back to indexKevin Scott: Microsoft CTO | Lex Fridman Podcast #30
link |
The following is a conversation with Kevin Scott,
link |
the CTO of Microsoft.
link |
Before that, he was the senior vice president
link |
of engineering and operations at LinkedIn,
link |
and before that, he oversaw mobile ads
link |
engineering at Google.
link |
He also has a podcast called Behind the Tech
link |
with Kevin Scott, which I'm a fan of.
link |
This was a fun and wide ranging conversation
link |
that covered many aspects of computing.
link |
It happened over a month ago,
link |
before the announcement of Microsoft's investment
link |
OpenAI that a few people have asked me about.
link |
I'm sure there'll be one or two people in the future
link |
that'll talk with me about the impact of that investment.
link |
This is the Artificial Intelligence podcast.
link |
If you enjoy it, subscribe on YouTube,
link |
give it five stars on iTunes,
link |
support it on a Patreon,
link |
or simply connect with me on Twitter,
link |
at Lex Freedman, spelled FRIDMAM.
link |
And I'd like to give a special thank you
link |
to Tom and Elanti Bighausen
link |
for their support of the podcast on Patreon.
link |
Thanks Tom and Elanti.
link |
Hope I didn't mess up your last name too bad.
link |
Your support means a lot,
link |
and inspires me to keep this series going.
link |
And now, here's my conversation with Kevin Scott.
link |
You've described yourself as a kid in a candy store
link |
at Microsoft because of all the interesting projects
link |
that are going on.
link |
Can you try to do the impossible task
link |
and give a brief whirlwind view
link |
of all the spaces that Microsoft is working in?
link |
Both research and product.
link |
If you include research, it becomes even more difficult.
link |
So, I think broadly speaking,
link |
Microsoft's product portfolio includes everything
link |
from big cloud business,
link |
like a big set of SaaS services.
link |
We have sort of the original,
link |
or like some of what are among the original
link |
productivity software products that everybody uses.
link |
We have an operating system business.
link |
We have a hardware business
link |
where we make everything from computer mice
link |
and headphones to high end,
link |
high end personal computers and laptops.
link |
We have a fairly broad ranging research group
link |
where we have people doing everything
link |
from economics research.
link |
So, there's this really smart young economist,
link |
Glenn Weil, who like my group works with a lot,
link |
who's doing this research on these things
link |
called radical markets.
link |
Like he's written an entire technical book
link |
about this whole notion of radical markets.
link |
So, like the research group sort of spans from that
link |
to human computer interaction, to artificial intelligence.
link |
And we have GitHub, we have LinkedIn.
link |
We have a search advertising and news business
link |
and like probably a bunch of stuff
link |
that I'm embarrassingly not recounting in this list.
link |
On gaming to Xbox and so on, right?
link |
Yeah, gaming for sure.
link |
Like I was having a super fun conversation
link |
this morning with Phil Spencer.
link |
So, when I was in college,
link |
there was this game that Lucas Arts made
link |
called Day of the Tentacle,
link |
that my friends and I played forever.
link |
And like we're doing some interesting collaboration now
link |
with the folks who made Day of the Tentacle.
link |
And I was like completely nerding out with Tim Schaeffer,
link |
like the guy who wrote Day of the Tentacle this morning,
link |
just a complete fanboy,
link |
which you know, sort of it like happens a lot.
link |
Like, you know, Microsoft has been doing so much stuff
link |
at such breadth for such a long period of time
link |
that, you know, like being CTO,
link |
like most of the time my job is very, very serious
link |
and sometimes that like I get caught up
link |
in like how amazing it is
link |
to be able to have the conversations
link |
that I have with the people I get to have them with.
link |
You had to reach back into the sentimental
link |
and what's the radical markets and the economics?
link |
So the idea with radical markets is like,
link |
can you come up with new market based mechanisms to,
link |
you know, I think we have this,
link |
we're having this debate right now,
link |
like does capitalism work, like free markets work?
link |
Can the incentive structures
link |
that are built into these systems produce outcomes
link |
that are creating sort of equitably distributed benefits
link |
for every member of society?
link |
You know, and I think it's a reasonable set of questions
link |
And so what Glenn, and so like, you know,
link |
one mode of thought there, like if you have doubts
link |
that the markets are actually working,
link |
you can sort of like tip towards like,
link |
okay, let's become more socialist
link |
and like have central planning and governments
link |
or some other central organization
link |
is like making a bunch of decisions
link |
about how sort of work gets done
link |
and like where the investments
link |
and where the outputs of those investments get distributed.
link |
Glenn's notion is like lean more
link |
into like the market based mechanism.
link |
So like for instance,
link |
this is one of the more radical ideas,
link |
like suppose that you had a radical pricing mechanism
link |
for assets like real estate
link |
where you could be bid out of your position
link |
in your home, you know, for instance.
link |
So like if somebody came along and said,
link |
you know, like I can find higher economic utility
link |
for this piece of real estate
link |
that you're running your business in,
link |
like then like you either have to, you know,
link |
sort of bid to sort of stay
link |
or like the thing that's got the higher economic utility,
link |
you know, sort of takes over the asset
link |
and which would make it very difficult
link |
to have the same sort of rent seeking behaviors
link |
that you've got right now
link |
because like if you did speculative bidding,
link |
like you would very quickly like lose a whole lot of money.
link |
And so like the prices of the assets would be sort of
link |
like very closely indexed to like the value
link |
that they can produce.
link |
And like because like you'd have this sort of real time
link |
mechanism that would force you to sort of mark the value
link |
of the asset to the market,
link |
then it could be taxed appropriately.
link |
Like you couldn't sort of sit on this thing and say,
link |
oh, like this house is only worth 10,000 bucks
link |
when like everything around it is worth 10 million.
link |
That's really interesting.
link |
So it's an incentive structure
link |
that where the prices match the value much better.
