadly speaking,
link |
we've been searching for a kind of meaning.
link |
So it feels like a human civilization
link |
has been going through a meaning crisis
link |
of different flavors throughout its history.
link |
Why is, how is this particular meaning crisis different?
link |
Or is it really a crisis and it wasn't previously?
link |
What's your sense?
link |
A lot of human history,
link |
there wasn't so much a meaning crisis.
link |
There was just a like food
link |
and not getting eaten by bears crisis, right?
link |
Once you get to a point where you can make food,
link |
there was the like not getting killed
link |
by other humans crisis.
link |
So sitting around wondering what is it all about,
link |
it's actually a relatively recent luxury.
link |
And to some extent, the meaning crisis coming out of that
link |
is precisely because, well, it's not precisely because,
link |
I believe that meaning is the consequence of
link |
when we make consequential decisions,
link |
it's tied to agency, right?
link |
When we make consequential decisions,
link |
that generates meaning.
link |
So if we make a lot of decisions,
link |
but we don't see the consequences of them,
link |
then it feels like what was the point, right?
link |
But if there's all these big things
link |
that we don't see the consequences of,
link |
right, but if there's all these big things happening,
link |
but we're just along for the ride,
link |
then it also does not feel very meaningful.
link |
Meaning, as far as I can tell,
link |
this is my working definition of CERCA 2021,
link |
is generally the result of a person
link |
making a consequential decision,
link |
acting on it and then seeing the consequences of it.
link |
So historically, just when humans are in survival mode,
link |
you're making consequential decisions all the time.
link |
So there's not a lack of meaning
link |
because like you either got eaten or you didn't, right?
link |
You got some food and that's great, you feel good.
link |
Like these are all consequential decisions.
link |
Only in the post fossil fuel and industrial revolution
link |
could we create a massive leisure class.
link |
I could sit around not being threatened by bears,
link |
not starving to death,
link |
making decisions somewhat,
link |
but a lot of times not seeing the consequences
link |
of any decisions they make.
link |
The general sort of sense of anomie,
link |
I think that is the French term for it,
link |
in the wake of the consumer society,
link |
in the wake of mass media telling everyone,
link |
hey, choosing between Hermes and Chanel
link |
is a meaningful decision.
link |
I don't know what either of those mean.
link |
Oh, they're high end luxury purses and crap like that.
link |
But the point is that we give people the idea
link |
that consumption is meaning,
link |
that making a choice of this team versus that team,
link |
spectating has meaning.
link |
So we produce all of these different things
link |
that are as if meaning, right?
link |
But really making a decision that has no consequences for us.
link |
And so that creates the meaning crisis.
link |
Well, you're saying choosing between Chanel
link |
and the other one has no consequence.
link |
I mean, why is one more meaningful than the other?
link |
It's not that it's more meaningful than the other.
link |
It's that you make a decision between these two brands
link |
and you're told this brand will make me look better
link |
in front of other people.
link |
If I buy this brand of car,
link |
if I wear that brand of apparel, right?
link |
Like a lot of decisions we make are around consumption,
link |
but consumption by itself doesn't actually yield meaning.
link |
Gaining social status does provide meaning.
link |
So that's why in this era of abundant production,
link |
so many things turn into status games.
link |
The NFT kind of explosion is a similar kind of thing.
link |
Everywhere there are status games
link |
because we just have so much excess production.
link |
But aren't those status games a source of meaning?
link |
Like why do the games we play have to be grounded
link |
in physical reality like they are
link |
when you're trying to run away from lions?
link |
Why can't we, in this virtuality world, on social media,
link |
why can't we play the games on social media,
link |
even the dark ones?
link |
Right, we can, we can.
link |
But you're saying that's creating a meaning crisis.
link |
Well, there's a meaning crisis
link |
in that there's two aspects of it.
link |
Number one, playing those kinds of status games
link |
oftentimes requires destroying the planet
link |
because it ties to consumption,
link |
consuming the latest and greatest version of a thing,
link |
buying the latest limited edition sneaker
link |
and throwing out all the old ones.
link |
Maybe it keeps in the old ones,
link |
but the amount of sneakers we have to cut up
link |
and destroy every year
link |
to create artificial scarcity for the next generation, right?
link |
This is kind of stuff that's not great.
link |
It's not great at all.
link |
So conspicuous consumption fueling status games
link |
is really bad for the planet, not sustainable.
link |
The second thing is you can play these kinds of status games,
link |
but then what it does is it renders you captured
link |
to the virtual environment.
link |
The status games that really wealthy people are playing
link |
are all around the hard resources
link |
where they're gonna build the factories,
link |
they're gonna have the fuel in the rare earths
link |
to make the next generation of robots.
link |
They're then going to run game,
link |
run circles around you and your children.
link |
So that's another reason not to play
link |
those virtual status games.
link |
So you're saying ultimately the big picture game is won
link |
by people who have access or control
link |
over actual hard resources.
link |
So you can't, you don't see a society
link |
where most of the games are played in the virtual space.
link |
They'll be captured in the physical space.
link |
It's just like the stack of human being, right?
link |
If you only play the game at the cultural
link |
and then intellectual level,
link |
then the people with the hard resources
link |
and access to layer zero physical are going to own you.
