Latest / a16z AI Policy Brief / The Real AI Race With China Is Who Sets the Default

Transcript

The Real AI Race With China Is Who Sets the Default · a16z AI Policy Brief

• 0:00
  SPEAKER_00: Generally,
• 0:00
  the Chinese Communist Party sees democracy as antithetical to
• 0:05
  its long-term
• 0:06
  SPEAKER_00: survival.
• 0:06
  SPEAKER_00: If you care about human rights,
• 0:09
  SPEAKER_00: democracy,
• 0:09
  SPEAKER_00: the ability to perceive objective reality,
• 0:12
  you have to be 100% committed to ensuring that America slash
• 0:16
  the West wins the AI
• 0:17
  SPEAKER_00: race.
• 0:18
  They want a reality where not only the truth is unknown,
• 0:22
  it becomes unknowable.
• 0:24
  SPEAKER_01: In this episode of the A16Z AI Policy Brief,
• 0:27
  SPEAKER_01: Jay Ramaswani sits down with Matt Cronin,
• 0:29
  Senior National Security Advisor at A16Z and a former federal
• 0:32
  prosecutor who worked
• 0:33
  SPEAKER_01: China-related issues across the Department of Justice,
• 0:36
  SPEAKER_01: the White House,
• 0:37
  SPEAKER_01: and the China Select Committee.
• 0:38
  SPEAKER_01: Recorded at A16Z's American Dynamism Summit in Washington,
• 0:42
  SPEAKER_01: D.C.,
• 0:43
  the conversation unpacks the incentives driving the Chinese
• 0:45
  Communist Party's push
• 0:46
  SPEAKER_01: to dominate AI and what's at stake.
• 0:49
  SPEAKER_01: This is an episode for you.
• 1:18
  SPEAKER_02: Today I'd like to introduce Matt Cronin to the A16Z AI policy brief.
• 1:24
  SPEAKER_02: Matt's a senior national security advisor with A16Z.
• 1:30
  SPEAKER_02: Matt, why don't you introduce yourself to the audience?
• 1:31
  SPEAKER_00: Oh, thanks, Jay.
• 1:32
  SPEAKER_00: Sure.
• 1:33
  SPEAKER_00: My name's Matt.
• 1:34
  SPEAKER_00: And in terms of background,
• 1:36
  SPEAKER_00: I was a federal prosecutor for years,
• 1:38
  SPEAKER_00: specializing and focusing on China-related issues.
• 1:41
  And then I had national security and senior roles at DOJ,
• 1:45
  main justice in D.C.,
• 1:47
  the White House.
• 1:48
  And then I was chief investigative counsel on the China Select
• 1:50
  Committee before I
• 1:50
  SPEAKER_00: came over here.
• 1:51
  SPEAKER_02: That's great.
• 1:52
  SPEAKER_02: Well, it's a pleasure to have you today.
• 1:53
  SPEAKER_02: Honored to be here.
• 1:55
  SPEAKER_02: To talk about things China-related.
• 1:59
  SPEAKER_02: I want to start off with an assumption,
• 2:02
  SPEAKER_02: and I don't think it's a far-fledged assumption,
• 2:04
  that there really are only two nations on Earth right now that
• 2:09
  have expertise and
• 2:11
  SPEAKER_02: advanced technology in AI foundation models.
• 2:16
  And I think given your background in the executive branch as
• 2:22
  well as with the China
• 2:23
  SPEAKER_02: Select Committee,
• 2:25
  SPEAKER_02: why don't you describe for us a little bit,
• 2:27
  SPEAKER_02: we hear a lot about China.
• 2:31
  and about China as being the sort of principal geopolitical
• 2:35
  rival in this space.
• 2:37
  SPEAKER_02: From what you've seen, what is motivating that rival?
• 2:41
  SPEAKER_02: We oftentimes don't really get into what's motivating the other side,
• 2:45
  but what's motivating the Chinese kind of drive to really
• 2:49
  dominate in this space?