link |
And Glenn does a much, much better job than I do
link |
at selling and I probably picked the world's worst example,
link |
you know, and, and, and, but like,
link |
and it's intentionally provocative, you know,
link |
so like this whole notion, like I, you know,
link |
like I'm not sure whether I like this notion
link |
that like we can have a set of market mechanisms
link |
where I could get bid out of, out of my property, you know,
link |
but, but, you know, like if you're thinking about something
link |
like Elizabeth Warren's wealth tax, for instance,
link |
like you would have, I mean, it'd be really interesting
link |
in like how you would actually set the price on the assets.
link |
And like you might have to have a mechanism like that
link |
if you put a tax like that in place.
link |
It's really interesting that that kind of research,
link |
at least tangentially touching Microsoft research.
link |
So if you're really thinking broadly,
link |
maybe you can speak to this connects to AI.
link |
So we have a candidate, Andrew Yang,
link |
who kind of talks about artificial intelligence
link |
and the concern that people have about, you know,
link |
automations impact on society.
link |
And arguably Microsoft is at the cutting edge
link |
of innovation in all these kinds of ways.
link |
And so it's pushing AI forward.
link |
How do you think about combining all our conversations
link |
together here with radical markets and socialism
link |
and innovation in AI that Microsoft is doing?
link |
And then Andrew Yang's worry that that will,
link |
that will result in job loss for the lower and so on.
link |
How do you think about that?
link |
I think it's sort of one of the most important questions
link |
in technology, like maybe even in society right now
link |
about how is AI going to develop over the course
link |
of the next several decades
link |
and like what's it gonna be used for
link |
and like what benefits will it produce
link |
and what negative impacts will it produce
link |
and you know, who gets to steer this whole thing?
link |
You know, I'll say at the highest level,
link |
one of the real joys of getting to do what I do at Microsoft
link |
is Microsoft has this heritage as a platform company.
link |
And so, you know, like Bill has this thing
link |
that he said a bunch of years ago
link |
where the measure of a successful platform
link |
is that it produces far more economic value
link |
for the people who build on top of the platform
link |
than is created for the platform owner or builder.
link |
And I think we have to think about AI that way.
link |
Like it has to be a platform that other people can use
link |
to build businesses, to fulfill their creative objectives,
link |
to be entrepreneurs, to solve problems that they have
link |
in their work and in their lives.
link |
It can't be a thing where there are a handful of companies
link |
sitting in a very small handful of cities geographically
link |
who are making all the decisions
link |
about what goes into the AI and like,
link |
and then on top of like all this infrastructure,
link |
then build all of the commercially valuable uses for it.
link |
So like, I think like that's bad from a, you know,
link |
sort of, you know, economics
link |
and sort of equitable distribution of value perspective,
link |
like, you know, sort of back to this whole notion of,
link |
you know, like, do the markets work?
link |
But I think it's also bad from an innovation perspective
link |
because like I have infinite amounts of faith
link |
in human beings that if you, you know,
link |
give folks powerful tools, they will go do interesting things.
link |
And it's more than just a few tens of thousands of people
link |
with the interesting tools,
link |
it should be millions of people with the tools.
link |
So it's sort of like, you know,
link |
you think about the steam engine
link |
and the late 18th century, like it was, you know,
link |
maybe the first large scale substitute for human labor
link |
that we've built like a machine.
link |
And, you know, in the beginning,
link |
when these things are getting deployed,
link |
the folks who got most of the value from the steam engines
link |
were the folks who had capital
link |
so they could afford to build them.
link |
And like they built factories around them in businesses
link |
and the experts who knew how to build and maintain them.
link |
But access to that technology democratized over time.
link |
Like now like an engine is not a,
link |
it's not like a differentiated thing.
link |
Like there isn't one engine company
link |
that builds all the engines
link |
and all of the things that use engines
link |
are made by this company.
link |
And like they get all the economics from all of that.
link |
Like, no, like fully demarcated.
link |
Like they're probably, you know,
link |
we're sitting here in this room
link |
and like even though they don't,
link |
they're probably things, you know,
link |
like the MIMS gyroscope that are in both of our,
link |
like there's like little engines, you know,
link |
sort of everywhere, they're just a component
link |
in how we build the modern world.
link |
Like AI needs to get there.
link |
Yeah, so that's a really powerful way to think.
link |
If we think of AI as a platform versus a tool
link |
that Microsoft owns as a platform
link |
that enables creation on top of it,
link |
that's the way to democratize it.
link |
That's really interesting actually.
link |
And Microsoft throughout its history
link |
has been positioned well to do that.
link |
And the, you know, the tieback to this radical markets thing,
link |
like the, so my team has been working with Glenn
link |
on this and Jaren Lanier actually.
link |
So Jaren is the like the sort of father of virtual reality.
link |
Like he's one of the most interesting human beings
link |
on the planet, like a sweet, sweet guy.
link |
And so Jaren and Glenn and folks in my team
link |
have been working on this notion of data as labor
link |
or like they call it data dignity as well.
link |
And so the idea is that if you, you know,
link |
again, going back to this, you know,
link |
sort of industrial analogy,
link |
if you think about data as the raw material
link |
that is consumed by the machine of AI
link |
in order to do useful things,
link |
then like we're not doing a really great job right now
link |
in having transparent marketplaces for valuing
link |
those data contributions.
link |
So like, and we all make them like explicitly,
link |
like you go to LinkedIn,
link |
you sort of set up your profile on LinkedIn,
link |
like that's an explicit contribution.
link |
Like, you know exactly the information
link |
that you're putting into the system.
link |
And like you put it there because you have
link |
some nominal notion of like what value
link |
you're going to get in return,
link |
but it's like only nominal.
link |
Like you don't know exactly what value
link |
you're getting in return, like services free, you know,
link |
like it's low amount of like perceived.
link |
And then you've got all this indirect contribution
link |
that you're making just by virtue of interacting
link |
with all of the technology that's in your daily life.