link |
But isn't money not connected to,
link |
or less and less connected to hard resources
link |
and money still seems to work?
link |
It's a virtual technology.
link |
There's different kinds of money.
link |
Part of the reason that some of the stuff is able
link |
to go a little unhinged is because the big sovereignties
link |
where one spends money and uses money
link |
and plays money games and inflates money,
link |
their ability to adjudicate the physical resources
link |
and hard resources and the resources
link |
and hard resources on land and things like that,
link |
those have not been challenged in a very long time.
link |
So, you know, we went off the gold standard.
link |
Most money is not connected to physical resources.
link |
And that idea is very closely connected to status.
link |
But it's also tied to like, it's actually tied to law.
link |
It is tied to some physical hard things
link |
so you have to pay your taxes.
link |
Yes, so it's always at the end going to be connected
link |
to the blockchain of physical reality.
link |
So in the case of law and taxes, it's connected to government
link |
and government is what violence is the,
link |
I'm playing with stacks of devil's advocates here
link |
and popping one devil off the stack at a time.
link |
Isn't ultimately, of course,
link |
it'll be connected to physical reality,
link |
but just because people control the physical reality,
link |
it doesn't mean the status.
link |
I guess LeBron James in theory could make more money
link |
than the owners of the teams in theory.
link |
And to me, that's a virtual idea.
link |
So somebody else constructed a game
link |
and now you're playing in the virtual space of the game.
link |
So it just feels like there could be games where status,
link |
we build realities that give us meaning in the virtual space.
link |
I can imagine such things being possible.
link |
Oh yeah, okay, so I see what you're saying.
link |
I think I see what you're saying there
link |
with the idea there, I mean, we'll take the LeBron James side
link |
and put in like some YouTube influencer.
link |
So the YouTube influencer, it is status games,
link |
but at a certain level, it precipitates into real dollars
link |
and into like, well, you look at Mr. Beast, right?
link |
He's like sending off half a million dollars
link |
worth of fireworks or something, right?
link |
Not a YouTube video.
link |
And also like saving, like saving trees and so on.
link |
Sure, right, trying to find a million trees
link |
with Mark Rober or whatever it was.
link |
Yeah, like it's not that those kinds of games
link |
can't lead to real consequences.
link |
It's that for the vast majority of people in consumer culture,
link |
they are incented by the, I would say mostly,
link |
I'm thinking about middle class consumers.
link |
They're incented by advertisements,
link |
they're scented by their memetic environment
link |
to treat the purchasing of certain things,
link |
the need to buy the latest model, whatever,
link |
the need to appear, however,
link |
the need to pursue status games as a driver of meaning.
link |
And my point would be that it's a very hollow
link |
driver of meaning.
link |
And that is what creates a meaning crisis.
link |
Because at the end of the day,
link |
it's like eating a lot of empty calories, right?
link |
Yeah, it tasted good going down, a lot of sugar,
link |
but man, it did not, it was not enough protein
link |
to help build your muscles.
link |
And you kind of feel that in your gut.
link |
And I think that's, I mean, so all this stuff aside
link |
and setting aside our discussion on currency,
link |
which I hope we get back to,
link |
that's what I mean about the meaning crisis,
link |
part of it being created by the fact that we don't,
link |
we're not encouraged to have more and more
link |
direct relationships.
link |
We're actually alienated from relating to,
link |
even our family members sometimes, right?
link |
We're encouraged to relate to brands.
link |
We're encouraged to relate to these kinds of things
link |
that then tell us to do things
link |
that are really of low consequence.
link |
And that's where the meaning crisis comes from.
link |
So the role of technology in this,
link |
so there's somebody you mentioned who's Jacques,
link |
his view of technology, he warns about the towering piles
link |
of technique, which I guess is a broad idea of technology.
link |
So I think, correct me if I'm wrong for him,
link |
technology is bad at moving away from human nature
link |
and it's ultimately is destructive.
link |
My question, broadly speaking, this meaning crisis,
link |
can technology, what are the pros and cons of technology?
link |
Yeah, I think it can be.
link |
I certainly think it can be a good thing.
link |
Can it be a good? Yeah, I think it can be.
link |
I certainly draw on some of Alol's ideas
link |
and I think some of them are pretty good.
link |
But the way he defines technique is,
link |
well, also Simondon as well.
link |
I mean, he speaks to the general mentality of efficiency,
link |
homogenized processes, homogenized production,
link |
homogenized labor to produce homogenized artifacts
link |
that then are not actually,
link |
they don't sit well in the environment.
link |
Essentially, you can think of it as the antonym of craft.
link |
Whereas a craftsman will come to a problem,
link |
maybe a piece of wood and they make into a chair.
link |
It may be a site to build a house or build a stable
link |
or build whatever.
link |
And they will consider how to bring various things in
link |
to build something well contextualized
link |
that's in right relationship with that environment.
link |
But the way we have driven technology
link |
over the last 100 and 150 years is not that at all.
link |
It is how can we make sure the input materials
link |
are homogenized, cut to the same size,
link |
diluted and doped to exactly the right alloy concentrations.
link |
How do we create machines that then consume exactly
link |
the right kind of energy to be able to run
link |
at this high speed to stamp out the same parts,
link |
which then go out the door,
link |
everyone gets the same tickle of Mielmo.