• 2:50
  SPEAKER_00: So I can talk first generally and then get into AI.
• 2:54
  SPEAKER_00: So generally,
• 2:55
  the Chinese Communist Party sees democracy as antithetical to
• 3:00
  its long-term
• 3:01
  SPEAKER_00: survival.
• 3:02
  SPEAKER_00: This is true of other totalitarian regimes.
• 3:03
  SPEAKER_00: The Soviets were the same.
• 3:05
  Their concern is that when democracies thrive,
• 3:08
  it makes the legitimacy of their rule change.
• 3:12
  SPEAKER_00: seem fake or insincere.
• 3:14
  SPEAKER_00: So the sheer fact that Taiwan,
• 3:16
  SPEAKER_00: for instance,
• 3:16
  exists and you have Han Chinese that are thriving and doing
• 3:19
  well in a free society
• 3:20
  SPEAKER_00: undermines their right to rule,
• 3:22
  which says you need us and our authoritarian rule for you to be
• 3:25
  prosperous and
• 3:26
  SPEAKER_00: safe.
• 3:27
  SPEAKER_00: So any thriving democracy,
• 3:29
  SPEAKER_00: to a certain extent,
• 3:30
  SPEAKER_00: undermines the legitimacy of their rule,
• 3:32
  SPEAKER_00: and they want to combat that.
• 3:34
  In terms of AI,
• 3:35
  the Chinese,
• 3:36
  with that we really mean the Chinese Communist Party,
• 3:38
  the CCP,
• 3:40
  SPEAKER_00: has seen AI as absolutely essential for years, longer than we have.
• 3:44
  Chairman Xi has stated that it is a technology of a
• 3:48
  civilizational impact.
• 3:49
  SPEAKER_00: It is the centerpiece of what they call new quality productive forces,
• 3:53
  SPEAKER_00: which makes kind of the new economy for them.
• 3:56
  SPEAKER_00: And they have been investing in it,
• 3:59
  SPEAKER_00: not just merely with words,
• 4:00
  SPEAKER_00: but they've been investing in it with funding and resourcing for years,
• 4:04
  SPEAKER_00: for over a decade,
• 4:05
  SPEAKER_00: well before we were engaged in it.
• 4:06
  SPEAKER_00: Yeah.
• 4:07
  And kind of drawing out that notion that they see this as a
• 4:12
  civilizational
• 4:13
  SPEAKER_02: imperative,
• 4:13
  it's always occurred to me that one of the things that the
• 4:18
  Chinese government,
• 4:20
  and I think it's important to distinguish between because
• 4:22
  there's Chinese diaspora
• 4:23
  SPEAKER_02: all over the world,
• 4:23
  SPEAKER_02: but we're talking about the CCP estate here at this juncture,
• 4:28
  SPEAKER_02: but that the Chinese government whiffed,
• 4:30
  SPEAKER_02: in a sense,
• 4:31
  SPEAKER_02: on the internet.
• 4:32
  SPEAKER_02: Yes.
• 4:33
  SPEAKER_02: And they recognized that.
• 4:34
  It was Western and largely US-based technology and open source
• 4:39
  technology,
• 4:39
  SPEAKER_02: quite frankly,
• 4:40
  SPEAKER_02: that created everything we know about the internet.
• 4:43
  SPEAKER_02: Now there's some criticism of where the internet has gone,
• 4:45
  SPEAKER_02: but if you think about it,
• 4:47
  SPEAKER_02: it is truly a communication marvel.
• 4:49
  but one that embodies values very antithetical to those that
• 4:55
  the Chinese state
• 4:56
  SPEAKER_02: would want to impose on its population and elsewhere.
• 4:59
  SPEAKER_02: But can you kind of explain why that miss
• 5:03
  SPEAKER_02: potentially explains why they were so all in on trying to lead in AI?
• 5:08
  SPEAKER_00: Absolutely.
• 5:09
  SPEAKER_00: So when the internet came out,
• 5:11
  SPEAKER_00: as you correctly noted,
• 5:12
  SPEAKER_00: China saw it as a threat and they distanced themselves from it.