link |
And so like what Glenn and Jaren
link |
and this data dignity team are trying to do is like,
link |
can we figure out a set of mechanisms
link |
that let us value those data contributions
link |
so that you could create an economy
link |
and like a set of controls and incentives
link |
that would allow people to like maybe even in the limit
link |
like earn part of their living
link |
through the data that they're creating.
link |
And like you can sort of see it in explicit ways.
link |
There are these companies like Scale AI
link |
and like they're a whole bunch of them in China right now
link |
that are basically data labeling companies.
link |
So like you're doing supervised machine learning,
link |
you need lots and lots of label training data.
link |
And like those people are getting like who work
link |
for those companies are getting compensated
link |
for their data contributions into the system.
link |
That's easier to put a number on their contribution
link |
because they're explicitly labeling data.
link |
But you're saying that we're all contributing data
link |
in different kinds of ways.
link |
And it's fascinating to start to explicitly try
link |
to put a number on it.
link |
Do you think that's possible?
link |
I don't know, it's hard.
link |
Because, you know, we don't have as much transparency
link |
as I think we need in like how the data is getting used.
link |
And it's, you know, super complicated.
link |
Like, you know, we, you know,
link |
I think as technologists sort of appreciate
link |
like some of the subtlety there.
link |
It's like, you know, the data, the data gets created
link |
and then it gets, you know, it's not valuable.
link |
Like the data exhaust that you give off
link |
or the, you know, the explicit data
link |
that I am putting into the system isn't valuable.
link |
It's super valuable atomically.
link |
Like it's only valuable when you sort of aggregate it together
link |
into, you know, sort of large numbers.
link |
It's true even for these like folks
link |
who are getting compensated for like labeling things.
link |
Like for supervised machine learning now,
link |
like you need lots of labels to train, you know,
link |
a model that performs well.
link |
And so, you know, I think that's one of the challenges.
link |
It's like, how do you, you know,
link |
how do you sort of figure out like
link |
because this data is getting combined in so many ways,
link |
like through these combinations,
link |
like how the value is flowing.
link |
Yeah, that's, that's fascinating.
link |
And it's fascinating that you're thinking about this.
link |
And I wasn't even going into this competition
link |
expecting the breadth of research really
link |
that Microsoft broadly is thinking about.
link |
You are thinking about in Microsoft.
link |
So if we go back to 89 when Microsoft released Office
link |
or 1990 when they released Windows 3.0,
link |
how's the, in your view,
link |
I know you weren't there the entire, you know,
link |
through its history, but how has the company changed
link |
in the 30 years since as you look at it now?
link |
The good thing is it's started off as a platform company.
link |
Like it's still a platform company,
link |
like the parts of the business that are like thriving
link |
and most successful or those that are building platforms,
link |
like the mission of the company now is,
link |
the mission's changed.
link |
It's like changing a very interesting way.
link |
So, you know, back in 89.90,
link |
like they were still on the original mission,
link |
which was like put a PC on every desk and in every home.
link |
Like, and it was basically about democratizing access
link |
to this new personal computing technology,
link |
which when Bill started the company,
link |
integrated circuit microprocessors were a brand new thing
link |
and like people were building, you know,
link |
homebrew computers, you know, from kits,
link |
like the way people build ham radios right now.
link |
And I think this is sort of the interesting thing
link |
for folks who build platforms in general.
link |
Bill saw the opportunity there
link |
and what personal computers could do.
link |
And it was like, it was sort of a reach.
link |
Like you just sort of imagined
link |
like where things were, you know,
link |
when they started the company
link |
versus where things are now.
link |
Like in success, when you democratize a platform,
link |
it just sort of vanishes into the platform.
link |
You don't pay attention to it anymore.
link |
Like operating systems aren't a thing anymore.
link |
Like they're super important, like completely critical.
link |
And like, you know, when you see one, you know, fail,
link |
like you just, you sort of understand,
link |
but like, you know, it's not a thing where you're,
link |
you're not like waiting for, you know,
link |
the next operating system thing
link |
in the same way that you were in 1995, right?
link |
Like in 1995, like, you know,
link |
we had Rolling Stones on the stage
link |
with the Windows 95 roll out.
link |
Like it was like the biggest thing in the world.
link |
Everybody would like lined up for it
link |
the way that people used to line up for iPhone.
link |
But like, you know, eventually,
link |
and like this isn't necessarily a bad thing.
link |
Like it just sort of, you know,
link |
the success is that it's sort of, it becomes ubiquitous.
link |
It's like everywhere and like human beings
link |
when their technology becomes ubiquitous,
link |
they just sort of start taking it for granted.
link |
So the mission now that Satya rearticulated
link |
five plus years ago now
link |
when he took over as CEO of the company,
link |
our mission is to empower every individual
link |
and every organization in the world to be more successful.
link |
And so, you know, again, like that's a platform mission.
link |
And like the way that we do it now is different.
link |
It's like we have a hyperscale cloud
link |
that people are building their applications on top of.
link |
Like we have a bunch of AI infrastructure
link |
that people are building their AI applications on top of.
link |
We have, you know, we have a productivity suite of software
link |
like Microsoft Dynamics, which, you know,
link |
some people might not think is the sexiest thing
link |
in the world, but it's like helping people figure out
link |
how to automate all of their business processes
link |
and workflows and to, you know, like help those businesses
link |
using it to like grow and be more successful.
link |
So it's a much broader vision in a way now
link |
than it was back then.
link |
Like it was sort of very particular thing.
link |
And like now, like we live in this world
link |
where technology is so powerful
link |
and it's like such a basic fact of life
link |
that it, you know, that it both exists
link |
and is going to get better and better over time
link |
or at least more and more powerful over time.
link |
So like, you know, what you have to do as a platform player
link |
is just much bigger.
link |
There's so many directions in which you can transform.
link |
You didn't mention mixed reality too.
link |
You know, that's probably early days
link |
or depends how you think of it.
link |
But if we think in a scale of centuries,
link |
it's the early days of mixed reality.
link |
And so yeah, with how it lands,
link |
the Microsoft is doing some really interesting work there.
link |
Do you touch that part of the effort?