link |
And the reason why everyone wants it
link |
is because we have broadcasts that tells everyone
link |
this is the cool thing.
link |
So we homogenize demand, right?
link |
And we're like Baudrillard and other critiques
link |
of modernity coming from that direction,
link |
the situation lists as well.
link |
It's that their point is that at this point in time,
link |
consumption is the thing that drives
link |
a lot of the economic stuff, not the need,
link |
but the need to consume and build status games on top.
link |
So we have homogenized, when we discovered,
link |
I think this is really like Bernays and stuff, right?
link |
In the early 20th century, we discovered we can create,
link |
we can create demand, we can create desire
link |
in a way that was not possible before
link |
because of broadcast media.
link |
And not only do we create desire,
link |
we don't create a desire for each person
link |
to connect to some bespoke thing,
link |
to build a relationship with their neighbor or their spouse.
link |
We are telling them, you need to consume this brand,
link |
you need to drive this vehicle,
link |
you gotta listen to this music,
link |
have you heard this, have you seen this movie, right?
link |
So creating homogenized demand makes it really cheap
link |
to create homogenized product.
link |
And now you have economics of scale.
link |
So we make the same tickle me Elmo,
link |
give it to all the kids and all the kids are like,
link |
hey, I got a tickle me Elmo, right?
link |
So this is ultimately where this ties in then
link |
to runaway hypercapitalism is that we then,
link |
capitalism is always looking for growth.
link |
It's always looking for growth
link |
and growth only happens at the margins.
link |
So you have to squeeze more and more demand out.
link |
You gotta make it cheaper and cheaper
link |
to make the same thing,
link |
but tell everyone they're still getting meaning from it.
link |
You're still like, this is still your tickle me Elmo, right?
link |
And we see little bits of this dripping critiques
link |
of this dripping in popular culture.
link |
You see it sometimes it's when Buzz Lightyear
link |
walks into the thing, he's like,
link |
oh my God, at the toy store, I'm just a toy.
link |
Like there's millions of other,
link |
or there's hundreds of other Buzz Lightyear's
link |
just like me, right?
link |
That is, I think, a fun Pixar critique
link |
on this homogenization dynamic.
link |
I agree with you on most of the things you're saying.
link |
So I'm playing devil's advocate here,
link |
but this homogenized machine of capitalism
link |
is also the thing that is able to fund,
link |
if channeled correctly, innovation, invention,
link |
and development of totally new things
link |
that in the best possible world,
link |
create all kinds of new experiences that can enrich lives,
link |
the quality of lives for all kinds of people.
link |
So isn't this the machine
link |
that actually enables the experiences
link |
and more and more experiences that will then give meaning?
link |
It has done that to some extent.
link |
I mean, it's not all good or bad in my perspective.
link |
We can always look backwards
link |
and offer a critique of the path we've taken
link |
to get to this point in time.
link |
But that's a different, that's somewhat different
link |
and informs the discussion,
link |
but it's somewhat different than the question
link |
of where do we go in the future, right?
link |
Is this still the same rocket we need to ride
link |
to get to the next point?
link |
Will it even get us to the next point?
link |
Well, how does this, so you're predicting the future,
link |
how does it go wrong in your view?
link |
We have the mechanisms,
link |
we have now explored enough technologies
link |
to where we can actually, I think, sustainably produce
link |
what most people in the world need to live.
link |
We have also created the infrastructures
link |
to allow continued research and development
link |
of additional science and medicine
link |
and various other kinds of things.
link |
The organizing principles that we use
link |
to govern all these things today have been,
link |
a lot of them have been just inherited
link |
from honestly medieval times.
link |
Some of them have refactored a little bit
link |
in the industrial era,
link |
but a lot of these modes of organizing people
link |
are deeply problematic.
link |
And furthermore, they're rooted in,
link |
I think, a very industrial mode perspective on human labor.
link |
And this is one of those things,
link |
I'm gonna go back to the open source thing.
link |
There was a point in time when,
link |
well, let me ask you this.
link |
If you look at the core SciPy sort of collection of libraries,
link |
so SciPy, NumPy, Matplotlib, right?
link |
There's iPython Notebook, let's throw pandas in there,
link |
scikit learn, a few of these things.
link |
How much value do you think, economic value,
link |
would you say they drive in the world today?
link |
That's one of the fascinating things
link |
about talking to you and Travis is like,
link |
it's a measure, it's like a...
link |
At least a billion dollars a day, maybe?
link |
A billion dollars, sure.
link |
I mean, it's like, it's similar question of like,
link |
how much value does Wikipedia create?
link |
It's like, all of it, I don't know.
link |
Well, I mean, if you look at it,
link |
all of it, I don't know.
link |
Well, I mean, if you look at our systems,
link |
when you do a Google search, right?
link |
Now, some of that stuff runs through TensorFlow,
link |
but when you look at Siri,
link |
when you do credit card transaction fraud,
link |
like just everything, right?
link |
Every intelligence agency under the sun,
link |
they're using some aspect of these kinds of tools.
link |
So I would say that these create billions of dollars
link |
Oh, you mean like direct use of tools
link |
that leverage this data?
link |
Yes, direct, yeah.