• 5:16
  SPEAKER_00: And it was only until late in the game,
• 5:17
  SPEAKER_00: around when Xi came in in 2012,
• 5:20
  SPEAKER_00: 2013,
• 5:29
  SPEAKER_00: 2014,
• 5:20
  they realized the internet was a neutral technology and could
• 5:23
  be used as a tool in
• 5:25
  SPEAKER_00: favor of oppression,
• 5:26
  SPEAKER_00: which is exactly against the original purpose of the internet.
• 5:30
  SPEAKER_00: When AI came around,
• 5:32
  SPEAKER_00: the idea of it came around at least,
• 5:34
  SPEAKER_00: kind of in universities and research centers,
• 5:36
  SPEAKER_00: they realized that
• 5:38
  SPEAKER_00: the same thing could apply here,
• 5:39
  SPEAKER_00: but what if rather it's starting than being pro-freedom,
• 5:42
  SPEAKER_00: pro-democracy,
• 5:43
  SPEAKER_00: pro-natural rights,
• 5:44
  SPEAKER_00: it instead started a pro-totalitarian bent.
• 5:47
  SPEAKER_00: And so they invested really heavily in AI to get there first,
• 5:52
  SPEAKER_00: both at the frontier level,
• 5:53
  SPEAKER_00: so the top-end technologies and models,
• 5:56
  SPEAKER_00: but also at broad diffusion,
• 5:57
  so that the world would essentially run on an AI ecosystem
• 6:01
  ultimately controlled by
• 6:02
  SPEAKER_00: the CCP.
• 6:04
  SPEAKER_00: Interestingly,
• 6:05
  SPEAKER_00: Their original investment,
• 6:06
  they put most of their eggs in the basket of kind of a human
• 6:14
  primate brain
• 6:15
  SPEAKER_00: simulacrum.
• 6:16
  SPEAKER_00: It's a very odd story.
• 6:17
  SPEAKER_00: You can get a diesel if you want.
• 6:19
  SPEAKER_00: I mean, I would love to hear this.
• 6:21
  SPEAKER_00: I'd like to talk about simulacrums.
• 6:22
  SPEAKER_00: Yeah,
• 6:22
  SPEAKER_00: I don't know if we're breaking news here,
• 6:23
  SPEAKER_00: but we got some,
• 6:25
  SPEAKER_00: you know,
• 6:25
  SPEAKER_00: sourcing came in with documents when I was on the China Select Committee.
• 6:28
  SPEAKER_00: Yeah.
• 6:29
  SPEAKER_02: And to be clear, you're not revealing any classified information.
• 6:32
  SPEAKER_00: Oh,
• 6:32
  SPEAKER_00: to be extremely clear,
• 6:33
  SPEAKER_00: nothing here is because it's all open sources available,
• 6:36
  SPEAKER_00: obtainable,
• 6:37
  SPEAKER_00: nothing from the government.
• 6:39
  But essentially what it comes down to,
• 6:41
  and you can see some research today in the open,
• 6:45
  The Chinese bet heavily on the idea of what if we can recreate
• 6:48
  the human brain
• 6:49
  SPEAKER_00: through digital means?
• 6:50
  And they did that through what could be best described as
• 6:52
  disturbing experiments
• 6:54
  SPEAKER_00: and research done on primates.
• 6:57
  And they bet so heavily on that that they kind of missed the
• 7:01
  mark on the LLMs and
• 7:03
  SPEAKER_00: other models that we were developing in the private sector.
• 7:05
  SPEAKER_00: If you had drawn an analogy,
• 7:06
  SPEAKER_00: if you think back,
• 7:07
  SPEAKER_00: you would see pictures of people trying to discover flight.
• 7:11
  SPEAKER_00: all these people had wingsuits that, oh, I'll just fly like a bird.
• 7:15
  Whereas the Wright brothers correctly intuited that they could
• 7:18
  hack physics
• 7:19
  SPEAKER_00: essentially through some changes to structures and different engineering
• 7:22
  SPEAKER_00: techniques.