link |
What's the thinking?
link |
Do you think of mixed reality as a platform too?
link |
When we look at what the platforms of the future could be.
link |
So like fairly obvious that like AI is one,
link |
like you don't have to, I mean, like that's,
link |
you know, you sort of say it to like someone
link |
and you know, like they get it.
link |
But like we also think of the like mixed reality
link |
and quantum is like these two interesting,
link |
you know, potentially.
link |
Quantum computing.
link |
Okay, so let's get crazy then.
link |
So you're talking about some futuristic things here.
link |
Well, the mixed reality Microsoft is really,
link |
it's not even futuristic, it's here.
link |
And look, and it's having an impact right now.
link |
Like one of the more interesting things
link |
that's happened with mixed reality over the past
link |
couple of years that I didn't clearly see
link |
is that it's become the computing device
link |
for folks who, for doing their work
link |
who haven't used any computing device at all
link |
to do their work before.
link |
So technicians and service folks
link |
and people who are doing like machine maintenance
link |
on factory floors.
link |
So like they, you know, because they're mobile
link |
and like they're out in the world
link |
and they're working with their hands
link |
and, you know, sort of servicing these
link |
like very complicated things.
link |
They're, they don't use their mobile phone
link |
and like they don't carry a laptop with them.
link |
And, you know, they're not tethered to a desk.
link |
And so mixed reality, like where it's getting
link |
traction right now, where HoloLens is selling
link |
a lot of units is for these sorts of applications
link |
for these workers and it's become like,
link |
I mean, like the people love it.
link |
They're like, oh my God, like this is like,
link |
for them like the same sort of productivity boosts
link |
that, you know, like an office worker had
link |
when they got their first personal computer.
link |
Yeah, but you did mention,
link |
it's certainly obvious AI as a platform,
link |
but can we dig into it a little bit?
link |
How does AI begin to infuse some of the products
link |
So currently providing training of, for example,
link |
neural networks in the cloud
link |
or providing pre trained models
link |
or just even providing computing resources
link |
and whatever different inference
link |
that you want to do using neural networks.
link |
Well, how do you think of AI infusing the,
link |
as a platform that Microsoft can provide?
link |
Yeah, I mean, I think it's, it's super interesting.
link |
It's like everywhere.
link |
And like we run these, we run these review meetings now
link |
where it's me and Satya and like members of Satya's
link |
leadership team and like a cross functional group
link |
of folks across the entire company
link |
who are working on like either AI infrastructure
link |
or like have some substantial part of their,
link |
of their product work using AI in some significant way.
link |
Now, the important thing to understand is like,
link |
when you think about like how the AI is going to manifest
link |
in like an experience for something
link |
that's going to make it better,
link |
like I think you don't want the AI in this
link |
to be the first order thing.
link |
It's like whatever the product is and like the thing
link |
that is trying to help you do,
link |
like the AI just sort of makes it better.
link |
And you know, this is a gross exaggeration,
link |
but like I, yeah, people get super excited about it.
link |
They're super excited about like where the AI is showing up
link |
in products and I'm like, do you get that excited
link |
about like where you're using a hash table like in your code?
link |
Like it's just another, it's a very interesting
link |
programming tool, but it's sort of like it's an engineering
link |
tool and so like it shows up everywhere.
link |
So like we've got dozens and dozens of features now
link |
in office that are powered by like fairly sophisticated
link |
machine learning, our search engine wouldn't work at all
link |
if you took the machine learning out of it.
link |
The like increasingly, you know,
link |
things like content moderation on our Xbox and xCloud
link |
When you mean moderation to me, like the recommender
link |
is like showing what you want to look at next.
link |
No, no, no, it's like anti bullying stuff.
link |
So the usual social network stuff that you have to deal with.
link |
But it's like really it's targeted,
link |
it's targeted towards a gaming audience.
link |
So it's like a very particular type of thing where,
link |
you know, the the line between playful banter
link |
and like legitimate bullying is like a subtle one.
link |
And like you have to, it's sort of tough.
link |
Like I have, I love to, if we could dig into it
link |
because you're also, you led the engineering efforts
link |
of LinkedIn and if we look at,
link |
if we look at LinkedIn as a social network
link |
and if we look at the Xbox gaming as the social components,
link |
the very different kinds of, I imagine communication
link |
going on on the two platforms, right?
link |
And the line in terms of bullying and so on
link |
is different on the two platforms.
link |
So how do you, I mean,
link |
such a fascinating philosophical discussion
link |
of where that line is.
link |
I don't think anyone knows the right answer.
link |
Twitter folks are under fire now,
link |
Jack at Twitter for trying to find that line.
link |
Nobody knows what that line is,
link |
but how do you try to find the line for,
link |
you know, trying to prevent abusive behavior
link |
and at the same time let people be playful
link |
and joke around and that kind of thing.
link |
I think in a certain way, like, you know,
link |
if you have what I would call vertical social networks,
link |
it gets to be a little bit easier.
link |
So like if you have a clear notion
link |
of like what your social network should be used for
link |
or like what you are designing a community around,
link |
then you don't have as many dimensions
link |
to your sort of content safety problem
link |
as, you know, as you do in a general purpose platform.
link |
I mean, so like on LinkedIn,
link |
like the whole social network is about
link |
connecting people with opportunity,
link |
whether it's helping them find a job
link |
or to, you know, sort of find mentors
link |
or to, you know, sort of help them
link |
like find their next sales lead
link |
or to just sort of allow them to broadcast
link |
their, you know, sort of professional identity
link |
to their network of peers and collaborators
link |
and, you know, sort of professional community.
link |
Like that is, I mean, like in some ways,
link |
like that's very, very broad,
link |
but in other ways, it's sort of, you know, it's narrow.
link |
And so like you can build AIs like machine learning systems
link |
that are, you know, capable with those boundaries
link |
of making better automated decisions about like,
link |
what is, you know, sort of inappropriate
link |
and offensive comment or dangerous comment
link |
or illegal content.