link |
Yeah, even that's billions a day, yeah.
link |
Yeah, right, easily, I think.
link |
Like the things they could not do
link |
if they didn't have these tools, right?
link |
So that's billions of dollars a day, great.
link |
I think that's about right.
link |
Now, if we take, how many people did it take
link |
to make that, right?
link |
And there was a point in time, not anymore,
link |
but there was a point in time when they could fit
link |
I could have fit them in my Mercedes winter, right?
link |
And so if you look at that, like, holy crap,
link |
literally a van of maybe a dozen people
link |
could create value to the tune of billions of dollars a day.
link |
What lesson do you draw from that?
link |
Well, here's the thing.
link |
What can we do to do more of that?
link |
Like that's open source.
link |
The way I've talked about this in other environments is
link |
when we use generative participatory crowdsourced
link |
approaches, we unlock human potential
link |
at a level that is better than what capitalism can do.
link |
I would challenge anyone to go and try to hire
link |
the right 12 people in the world
link |
to build that entire stack
link |
the way those 12 people did that, right?
link |
They would be very, very hard pressed to do that.
link |
If a hedge fund could just hire a dozen people
link |
and create like something that is worth
link |
billions of dollars a day,
link |
every single one of them would be racing to do it, right?
link |
But finding the right people,
link |
fostering the right collaborations,
link |
getting it adopted by the right other people
link |
to then refine it,
link |
that is a thing that was organic in nature.
link |
That took crowdsourcing.
link |
That took a lot of the open source ethos
link |
and it took the right kinds of people, right?
link |
Now those people who started that said,
link |
I need to have a part of a multi billion dollar a day
link |
sort of enterprise.
link |
They're like, I'm doing this cool thing
link |
to solve my problem for my friends, right?
link |
So the point of telling the story
link |
is to say that our way of thinking about value,
link |
our way of thinking about allocation of resources,
link |
our ways of thinking about property rights
link |
and all these kinds of things,
link |
they come from finite game, scarcity mentality,
link |
medieval institutions.
link |
As we are now entering,
link |
to some extent we're sort of in a post scarcity era,
link |
although some people are hoarding a whole lot of stuff.
link |
We are at a point where if not now soon,
link |
we'll be in a post scarcity era.
link |
The question of how we allocate resources
link |
has to be revisited at a fundamental level
link |
because the kind of software these people built,
link |
the modalities that those human ecologies
link |
that built that software,
link |
it treats offers unproperty.
link |
Actually sharing creates value.
link |
Restricting and forking reduces value.
link |
So that's different than any other physical resource
link |
that we've ever dealt with.
link |
It's different than how most corporations
link |
treat software IP, right?
link |
So if treating software in this way
link |
created this much value so efficiently, so cheaply,
link |
because feeding a dozen people for 10 years
link |
is really cheap, right?
link |
That's the reason I care about this right now
link |
is because looking forward
link |
when we can automate a lot of labor,
link |
where we can in fact,
link |
the programming for your robot in your part,
link |
neck of the woods and your part of the Amazon
link |
to build something sustainable for you
link |
and your tribe to deliver the right medicines,
link |
to take care of the kids,
link |
that's just software, that's just code
link |
that could be totally open sourced, right?
link |
So we can actually get to a mode
link |
where all of this additional generative things
link |
that humans are doing,
link |
they don't have to be wrapped up in a container
link |
and then we charge for all the exponential dynamics
link |
That's what Facebook did.
link |
That's what modern social media did, right?
link |
Because the old internet was connecting people just fine.
link |
So Facebook came along and said,
link |
well, anyone can post a picture,
link |
anyone can post some text
link |
and we're gonna amplify the crap out of it to everyone else.
link |
And it exploded this generative network
link |
of human interaction.
link |
And then I said, how do I make money off that?
link |
Oh yeah, I'm gonna be a gatekeeper
link |
on everybody's attention.
link |
And that's how I'm gonna make money.
link |
So how do we create more than one van?
link |
How do we have millions of vans full of people
link |
that create NumPy, SciPy, that create Python?
link |
So the story of those people is often they have
link |
some kind of job outside of this.
link |
This is what they're doing for fun.
link |
Don't you need to have a job?
link |
Don't you have to be connected,
link |
plugged in to the capitalist system?
link |
isn't this consumerism,
link |
the engine that results in the individuals
link |
that kind of take a break from it every once in a while
link |
to create something magical?
link |
Like at the edges is where the innovation happens.
link |
There's a surplus, right, this is the question.
link |
Like if everyone were to go and run their own farm,
link |
no one would have time to go and write NumPy, SciPy, right?
link |
Maybe, but that's what I'm talking about
link |
when I say we're maybe at a post scarcity point
link |
for a lot of people.
link |
The question that we're never encouraged to ask
link |
in a Super Bowl ad is how much do you need?
link |
How much is enough?
link |
Do you need to have a new car every two years, every five?
link |
If you have a reliable car,
link |
can you drive one for 10 years, is that all right?
link |
I had a car for 10 years and it was fine.
link |
Your iPhone, do you have to upgrade every two years?
link |
I mean, it's sort of, you're using the same apps
link |
you did four years ago, right?
link |
This should be a Super Bowl ad.