• 7:23
  SPEAKER_00: We basically did the same thing in the West,
• 7:25
  SPEAKER_00: particularly United States,
• 7:26
  SPEAKER_00: where we kind of hacked physics
• 7:28
  intelligence through these kind of clever coding tricks that
• 7:32
  led us to racing ahead
• 7:33
  SPEAKER_00: of them,
• 7:34
  SPEAKER_00: even though they've invested trillions,
• 7:36
  SPEAKER_00: won trillions of dollars in AI to win that race earlier on.
• 7:39
  SPEAKER_00: And they kind of were caught flat-footed,
• 7:41
  SPEAKER_00: had to quickly pivot over to more of an LLM-based model.
• 7:44
  SPEAKER_00: They still have this kind of digital simulacrum model as well.
• 7:48
  It's interesting that you draw that distinction because I think
• 7:52
  it also speaks to a
• 7:54
  difference in the way that research early-stage companies are
• 8:00
  funded in both worlds
• 8:03
  SPEAKER_02: and draws both the strengths and the weaknesses of each system.
• 8:08
  One thing that would be great for you to discuss is it feels
• 8:11
  like one of the big
• 8:11
  SPEAKER_02: distinctions is because the Chinese system is so state-driven,
• 8:18
  you can have some of these misfires where once the state is
• 8:21
  committed to a
• 8:21
  SPEAKER_02: particular way of viewing research,
• 8:25
  SPEAKER_02: if that ends up being a dry hole,
• 8:28
  SPEAKER_02: you then have to reorient and re-pivot.
• 8:30
  SPEAKER_02: Whereas the model that's kind of,
• 8:32
  SPEAKER_02: you know,
• 8:33
  SPEAKER_02: we talk about we in the West did X and Y,
• 8:35
  SPEAKER_02: but the we is a distributed model.
• 8:37
  SPEAKER_02: It's not like a government saying you do X.
• 8:40
  It's more hundreds of entrepreneurs trying to figure out what's
• 8:43
  going on and
• 8:46
  SPEAKER_02: hitting on something.
• 8:47
  SPEAKER_02: And so,
• 8:47
  SPEAKER_02: it'll be interesting for you to sort of talk about the differences,
• 8:52
  SPEAKER_02: the strengths and weaknesses of those models.
• 8:55
  SPEAKER_02: And as we sort of pivot to what can the United States do,
• 8:58
  SPEAKER_02: what are our particular strengths that we should be leaning into?
• 9:01
  SPEAKER_00: Yeah, absolutely.
• 9:02
  SPEAKER_00: And you hit the nail on the head.
• 9:03
  SPEAKER_00: So, for China, they do have competition within markets.
• 9:09
  SPEAKER_00: And it can be extreme competition.
• 9:11
  SPEAKER_00: Extreme competition,
• 9:11
  SPEAKER_00: which led to what they currently call involution,
• 9:13
  SPEAKER_00: which is a fancy word for just hyper-productivity,
• 9:15
  SPEAKER_00: hyper-supply,
• 9:16
  SPEAKER_00: because it's all state-subsidized.
• 9:18
  SPEAKER_00: So that's why you have EVs that cost like negative $5,
• 9:21
  SPEAKER_00: because they subsidize the heck out of it,
• 9:23
  SPEAKER_00: and a bunch of people are competing over it.
• 9:25
  SPEAKER_00: So it exists to a certain degree,
• 9:27
  SPEAKER_00: but again,
• 9:27
  SPEAKER_00: the competition focuses on productivity,
• 9:29
  SPEAKER_00: like just making stuff,
• 9:30
  SPEAKER_00: pushing stuff out.
• 9:31
  SPEAKER_00: Those are the metrics.
• 9:32
  SPEAKER_00: So make an LLM, have one person use it, you get state subsidies.