link |
When you have some constraints,
link |
you know, same thing with, you know,
link |
same thing with like the gaming social network.
link |
So for instance, like it's about playing games,
link |
about having fun and like the thing
link |
that you don't want to have happen on the platform.
link |
It's why bullying is such an important thing.
link |
Like bullying is not fun.
link |
So you want to do everything in your power
link |
to encourage that not to happen.
link |
And yeah, but I think that's a really important thing
link |
but I think it's sort of a tough problem in general.
link |
It's one where I think, you know,
link |
eventually we're gonna have to have
link |
some sort of clarification from our policy makers
link |
about what it is that we should be doing,
link |
like where the lines are, because it's tough.
link |
Like you don't, like in democracy, right?
link |
Like you don't want, you want some sort
link |
of democratic involvement.
link |
Like people should have a say
link |
in like where the lines are drawn.
link |
Like you don't want a bunch of people
link |
making like unilateral decisions.
link |
And like we are in a state right now
link |
for some of these platforms where you actually
link |
do have to make unilateral decisions
link |
where the policy making isn't gonna happen fast enough
link |
in order to like prevent very bad things from happening.
link |
But like we need the policy making side of that
link |
to catch up I think as quickly as possible
link |
because you want that whole process
link |
to be a democratic thing,
link |
not a, you know, not some sort of weird thing
link |
where you've got a non representative group
link |
of people making decisions that have, you know,
link |
like national and global impact.
link |
And it's fascinating because the digital space
link |
is different than the physical space
link |
in which nations and governments were established.
link |
And so what policy looks like globally,
link |
what bullying looks like globally,
link |
what healthy communication looks like globally
link |
is an open question and we're all figuring it out together.
link |
Which is fascinating.
link |
Yeah, I mean with, you know, sort of fake news for instance
link |
and deep fakes and fake news generated by humans.
link |
Yeah, so we can talk about deep fakes.
link |
Like I think that is another like, you know,
link |
sort of very interesting level of complexity.
link |
But like if you think about just the written word, right?
link |
Like we have, you know, we invented Papyrus
link |
what 3000 years ago where we, you know,
link |
you could sort of put word on paper.
link |
And then 500 years ago, like we get the printing press
link |
like where the word gets a little bit more ubiquitous.
link |
And then like you really, really didn't get ubiquitous
link |
printed word until the end of the 19th century
link |
when the offset press was invented.
link |
And then, you know, just sort of explodes
link |
and like, you know, the cross product of that
link |
and the industrial revolutions need
link |
for educated citizens resulted in like
link |
this rapid expansion of literacy
link |
and the rapid expansion of the word.
link |
But like we had 3000 years up to that point
link |
to figure out like how to, you know, like what's,
link |
what's journalism, what's editorial integrity?
link |
Like what's, you know, what's scientific peer review?
link |
And so like you built all of this mechanism
link |
to like try to filter through all of the noise
link |
that the technology made possible to like, you know,
link |
sort of getting to something that society could cope with.
link |
And like, if you think about just the piece,
link |
the PC didn't exist 50 years ago.
link |
And so in like this span of, you know,
link |
like half a century, like we've gone from no digital,
link |
you know, no ubiquitous digital technology
link |
to like having a device that sits in your pocket
link |
where you can sort of say whatever is on your mind
link |
to like what would Mary have
link |
and Mary Meeker just released her new like slide deck last week.
link |
You know, we've got 50% penetration of the internet
link |
to the global population.
link |
Like there are like three and a half billion people
link |
who are connected now.
link |
So it's like, it's crazy, crazy.
link |
They're like inconceivable,
link |
like how fast all of this happened.
link |
So, you know, it's not surprising
link |
that we haven't figured out what to do yet,
link |
but like we gotta really like lean into this set of problems
link |
because like we basically have three millennia worth of work
link |
to do about how to deal with all of this
link |
and like probably what amounts to the next decade
link |
So since we're on the topic of tough, you know,
link |
tough challenging problems,
link |
let's look at more on the tooling side in AI
link |
that Microsoft is looking at as face recognition software.
link |
So there's a lot of powerful positive use cases
link |
for face recognition, but there's some negative ones
link |
and we're seeing those in different governments
link |
So how do you, how does Microsoft think
link |
about the use of face recognition software
link |
as a platform in governments and companies?
link |
Yeah, how do we strike an ethical balance here?
link |
Yeah, I think we've articulated a clear point of view.
link |
So Brad Smith wrote a blog post last fall,
link |
I believe that sort of like outline,
link |
like very specifically what, you know,
link |
what our point of view is there.
link |
And, you know, I think we believe that there are certain uses
link |
to which face recognition should not be put
link |
and we believe again that there's a need for regulation there.
link |
Like the government should like really come in and say
link |
that, you know, this is where the lines are.
link |
And like we very much wanted to like figuring out
link |
where the lines are should be a democratic process.
link |
But in the short term, like we've drawn some lines
link |
where, you know, we push back against uses
link |
of face recognition technology.
link |
You know, like this city of San Francisco, for instance,
link |
I think has completely outlawed any government agency
link |
from using face recognition tech.
link |
And like that may prove to be a little bit overly broad.
link |
But for like certain law enforcement things,
link |
like you really, I would personally rather be overly
link |
sort of cautious in terms of restricting use of it
link |
until like we have, you know,
link |
sort of defined a reasonable, you know,
link |
democratically determined regulatory framework
link |
for like where we could and should use it.
link |
And, you know, the other thing there is
link |
like we've got a bunch of research that we're doing
link |
and a bunch of progress that we've made on bias there.
link |
And like there are all sorts of like weird biases
link |
that these models can have like all the way
link |
from like the most noteworthy one where, you know,
link |
you may have underrepresented minorities
link |
who are like underrepresented in the training data.
link |
And then you start learning like strange things.
link |
But like they're even, you know, other weird things
link |
like we've, I think we've seen in the public research
link |
like models can learn strange things
link |
like all doctors or men for instance.