link |
This should be a Super Bowl ad, that's great.
link |
Maybe somebody. Do you really need a new iPhone?
link |
Maybe one of our listeners will fund something like this
link |
of like, no, but just actually bringing it back,
link |
bringing it back to actually the question
link |
of what do you need?
link |
How do we create the infrastructure
link |
for collectives of people to live on the basis
link |
of providing what we need, meeting people's needs
link |
with a little bit of access to handle emergencies,
link |
things like that, pulling our resources together
link |
to handle the really, really big emergencies,
link |
somebody with a really rare form of cancer
link |
or some massive fire sweeps through half the village
link |
or whatever, but can we actually unscale things
link |
and solve for people's needs
link |
and then give them the capacity to explore
link |
how to be the best version of themselves?
link |
And for Travis, that was throwing away his shot of tenure
link |
in order to write NumPy.
link |
For others, there is a saying in the SciFi community
link |
that SciFi advances one failed postdoc at a time.
link |
And that's, we can do these things.
link |
We can actually do this kind of collaboration
link |
because code, software, information, organization,
link |
Those bits are very cheap to fling across the oceans.
link |
So you mentioned Travis.
link |
We've been talking and we'll continue to talk
link |
about open source.
link |
Maybe you can comment.
link |
How did you meet Travis?
link |
Who is Travis Aliphant?
link |
What's your relationship been like through the years?
link |
Where did you work together?
link |
What's the present and the future look like?
link |
Yeah, so the first time I met Travis
link |
was at a SciFi conference in Pasadena.
link |
Do you remember the year?
link |
I was working at, again, at nthought,
link |
working on scientific computing consulting.
link |
And a couple of years later,
link |
he joined us at nthought, I think 2007.
link |
And he came in as the president.
link |
One of the founders of nthought was the CEO, Eric Jones.
link |
And we were all very excited that Travis was joining us
link |
and that was great fun.
link |
And so I worked with Travis
link |
on a number of consulting projects
link |
and we worked on some open source stuff.
link |
I mean, it was just a really, it was a good time there.
link |
It was primarily Python related?
link |
Oh yeah, it was all Python, NumPy, SciFi consulting
link |
Towards the end of that time,
link |
we started getting called into more and more finance shops.
link |
They were adopting Python pretty heavily.
link |
I did some work on like a high frequency trading shop,
link |
working on some stuff.
link |
And then we worked together on some,
link |
at a couple of investment banks in Manhattan.
link |
And so we started seeing that there was a potential
link |
to take Python in the direction of business computing,
link |
more than just being this niche like MATLAB replacement
link |
for big vector computing.
link |
What we were seeing was, oh yeah,
link |
you could actually use Python as a Swiss army knife
link |
to do a lot of shadow data transformation kind of stuff.
link |
So that's when we realized the potential is much greater.
link |
And so we started Anaconda,
link |
I mean, it was called Continuum Analytics at the time,
link |
but we started in January of 2012
link |
with a vision of shoring up the parts of Python
link |
that needed to get expanded to handle data at scale,
link |
to do web visualization, application development, et cetera.
link |
And that was that, yeah.
link |
So he was CEO and I was president for the first five years.
link |
And then we raised some money and then the board,
link |
it was sort of put in a new CEO.
link |
They hired a kind of professional CEO.
link |
And then Travis, you laugh at that.
link |
I took over the CTO role.
link |
Travis then left after a year to do his own thing,
link |
to do Quonsight, which was more oriented
link |
around some of the bootstrap years that we did at Continuum
link |
where it was open source and consulting.
link |
It wasn't sort of like gung ho product development.
link |
And it wasn't focused on,
link |
we accidentally stumbled
link |
into the package management problem at Anaconda,
link |
but we had a lot of other visions of other technology
link |
that we built in the open source.
link |
And Travis was really trying to push,
link |
again, the frontiers of numerical computing,
link |
handling things like auto differentiation and stuff
link |
intrinsically in the open ecosystem.
link |
So I think that's kind of the direction
link |
he's working on in some of his work.
link |
We remain great friends and colleagues and collaborators,
link |
even though he's no longer day to day working at Anaconda,
link |
but he gives me a lot of feedback
link |
about this and that and the other.
link |
What's a big lesson you've learned from Travis
link |
about life or about programming or about leadership?
link |
Wow, there's a lot.
link |
Travis is a really, really good guy.
link |
He really, his heart is really in it.
link |
I've gotten that sense having to interact with him.
link |
It's so interesting.
link |
Such a good human being.
link |
He's a really good dude.
link |
And he and I, it's so interesting.
link |
We come from very different backgrounds.
link |
We're quite different as people,
link |
but I think we can like not talk for a long time
link |
and then be on a conversation
link |
and be eye to eye on like 90% of things.
link |
And so he's someone who I believe
link |
no matter how much fog settles in over the ocean,
link |
his ship, my ship are pointed
link |
sort of in the same direction of the same star.
link |
Wow, that's a beautiful way to phrase it.
link |
No matter how much fog there is,
link |
we're pointed at the same star.
link |
Yeah, and I hope he feels the same way.
link |
I mean, I hope he knows that over the years now.
link |
We both care a lot about the community.
link |
For someone who cares so deeply,
link |
I would say this about Travis that's interesting.