• 9:37
  SPEAKER_00: And that can lead to incredible productivity,
• 9:40
  SPEAKER_00: but because there's less of a focus on profit,
• 9:42
  SPEAKER_00: which basically just means productivity that's efficient,
• 9:45
  SPEAKER_00: it leads to just all sorts of wasted capital,
• 9:48
  SPEAKER_00: inefficient models,
• 9:49
  SPEAKER_00: and a lot of dry holes.
• 9:51
  Which is like the fundamental problem of authoritarian,
• 9:52
  particularly Marxist systems.
• 9:54
  And so you will have all sorts of innovations,
• 9:58
  all sorts of things that will come through.
• 9:59
  SPEAKER_00: And some of it is really remarkable, particularly in autonomy.
• 10:02
  SPEAKER_00: They've done a lot of great work.
• 10:04
  SPEAKER_00: But it ends up being kind of,
• 10:07
  it's perverting the course of research in some sense by kind of
• 10:10
  putting too much
• 10:11
  incentives in beyond either what the market can bear or what
• 10:14
  actually is there.
• 10:15
  And so it ends up kind of going to a bunch of kind of dead ends
• 10:18
  or just no longer
• 10:19
  SPEAKER_00: being productive or helpful.
• 10:21
  On the Western side,
• 10:22
  they see what we do is extraordinarily chaotic and frankly
• 10:26
  stupid.
• 10:27
  SPEAKER_00: And there's an argument to be made to a certain extent.
• 10:29
  But because we have a dynamic system with multiple layers of
• 10:33
  competition within
• 10:35
  SPEAKER_00: certain rule of law strictures,
• 10:37
  SPEAKER_00: which they don't have on their side,
• 10:39
  SPEAKER_00: it can allow us to rapidly compete and find solutions.
• 10:43
  SPEAKER_00: It creates more aligned incentives to long-term human ingenuity.
• 10:48
  SPEAKER_02: It's interesting because there was a time when,
• 10:51
  SPEAKER_02: not too long ago,
• 10:52
  SPEAKER_02: when the Chinese were trying to develop a sort of Silicon Valley model.
• 10:58
  SPEAKER_02: And it seemed that you had venture capitalists,
• 11:01
  SPEAKER_02: you had others,
• 11:02
  SPEAKER_02: but that seems to have gone by the wayside as well.
• 11:04
  SPEAKER_02: So what accounts for that sort of transition or change?
• 11:08
  SPEAKER_00: Yeah.
• 11:08
  SPEAKER_00: So there's two things there.
• 11:09
  So one,
• 11:11
  China actually had a VC community that they very carefully
• 11:15
  cultivated for decades.
• 11:17
  SPEAKER_00: And about two or three years ago,
• 11:19
  SPEAKER_00: when a lot of debt was coming due and the economy wasn't doing so well,
• 11:22
  they made a big mistake where they allowed for all the VC firms
• 11:27
  to activate a
• 11:27
  clause that I think we can be bylaws required in these types of
• 11:30
  contracts,
• 11:31
  SPEAKER_00: or at least it was very commonly used customarily in these contracts,
• 11:35
  SPEAKER_00: where it would say,
• 11:36
  SPEAKER_00: if your company doesn't work out,
• 11:39
  SPEAKER_00: if you don't make a profit,
• 11:40
  SPEAKER_00: I can call you up and get my money back immediately.
• 11:43
  SPEAKER_00: And the government was putting, inserting those.
• 11:46
  The government allowed for state-backed VCs in China to put
• 11:51
  those clauses in.
• 11:52
  SPEAKER_00: And then when things started going south,
• 11:54
  SPEAKER_00: all these VCs started going to founders and effectively bankrupting them
• 11:57
  SPEAKER_00: immediately,
• 11:59
  SPEAKER_00: even though there was an understanding,
• 12:00
  SPEAKER_00: at least implicitly,
• 12:01
  SPEAKER_00: as is the case here,
• 12:02
  SPEAKER_00: that if you're backing a founder,
• 12:03
  SPEAKER_00: there's a good chance you're not going to get an ROI.
• 12:05
  SPEAKER_00: Some of them will be 1,000x, some will be 0x, just how it works.