link |
Yeah, I mean, and so like it really is a thing where
link |
it's very important for everybody
link |
who is working on these things before they push publish,
link |
they launch the experiment, they, you know, push the code
link |
to, you know, online or they even publish the paper
link |
that they are at least starting to think
link |
about what some of the potential negative consequences
link |
are some of this stuff.
link |
I mean, this is where, you know, like the deep fake stuff
link |
I find very worrisome just because
link |
they're going to be some very good beneficial uses
link |
of like GAN generated imagery.
link |
And like, and funny enough, like one of the places
link |
where it's actually useful is we're using the technology
link |
right now to generate synthetic, synthetic visual data
link |
for training some of the face recognition models
link |
to get rid of the bias.
link |
So like that's one like super good use of the tech,
link |
but like, you know, it's getting good enough now
link |
where, you know, it's going to sort of challenge
link |
a normal human beings ability to like now you're just sort
link |
of say like it's very expensive for someone
link |
to fabricate a photorealistic fake video.
link |
And like GANs are going to make it fantastically cheap
link |
to fabricate a photorealistic fake video.
link |
And so like what you assume you can sort of trust
link |
is true versus like be skeptical about is about to change.
link |
And like we're not ready for it, I don't think.
link |
The nature of truth, right?
link |
That's, it's also exciting because I think both you
link |
and I probably would agree that the way to solve,
link |
to take on that challenge is with technology.
link |
There's probably going to be ideas of ways to verify
link |
which kind of video is legitimate, which kind is not.
link |
So to me, that's an exciting possibility.
link |
Most likely for just the comedic genius
link |
that the internet usually creates with these kinds of videos.
link |
And hopefully will not result in any serious harm.
link |
Yeah. And it could be, you know, like I think
link |
we will have technology to that may be able to detect
link |
whether or not something's fake or real.
link |
Although the fakes are pretty convincing
link |
even like when you subject them to machine scrutiny.
link |
But, you know, we also have these increasingly
link |
interesting social networks, you know,
link |
that are under fire right now for some of the bad things
link |
Like one of the things you could choose to do
link |
with a social network is like you could,
link |
you could use crypto and the networks
link |
to like have content signed where you could have a like
link |
full chain of custody that accompanied
link |
every piece of content.
link |
So like when you're viewing something
link |
and like you want to ask yourself like how, you know,
link |
how much can I trust this?
link |
Like you can click something
link |
and like have a verified chain of custody that shows like,
link |
oh, this is coming from, you know, from this source.
link |
And it's like signed by like someone whose identity I trust.
link |
Yeah, I think having that, you know,
link |
having that chain of custody like being able to like say,
link |
oh, here's this video, like it may or may not
link |
been produced using some of this deep fake technology.
link |
But if you've got a verified chain of custody
link |
where you can sort of trace it all the way back
link |
to an identity and you can decide whether or not
link |
like I trust this identity.
link |
Like, oh no, this is really from the White House
link |
or like this is really from the, you know,
link |
the office of this particular presidential candidate
link |
or it's really from, you know,
link |
Jeff Wiener CEO of LinkedIn or Satya Nadella CEO of Microsoft.
link |
Like that might be like one way
link |
that you can solve some of the problems.
link |
So like that's not the super high tech.
link |
Like we've had all of this technology forever.
link |
And but I think you're right.
link |
Like it has to be some sort of technological thing
link |
because the underlying tech that is used to create this
link |
is not going to do anything but get better over time
link |
and the genie is sort of out of the bottle.
link |
There's no stuffing it back in.
link |
And there's a social component
link |
which I think is really healthy for democracy
link |
where people will be skeptical about the thing they watch.
link |
In general, so, you know, which is good.
link |
Skepticism in general is good for your personal content.
link |
So deep fakes in that sense are creating
link |
global skepticism about can they trust what they read?
link |
It encourages further research.
link |
I come from the Soviet Union
link |
where basically nobody trusted the media
link |
because you knew it was propaganda.
link |
And that kind of skepticism encouraged further research
link |
about ideas supposed to just trusting anyone's source.
link |
Well, like I think it's one of the reasons why the,
link |
you know, the scientific method and our apparatus
link |
of modern science is so good.
link |
Like because you don't have to trust anything.
link |
Like you, like the whole notion of, you know,
link |
like modern science beyond the fact that, you know,
link |
this is a hypothesis and this is an experiment
link |
to test the hypothesis.
link |
And, you know, like this is a peer review process
link |
for scrutinizing published results.
link |
But like stuff's also supposed to be reproducible.
link |
So like, you know, it's been vetted by this process,
link |
but like you also are expected to publish enough detail
link |
where, you know, if you are sufficiently skeptical
link |
of the thing, you can go try to like reproduce it yourself.
link |
And like, I don't know what it is.
link |
Like, I think a lot of engineers are like this
link |
where like, you know, sort of this, like your brain
link |
is sort of wired for skepticism.
link |
Like you don't just first order trust everything
link |
that you see and encounter.
link |
And like you're sort of curious to understand,
link |
you know, the next thing.
link |
But like, I think it's an entirely healthy thing.
link |
And like we need a little bit more of that right now.
link |
So I'm not a large business owner.
link |
So I'm just, I'm just a huge fan of many of Microsoft products.
link |
I mean, I still, actually in terms of,
link |
I generate a lot of graphics and images
link |
and I still use PowerPoint to do that.
link |
It beats Illustrator for me.
link |
Even professional sort of, it's fascinating.
link |
So I wonder what is the future of, let's say,
link |
windows and office look like?
link |
I mean, I remember looking forward to XP.
link |
Was it exciting when XP was released?
link |
Just like you said, I don't remember when 95 was released.
link |
But XP for me was a big celebration.
link |
And when 10 came out, I was like,
link |
okay, well, it's nice, it's a nice improvement.
link |
But so what do you see the future of these products?
link |
You know, I think there's a bunch of excitement.