link |
For someone who cares so deeply about the nerd details
link |
of like type system design and vector computing
link |
and efficiency of expressing this and that and the other,
link |
memory layouts and all that stuff,
link |
he cares even more about the people
link |
in the ecosystem, the community.
link |
And I have a similar kind of alignment.
link |
I care a lot about the tech, I really do.
link |
But for me, the beauty of what this human ecology
link |
has produced is I think a touchstone.
link |
It's an early version, we can look at it and say,
link |
how do we replicate this for humanity at scale?
link |
What this open source collaboration was able to produce?
link |
How can we be generative in human collaboration
link |
moving forward and create that
link |
as a civilizational kind of dynamic?
link |
Like, can we seize this moment to do that?
link |
Because like a lot of the other open source movements,
link |
it's all nerds nerding out on code for nerds.
link |
And this because it's scientists,
link |
because it's people working on data,
link |
that all of it faces real human problems.
link |
I think we have an opportunity
link |
to actually make a bigger impact.
link |
Is there a way for this kind of open source vision
link |
To fund the people involved?
link |
Is that an essential part of it?
link |
It's hard, but we're trying to do that
link |
in our own way at Anaconda,
link |
because we know that business users,
link |
as they use more of the stuff, they have needs,
link |
like business specific needs around security, provenance.
link |
They really can't tell their VPs and their investors,
link |
hey, we're having, our data scientists
link |
are installing random packages from who knows where
link |
and running on customer data.
link |
So they have to have someone to talk to you.
link |
And that's what Anaconda does.
link |
So we are a governed source of packages for them,
link |
and that's great, that makes some money.
link |
We take some of that and we just take that as a dividend.
link |
We take a percentage of our revenues
link |
and write that as a dividend for the open source community.
link |
But beyond that, I really see the development
link |
of a marketplace for people to create notebooks,
link |
models, data sets, curation of these different kinds
link |
of things, and to really have
link |
a long tail marketplace dynamic with that.
link |
Can you speak about this problem
link |
that you stumbled into of package management,
link |
Python package management?
link |
A lot of people speak very highly of Conda,
link |
which is part of Anaconda, which is a package manager.
link |
There's a ton of packages.
link |
So first, what are package managers?
link |
And second, what was there before?
link |
And why is Conda more awesome?
link |
The package problem is this, which is that
link |
in order to do numerical computing efficiently with Python,
link |
there are a lot of low level libraries
link |
that need to be compiled, compiled with a C compiler
link |
or C++ compiler or Fortran compiler.
link |
They need to not just be compiled,
link |
but they need to be compiled with all of the right settings.
link |
And oftentimes those settings are tuned
link |
for specific chip architectures.
link |
And when you add GPUs to the mix,
link |
when you look at different operating systems,
link |
you may be on the same chip,
link |
but if you're running Mac versus Linux versus Windows
link |
on the same x86 chip, you compile and link differently.
link |
All of this complexity is beyond the capability
link |
of most data scientists to reason about.
link |
And it's also beyond what most of the package developers
link |
want to deal with too.
link |
Because if you're a package developer,
link |
you're like, I code on Linux.
link |
This works for me, I'm good.
link |
It is not my problem to figure out how to build this
link |
on an ancient version of Windows, right?
link |
That's just simply not my problem.
link |
So what we end up with is we have a creator economy
link |
or create a very creative crowdsourced environment
link |
where people want to use this stuff, but they can't.
link |
And so we ended up creating a new set of technologies
link |
like a build recipe system, a build system
link |
and an installer system that is able to,
link |
well, to put it simply,
link |
it's able to build these packages correctly
link |
on each of these different kinds of platforms
link |
and operating systems,
link |
and make it so when people want to install something,
link |
they can, it's just one command.
link |
They don't have to set up a big compiler system
link |
and do all these things.
link |
So when it works well, it works great.
link |
Now, the difficulty is we have literally thousands
link |
of people writing code in the ecosystem,
link |
building all sorts of stuff and each person writing code,
link |
they may take a dependence on something else.
link |
And so you have all this web,
link |
incredibly complex web of dependencies.
link |
So installing the correct package
link |
for any given set of packages you want,
link |
getting that right subgraph is an incredibly hard problem.
link |
And again, most data scientists
link |
don't want to think about this.
link |
They're like, I want to install NumPy and pandas.
link |
I want this version of some like geospatial library.
link |
I want this other thing.
link |
Like, why is this hard?
link |
These exist, right?
link |
And it is hard because it's, well,
link |
you're installing this on a version of Windows, right?
link |
And half of these libraries are not built for Windows
link |
or the latest version isn't available,
link |
but the old version was.
link |
And if you go to the old version of this library,
link |
that means you need to go to a different version
link |
And so the Python ecosystem,
link |
by virtue of being crowdsourced,
link |
we were able to fill a hundred thousand different niches.
link |
But then we also suffer this problem
link |
that because it's crowdsourced and no one,
link |
it's like a tragedy of the commons, right?
link |
No one really needs, wants to support
link |
their thousands of other dependencies.
link |
So we end up sort of having to do a lot of this.
link |
And of course the conda forge community
link |
also steps up as an open source community that,
link |
you know, maintain some of these recipes.
link |
That's what conda does.