• 12:09
  But in doing so,
• 12:10
  that told a lot of founders,
• 12:12
  I should not work with these VCs.
• 12:14
  SPEAKER_00: I don't want to ruin my life.
• 12:15
  SPEAKER_00: And then,
• 12:15
  SPEAKER_00: like,
• 12:15
  SPEAKER_00: you know,
• 12:16
  SPEAKER_00: collect debt from me and my family,
• 12:17
  SPEAKER_00: have the local government harass my family,
• 12:19
  SPEAKER_00: seize my bank accounts.
• 12:21
  So that really seized up a lot of VC-oriented startups,
• 12:25
  and they never really recovered.
• 12:26
  SPEAKER_00: Now, on the other side, there's stake-backed capital models.
• 12:30
  SPEAKER_00: And that, while being in many ways inefficient, has been more effective.
• 12:33
  SPEAKER_00: So in the AI space, for instance, they don't just give money.
• 12:37
  They'll make entire industrial parks and just invite in startup
• 12:39
  companies and say,
• 12:40
  SPEAKER_00: you have free power.
• 12:42
  SPEAKER_00: You have negligible cost compute.
• 12:44
  SPEAKER_00: So heavily subsidize.
• 12:45
  SPEAKER_00: Yeah, heavily subsidize.
• 12:46
  SPEAKER_00: And just make stuff.
• 12:49
  SPEAKER_00: And the other part of it is that whereas we really focus on frontier and
• 12:52
  SPEAKER_00: enterprise,
• 12:53
  SPEAKER_00: they hyper-focus on consumer,
• 12:57
  SPEAKER_00: not just because they want the consumer to do well,
• 12:59
  SPEAKER_00: because that's an area we've not focused on as much.
• 13:03
  SPEAKER_02: Meaning they're making strategic decisions as to where...
• 13:06
  SPEAKER_02: enterprises can have the biggest impact?
• 13:09
  SPEAKER_00: Yeah, kind of.
• 13:10
  SPEAKER_00: And so also,
• 13:11
  because it's more of an internet culture and kind of a
• 13:14
  smartphone native culture in
• 13:15
  SPEAKER_00: China than here,
• 13:16
  SPEAKER_00: so it's that plus the smartphone native culture,
• 13:20
  SPEAKER_00: they will try to make
• 13:22
  SPEAKER_00: a small area, like 1,000x better, due to AI.
• 13:25
  And it's all this one company,
• 13:26
  this one state-backed,
• 13:28
  directly or indirectly model will do.
• 13:30
  SPEAKER_00: And then when they do that,
• 13:31
  SPEAKER_00: that allows for this nationwide diffusion of AI,
• 13:35
  SPEAKER_00: which is a huge goal for them.
• 13:37
  SPEAKER_00: Their goal is...
• 13:39
  SPEAKER_00: To get to one of the frontier,
• 13:40
  they actually have what they call a Manhattan project for them
• 13:43
  of making a new
• 13:45
  SPEAKER_00: lithography facility.
• 13:46
  The Reuters thing reported on it earlier this year that's been
• 13:49
  technically under wraps.
• 13:50
  SPEAKER_02: Lithography facility?
• 13:51
  SPEAKER_00: Lithography.
• 13:52
  SPEAKER_00: Lithography.
• 13:53
  And so it's basically to make these chips using Dutch
• 13:56
  technology that they don't
• 13:57
  SPEAKER_00: have access to otherwise,
• 13:58
  SPEAKER_00: to recreate it.
• 13:59
  SPEAKER_00: So they're trying that.
• 14:01
  SPEAKER_00: But the other side of it is that they're trying to get broad diffusion.
• 14:03
  They think the goal is to have 70% of all industries fully
• 14:07
  adopted AI by 2030,
• 14:08
  which is crazy.
• 14:10
  SPEAKER_00: And they're really pushing for that.