link |
I mean, on the office front,
link |
there's going to be this like increasing productivity
link |
wins that are coming out of some of these AI powered features
link |
that are coming, like the products will sort of get
link |
smarter and smarter in like a very subtle way.
link |
Like there's not going to be this big bang moment
link |
where, you know, like Clippy is going to reemerge
link |
and it's going to be...
link |
Okay, well, I have to wait, wait, wait.
link |
It's Clippy coming back.
link |
Well, quite seriously.
link |
So injection of AI, there's not much,
link |
or at least I'm not familiar,
link |
sort of assistive type of stuff going on
link |
inside the office products,
link |
like a Clippy style assistant, personal assistant.
link |
Do you think that there's a possibility
link |
of that in the future?
link |
So I think there are a bunch of like very small ways
link |
in which like machine learning power
link |
and assistive things are in the product right now.
link |
So there are a bunch of interesting things,
link |
like the auto response stuff's getting better and better
link |
and it's like getting to the point where, you know,
link |
it can auto respond with like, okay,
link |
let this person is clearly trying to schedule a meeting
link |
so it looks at your calendar and it automatically
link |
like tries to find like a time and a space
link |
that's mutually interesting.
link |
Like we have this notion of Microsoft search
link |
where it's like not just web search,
link |
but it's like search across like all of your information
link |
that's sitting inside of like your Office 365 tenant
link |
and like, you know, potentially in other products.
link |
And like we have this thing called the Microsoft Graph
link |
that is basically a API federator that, you know,
link |
sort of like gets you hooked up across the entire breadth
link |
of like all of the, you know,
link |
like what were information silos
link |
before they got woven together with the graph.
link |
Like that is like getting increasing
link |
with increasing effectiveness,
link |
sort of plumbed into the,
link |
into some of these auto response things
link |
where you're going to be able to see the system
link |
like automatically retrieve information for you.
link |
Like if, you know, like I frequently send out,
link |
you know, emails to folks where like I can't find a paper
link |
or a document or whatnot.
link |
There's no reason why the system won't be able
link |
to do that for you.
link |
And like, I think the,
link |
it's building towards like having things that look more
link |
like like a fully integrated, you know, assistant,
link |
but like you'll have a bunch of steps
link |
that you will see before you,
link |
like it will not be this like big bang thing
link |
where like Clippy comes back and you've got this like,
link |
you know, manifestation of, you know,
link |
like a fully, fully powered assistant.
link |
So I think that's, that's definitely coming out.
link |
Like all of the, you know, collaboration,
link |
co authoring stuff's getting better.
link |
You know, it's like really interesting.
link |
Like if you look at how we use the office product portfolio
link |
at Microsoft, like more and more of it is happening
link |
inside of like teams as a canvas.
link |
And like it's this thing where, you know,
link |
that you've got collaboration is like
link |
at the center of the product.
link |
And like we, we, we built some like really cool stuff
link |
that's some of, which is about to be open source
link |
that are sort of framework level things for doing,
link |
for doing co authoring.
link |
So in, is there a cloud component to that?
link |
So on the web or is it,
link |
forgive me if I don't already know this,
link |
but with office 365,
link |
we still, the collaboration we do, if we're doing Word,
link |
we're still sending the file around.
link |
we're already a little bit better than that.
link |
And like, you know, so like the fact that you're unaware
link |
of it means we've got a better job to do,
link |
like helping you discover, discover this stuff.
link |
But yeah, I mean, it's already like got a huge,
link |
huge cloud component.
link |
And like part of, you know, part of this framework stuff,
link |
I think we're calling it, like I,
link |
like we've been working on it for a couple of years.
link |
So like, I know the, the internal OLA code name for it,
link |
but I think when we launched it to build,
link |
it's called the fluid framework.
link |
And, but like what fluid lets you do is like,
link |
you can go into a conversation that you're having in teams
link |
and like reference, like part of a spreadsheet
link |
that you're working on,
link |
where somebody's like sitting in the Excel canvas,
link |
like working on the spreadsheet with a, you know,
link |
And like, you can sort of embed like part of the spreadsheet
link |
in the team's conversation,
link |
where like you can dynamically update in like all
link |
of the changes that you're making to the,
link |
to this object or like, you know,
link |
coordinate and everything is sort of updating in real time.
link |
So like you can be in whatever canvas is most convenient
link |
for you to get your work done.
link |
So out of my own sort of curiosity as an engineer,
link |
I know what it's like to sort of lead a team
link |
of 10, 15 engineers.
link |
Microsoft has, I don't know what the numbers are,
link |
maybe 15, maybe 60,000 engineers, maybe 40.
link |
I don't know exactly what the number is.
link |
It's tens of thousands.
link |
Right. This is more than 10 or 15.
link |
I mean, you've led different sizes,
link |
mostly large sizes of engineers.
link |
What does it take to lead such a large group
link |
into a continue innovation,
link |
continue being highly productive
link |
and yet develop all kinds of new ideas
link |
and yet maintain like, what does it take
link |
to lead such a large group of brilliant people?
link |
I think the thing that you learn
link |
as you manage larger and larger scale
link |
is that there are three things
link |
that are like very, very important
link |
for big engineering teams.
link |
Like one is like having some sort of forethought
link |
about what it is that you're going to be building
link |
over large periods of time.
link |
Like you don't need to know that like,
link |
I'm putting all my chips on this one product
link |
and like this is going to be the thing.
link |
But it's useful to know what sort of capabilities
link |
you think you're going to need to have
link |
to build the products of the future
link |
and then like invest in that infrastructure.
link |
Like whether, and I'm not just talking about storage systems
link |
or cloud APIs, it's also like,
link |
what is your development process look like?
link |
What tools do you want?
link |
Like what culture do you want to build
link |
around like how you're sort of collaborating together
link |
to like make complicated technical things?
link |
And so like having an opinion and investing in that
link |
is like, it just gets more and more important.
link |
And like the sooner you can get a concrete set of opinions,
link |
like the better you're going to be.
link |
Like you can wing it for a while at small scales.
link |
Like, you know, when you start a company,
link |
like you don't have to be like super specific about it.