link |
Now, pip is a tool that came along after conda,
link |
to some extent, it came along as an easier way
link |
for the Python developers writing Python code
link |
that didn't have as much compiled, you know, stuff.
link |
They could then install different packages.
link |
And what ended up happening in the Python ecosystem
link |
was that a lot of the core Python and web Python developers,
link |
they never ran into any of this compilation stuff at all.
link |
So even we have, you know, on video,
link |
we have Guido van Rossum saying,
link |
you know what, the scientific community's packaging problems
link |
are just too exotic and different.
link |
I mean, you're talking about Fortran compilers, right?
link |
Like you guys just need to build your own solution
link |
So the Python core Python community went
link |
and built its own sort of packaging technologies,
link |
not really contemplating the complexity
link |
of this stuff over here.
link |
And so now we have the challenge where
link |
you can pip install some things, some libraries,
link |
if you just want to get started with them,
link |
you can pip install TensorFlow and that works great.
link |
The instant you want to also install some other packages
link |
that use different versions of NumPy
link |
or some like graphics library or some OpenCV thing
link |
or some other thing, you now run into dependency hell
link |
because you cannot, you know,
link |
OpenCV can have a different version of libjpeg over here
link |
than PyTorch over here.
link |
Like they actually, they all have to use the,
link |
if you want to use GPU acceleration,
link |
they have to all use the same underlying drivers
link |
and same GPU CUDA things.
link |
So it's, it gets to be very gnarly
link |
and it's a level of technology
link |
that both the makers and the users
link |
don't really want to think too much about.
link |
And that's where you step in and try to solve this.
link |
We try to solve it.
link |
Subgraph problems.
link |
I mean, you said that you don't want to think,
link |
they don't want to think about it,
link |
but how much is it a little bit on the developer
link |
and providing them tools to be a little bit more clear
link |
of that subgraph of dependency that's necessary?
link |
It is getting to a point where we do have to think about,
link |
look, can we pull some of the most popular packages together
link |
and get them to work on a coordinated release timeline,
link |
get them to build against the same test matrix,
link |
et cetera, et cetera, right?
link |
And there is a little bit of dynamic around this,
link |
but again, it is a volunteer community.
link |
You know, people working on these different projects
link |
have their own timelines
link |
and their own things they're trying to meet.
link |
So we end up trying to pull these things together.
link |
And then it's this incredibly,
link |
and I would recommend just as a business tip,
link |
don't ever go into business
link |
where when your hard work works, you're invisible.
link |
And when it breaks because of someone else's problem,
link |
you get flagged for it.
link |
Because that's in our situation, right?
link |
When something doesn't condensate all properly,
link |
usually it's some upstream issue,
link |
but it looks like condensate is broken.
link |
It looks like, you know, Anaconda screwed something up.
link |
When things do work though, it's like, oh yeah, cool.
link |
Assuming naturally, of course,
link |
it's very easy to make that work, right?
link |
So we end up in this kind of problematic scenario,
link |
but it's okay because I think we're still,
link |
you know, our heart's in the right place.
link |
We're trying to move this forward
link |
as a community sort of affair.
link |
I think most of the people in the community
link |
also appreciate the work we've done over the years
link |
to try to move these things forward
link |
in a collaborative fashion, so.
link |
One of the subgraphs of dependencies
link |
that became super complicated
link |
is the move from Python 2 to Python 3.
link |
So there's all these ways to mess
link |
with these kinds of ecosystems of packages and so on.
link |
So I just want to ask you about that particular one.
link |
What do you think about the move from Python 2 to 3?
link |
Why did it take so long?
link |
What were, from your perspective,
link |
just seeing the packages all struggle
link |
and the community all struggle through this process,
link |
what lessons do you take away from it?
link |
Why did it take so long?
link |
Looking back, some people perhaps underestimated
link |
how much adoption Python 2 had.
link |
I think some people also underestimated how much,
link |
or they overestimated how much value
link |
some of the new features in Python 3 really provided.
link |
Like the things they really loved about Python 3
link |
just didn't matter to some of these people in Python 2.
link |
Because this change was happening as Python, SciPy,
link |
was starting to take off really like past,
link |
like a hockey stick of adoption
link |
in the early data science era, in the early 2010s.
link |
A lot of people were learning and onboarding
link |
in whatever just worked.
link |
And the teachers were like,
link |
well, yeah, these libraries I need
link |
are not supported in Python 3 yet,
link |
I'm going to teach you Python 2.
link |
Took a lot of advocacy to get people
link |
to move over to Python 3.
link |
So I think it wasn't any particular single thing,
link |
but it was one of those death by a dozen cuts,
link |
which just really made it hard to move off of Python 2.
link |
And also Python 3 itself,
link |
as they were kind of breaking things
link |
and changing things around
link |
and reorganizing the standard library,
link |
there's a lot of stuff that was happening there
link |
that kept giving people an excuse to say,
link |
I'll put off till the next version.
link |
2 is working fine enough for me right now.
link |
So I think that's essentially what happened there.
link |
And I will say this though,
link |
the strength of the Python data science movement,
link |
I think is what kept Python alive in that transition.
link |
Because a lot of languages have died
link |
and left their user bases behind.