• 14:13
  SPEAKER_00: The rationale for that is many times historically,
• 14:16
  SPEAKER_00: the nation that wins at the frontier,
• 14:18
  SPEAKER_00: if it doesn't win in ecosystem diffusion,
• 14:21
  SPEAKER_00: will actually ultimately lose.
• 14:22
  SPEAKER_00: So a historical example would be
• 14:26
  SPEAKER_00: The Germans were ahead of us in chemistry.
• 14:28
  They were the pioneers in all sorts of chemical processes and
• 14:31
  industrial chemistry ingenuity.
• 14:33
  SPEAKER_00: But we,
• 14:34
  SPEAKER_00: as Americans,
• 14:35
  trained everyone to a high school degree of chemical
• 14:37
  understanding and then
• 14:39
  fostered development in many dozens and dozens of industries of
• 14:42
  different chemical
• 14:43
  SPEAKER_00: processes.
• 14:43
  SPEAKER_00: We won.
• 14:44
  SPEAKER_00: They lost.
• 14:45
  SPEAKER_00: Where we won would be where we were at the frontier of electricity,
• 14:48
  but then we made sure there was broad diffusion of electricity
• 14:51
  across industries.
• 14:52
  SPEAKER_00: And the third point,
• 14:53
  SPEAKER_00: which is critical,
• 14:53
  is we went out of our way to make sure the electricity didn't
• 14:56
  just benefit elite
• 14:58
  SPEAKER_00: level.
• 14:59
  SPEAKER_00: It benefited everybody.
• 15:00
  SPEAKER_00: It helped the entirety of democracy.
• 15:02
  SPEAKER_00: When you've got a washing machine in your house, that benefited everyone.
• 15:06
  SPEAKER_00: Every single individual saw a huge gain in productivity.
• 15:10
  SPEAKER_00: So they're not necessarily doing that,
• 15:11
  but they are very much focused on ecosystem mass adoption in
• 15:14
  the country and
• 15:15
  SPEAKER_00: ultimately globally.
• 15:16
  SPEAKER_02: Yeah.
• 15:17
  SPEAKER_02: I remember the apocryphal story.
• 15:18
  I can't remember if it was Lennon or Trotsky who visited New
• 15:21
  York City in like 1905
• 15:23
  or just at the turn of the century and was astounded at the
• 15:29
  sort of availability,
• 15:30
  SPEAKER_02: the general availability of some of these technologies to create.
• 15:34
  consumers in the United States,
• 15:37
  which I think is kind of what you're- That's exactly right.
• 15:40
  SPEAKER_00: And so they're trying to get that,
• 15:42
  SPEAKER_00: but they're doing it through their own model,
• 15:44
  SPEAKER_00: which exacerbates their incredible debt problems.
• 15:47
  We can get to the next issue of why it's important for them is
• 15:49
  they see AI as their
• 15:52
  SPEAKER_00: escape hatch to get around what are really
• 15:56
  SPEAKER_00: systemic huge problems that they have.
• 15:58
  SPEAKER_00: They've been building up,
• 15:59
  SPEAKER_00: whether it's debt,
• 16:01
  SPEAKER_00: a demography time bomb,
• 16:02
  SPEAKER_00: really going to get old before they get rich,
• 16:05
  SPEAKER_00: all sorts of other problems like double digit plus unemployment,
• 16:09
  SPEAKER_00: particularly for the youth.
• 16:11
  SPEAKER_00: And so,
• 16:12
  SPEAKER_00: you know,
• 16:13
  SPEAKER_00: it's hard for us to understand that in the West because what we see,
• 16:16
  SPEAKER_00: in part due to clever propaganda,
• 16:20
  is we see China with like dancing robots and amazing 25th
• 16:24
  century cities and bullet
• 16:26
  SPEAKER_00: trains going everywhere.
• 16:28
  SPEAKER_00: But at the same time,
• 16:29
  there's over 500 million Chinese citizens that have an eighth
• 16:34
  grade or less
• 16:34
  SPEAKER_00: education.
• 16:35
  SPEAKER_00: The hukou system essentially creates… They don't have running water.