link |
But like the biggest miseries that I've ever seen
link |
as an engineering leader are in places
link |
where you didn't have a clear enough opinion
link |
about those things soon enough.
link |
And then you just sort of go create a bunch of technical debt
link |
and like culture debt that is excruciatingly painful
link |
So like that's one bundle of things.
link |
Like the other, you know, another bundle of things is
link |
like it's just really, really important to
link |
like have a clear mission that's not just some cute crap
link |
you say because like you think you should have a mission,
link |
but like something that clarifies for people
link |
like where it is that you're headed together.
link |
Like I know it's like probably
link |
like a little bit too popular right now,
link |
but Yval Harari's book, Sapiens,
link |
one of the central ideas in his book is that
link |
like storytelling is like the quintessential thing
link |
for coordinating the activities of large groups of people.
link |
Like once you get past Dunbar's number
link |
and like I've really, really seen that
link |
just managing engineering teams.
link |
Like you can just brute force things
link |
when you're less than 120, 150 folks
link |
where you can sort of know and trust
link |
and understand what the dynamics are between all the people.
link |
But like past that,
link |
like things just sort of start to catastrophically fail
link |
if you don't have some sort of set of shared goals
link |
that you're marching towards.
link |
And so like even though it sounds touchy feely
link |
and you know, like a bunch of technical people
link |
will sort of balk at the idea that like you need
link |
to like have a clear, like the missions
link |
like very, very, very important.
link |
Yval's right, right?
link |
Stories, that's how our society,
link |
that's the fabric that connects us all of us
link |
is these powerful stories.
link |
And that works for companies too, right?
link |
It works for everything.
link |
Like I mean, even down to like, you know,
link |
you sort of really think about like our currency
link |
for instance is a story.
link |
Our constitution is a story, our laws are story.
link |
I mean, like we believe very, very, very strongly in them
link |
and thank God we do.
link |
But like they are, they're just abstract things.
link |
Like they're just words.
link |
Like if we don't believe in them, they're nothing.
link |
And in some sense, those stories are platforms
link |
and the kinds some of which Microsoft is creating, right?
link |
Yeah, platforms in which we define the future.
link |
So last question, what do you,
link |
let's get philosophical maybe,
link |
bigger than even Microsoft.
link |
What do you think the next 2030 plus years
link |
looks like for computing, for technology, for devices?
link |
Do you have crazy ideas about the future of the world?
link |
Yeah, look, I think we, you know,
link |
we're entering this time where we've got,
link |
we have technology that is progressing
link |
at the fastest rate that it ever has.
link |
And you've got, you get some really big social problems
link |
like society scale problems that we have to tackle.
link |
And so, you know, I think we're gonna rise to the challenge
link |
and like figure out how to intersect
link |
like all of the power of this technology
link |
with all of the big challenges that are facing us,
link |
whether it's, you know, global warming,
link |
whether it's like the biggest remainder of the population
link |
boom is in Africa for the next 50 years or so.
link |
And like global warming is gonna make it increasingly
link |
difficult to feed the global population in particular,
link |
like in this place where you're gonna have
link |
like the biggest population boom.
link |
I think we, you know, like AI is gonna,
link |
like if we push it in the right direction,
link |
like it can do like incredible things
link |
to empower all of us to achieve our full potential
link |
and to, you know, like live better lives.
link |
But like that also means focus on like
link |
some super important things,
link |
like how can you apply it to healthcare
link |
to make sure that, you know, like our quality and cost of,
link |
and sort of ubiquity of health coverage
link |
is better and better over time.
link |
Like that's more and more important every day
link |
is like in the United States
link |
and like the rest of the industrialized world.
link |
So Western Europe, China, Japan, Korea,
link |
like you've got this population bubble
link |
of like aging working, you know, working age folks
link |
who are, you know, at some point over the next 20, 30 years
link |
they're gonna be largely retired
link |
and like you're gonna have more retired people
link |
than working age people.
link |
And then like you've got, you know,
link |
sort of natural questions about who's gonna take care
link |
of all the old folks and who's gonna do all the work.
link |
And the answers to like all of these sorts of questions
link |
like where you're sort of running into, you know,
link |
like constraints of the, you know,
link |
the world and of society has always been like
link |
what tech is gonna like help us get around this.
link |
You know, like when I was a kid in the 70s and 80s,
link |
like we talked all the time about like population boom,
link |
population boom, like we're gonna,
link |
like we're not gonna be able to like feed the planet.
link |
And like we were like right in the middle
link |
of the green revolution
link |
where like this massive technology driven increase
link |
and crop productivity like worldwide.
link |
And like some of that was like taking some of the things
link |
that we knew in the West and like getting them distributed
link |
to the, you know, to the developing world.
link |
And like part of it were things like, you know,
link |
just smarter biology like helping us increase.
link |
And like we don't talk about like, yeah,
link |
overpopulation anymore because like we can more or less,
link |
we sort of figured out how to feed the world.
link |
Like that's a technology story.
link |
And so like I'm super, super hopeful about the future
link |
and in the ways where we will be able to apply technology
link |
to solve some of these super challenging problems.
link |
Like I've, like one of the things
link |
that I'm trying to spend my time doing right now
link |
is trying to get everybody else to be hopeful
link |
as well because, you know, back to Harari,
link |
like we are the stories that we tell.
link |
Like if we, you know, if we get overly pessimistic right now
link |
about like the potential future of technology, like we,
link |
you know, like we may fail to fail to get all the things
link |
in place that we need to like have our best possible future.
link |
And that kind of hopeful optimism.
link |
I'm glad that you have it because you're leading large groups
link |
of engineers that are actually defining
link |
that are writing that story,
link |
that are helping build that future,
link |
which is super exciting.
link |
And I agree with everything you said,
link |
except I do hope Clippy comes back.
link |
I speak for the people.
link |
So, Kellen, thank you so much for talking to me.
link |
Thank you so much for having me.
link |
It was a pleasure.