link |
If there wasn't the use of Python for data,
link |
there's a good chunk of Python users
link |
that during that transition,
link |
would have just left for Go and Rust and stayed.
link |
In fact, some people did.
link |
They moved to Go and Rust and they just never looked back.
link |
The fact that we were able to grow by millions of users,
link |
the Python data community,
link |
that is what kept the momentum for Python going.
link |
And now the usage of Python for data is over 50%
link |
of the overall Python user base.
link |
So I'm happy to debate that on stage somewhere,
link |
I don't know if they really wanna take issue
link |
with that statement, but from where I sit,
link |
I think that's true.
link |
The statement there, the idea is that the switch
link |
from Python 2 to Python 3 would have probably
link |
destroyed Python if it didn't also coincide with Python
link |
for whatever reason,
link |
just overtaking the data science community,
link |
anything that processes data.
link |
So like the timing was perfect that this maybe
link |
imperfect decision was coupled with a great timing
link |
on the value of data in our world.
link |
I would say the troubled execution of a good decision.
link |
It was a decision that was necessary.
link |
It's possible if we had more resources,
link |
we could have done in a way that was a little bit smoother,
link |
but ultimately, the arguments for Python 3,
link |
I bought them at the time and I buy them now, right?
link |
Having great text handling is like a nonnegotiable
link |
table stakes thing you need to have in a language.
link |
So that's great, but the execution,
link |
Python is the, it's volunteer driven.
link |
It's like now the most popular language on the planet,
link |
but it's all literally volunteers.
link |
So the lack of resources meant that they had to really,
link |
they had to do things in a very hamstrung way.
link |
And I think to carry the Python momentum in the language
link |
through that time, the data movement
link |
was a critical part of that.
link |
So some of it is carrot and stick, I actually have to
link |
shamefully admit that it took me a very long time
link |
to switch from Python 2 and Python 3
link |
because I'm a machine learning person.
link |
It was just for the longest time,
link |
you could just do fine with Python 2.
link |
But I think the moment where I switched everybody
link |
I worked with and switched myself for small projects
link |
and big is when finally, when NumPy announced
link |
that they're going to end support like in 2020
link |
or something like that.
link |
So like when I realized, oh, this isn't going,
link |
this is going to end.
link |
So that's the stick, that's not a carrot.
link |
That's not, so for the longest time it was carrots.
link |
It was like all of these packages were saying,
link |
okay, we have Python 3 support now, come join us.
link |
We have Python 2 and Python 3, but when NumPy,
link |
one of the packages I sort of love and depend on
link |
said like, nope, it's over.
link |
That's when I decided to switch.
link |
I wonder if you think it was possible much earlier
link |
for somebody like NumPy or some major package
link |
to step into the cold and say like we're ending this.
link |
Well, it's a chicken and egg problem too, right?
link |
You don't want to cut off a lot of users
link |
unless you see the user momentum going too.
link |
So the decisions for the scientific community
link |
for each of the different projects,
link |
you know, there's not a monolith.
link |
Some projects are like, we'll only be releasing
link |
new features on Python 3.
link |
And that was more of a sticky carrot, right?
link |
A firm carrot, if you will, a firm carrot.
link |
A stick shaped carrot.
link |
But then for others, yeah, NumPy in particular,
link |
cause it's at the base of the dependency stack
link |
for so many things, that was the final stick.
link |
That was a stick shaped stick.
link |
People were saying, look, if I have to keep maintaining
link |
my releases for Python 2, that's that much less energy
link |
that I can put into making things better
link |
for the Python 3 folks or in my new version,
link |
which is of course going to be Python 3.
link |
So people were also getting kind of pulled by this tension.
link |
So the overall community sort of had a lot of input
link |
into when the NumPy core folks decided
link |
that they would end of life on Python 2.
link |
So as these numbers are a little bit loose,
link |
but there are about 10 million Python programmers
link |
in the world, you could argue that number,
link |
but let's say 10 million.
link |
That's actually where I was looking,
link |
said 27 million total programmers, developers in the world.
link |
You mentioned in a talk that changes need to be made
link |
for there to be 100 million Python programmers.
link |
So first of all, do you see a future
link |
where there's 100 million Python programmers?
link |
And second, what kind of changes need to be made?
link |
So Anaconda and Miniconda get downloaded
link |
about a million times a week.
link |
So I think the idea that there's only
link |
10 million Python programmers in the world
link |
is a little bit undercounting.
link |
There are a lot of people who escape traditional counting
link |
that are using Python and data in their jobs.
link |
I do believe that the future world for it to,
link |
well, the world I would like to see
link |
is one where people are data literate.
link |
So they are able to use tools
link |
that let them express their questions and ideas fluidly.
link |
And the data variety and data complexity will not go down.
link |
It will only keep increasing.
link |
So I think some level of code or code like things
link |
will continue to be relevant.
link |
And so my hope is that we can build systems
link |
that allow people to more seamlessly integrate
link |
Python kinds of expressivity with data systems
link |
and operationalization methods that are much more seamless.
link |
And what I mean by that is, you know,
link |
right now you can't punch Python code into an Excel cell.
link |
I mean, there's some tools you can do to kind of do this.
link |
We didn't build a thing for doing this back in the day,
link |
but I feel like the