• 16:39
  SPEAKER_00: Simple things.
• 16:39
  SPEAKER_00: Yeah.
• 16:40
  SPEAKER_00: Huge environmental problems.
• 16:41
  And so they see and slash hope AI as the means to minimize
• 16:46
  those problems,
• 16:47
  SPEAKER_00: jump us to the next tier of technology and industry,
• 16:50
  SPEAKER_00: and then make that there's the global standard,
• 16:52
  SPEAKER_00: which may make them economically powerful,
• 16:55
  SPEAKER_00: but also politically powerful,
• 16:56
  SPEAKER_00: because then they'll have a pro-authoritarian bent worldwide.
• 16:58
  SPEAKER_00: Because it's a knowledge technology, not like a dumb technology.
• 17:01
  SPEAKER_02: Yeah, so let's pivot to that for a moment.
• 17:06
  SPEAKER_02: What does victory look like for the United States, for China in this?
• 17:11
  SPEAKER_02: And then I know we're both former prosecutors,
• 17:13
  SPEAKER_02: so I'm going to violate the cardinal tenet of the interview,
• 17:16
  SPEAKER_02: which I'll ask you a compound question.
• 17:19
  SPEAKER_02: Five parts.
• 17:20
  SPEAKER_02: Yeah, exactly.
• 17:22
  SPEAKER_02: But in addition to what does victory look like,
• 17:26
  SPEAKER_02: you know,
• 17:27
  SPEAKER_02: what should we be thinking of as the stakes?
• 17:32
  SPEAKER_02: Yeah.
• 17:33
  SPEAKER_02: Whether we win or not.
• 17:34
  SPEAKER_02: Because I think some people would say,
• 17:35
  SPEAKER_02: okay,
• 17:37
  SPEAKER_02: I use a Chinese vacuum cleaner and I'll use a Chinese model.
• 17:41
  SPEAKER_02: So talk to us a little bit about the stakes as well.
• 17:43
  SPEAKER_00: One, that Chinese vacuum cleaner is spying on you.
• 17:45
  SPEAKER_00: Ironically, there's a news report out from us two weeks ago.
• 17:48
  It was like a critical flaw in Chinese vacuums that allows them
• 17:51
  to map your home
• 17:52
  SPEAKER_00: and spy on you through your router.
• 17:53
  SPEAKER_00: So noting that for the public.
• 17:55
  SPEAKER_00: But yes, it's a great question.
• 17:58
  SPEAKER_00: Super important.
• 17:59
  SPEAKER_00: So start with the stakes.
• 18:01
  It's really important to understand that this is not a
• 18:04
  competition between,
• 18:05
  SPEAKER_00: say,
• 18:05
  SPEAKER_00: United States and Japan,
• 18:06
  SPEAKER_00: United States and Switzerland,
• 18:08
  SPEAKER_00: that whoever wins,
• 18:10
  SPEAKER_00: the rough parameters,
• 18:11
  SPEAKER_00: the moral underpinnings are about the same.
• 18:14
  SPEAKER_00: One country gets richer than the other, money changes hands, no big deal.
• 18:18
  I don't want to overstate it,
• 18:19
  but it really is existential as to the future of humanity.
• 18:22
  SPEAKER_00: I mean,
• 18:23
  SPEAKER_00: if you care about human rights,
• 18:25
  SPEAKER_00: democracy,
• 18:26
  SPEAKER_00: the ability to perceive objective reality,
• 18:29
  you have to be 100% committed to ensuring that America slash
• 18:32
  the West wins the AI
• 18:34
  SPEAKER_00: race.
• 18:35
  What China wants is a future where a mid-level politician or a
• 18:40
  business person or
• 18:41
  SPEAKER_00: engineer in Burkina Faso,
• 18:44
  SPEAKER_00: India,
• 18:44
  SPEAKER_00: Australia,
• 18:45
  SPEAKER_00: you name it,
• 18:46
  SPEAKER_00: will use by default their platform.
• 18:49
  SPEAKER_00: And that platform will give them information
• 18:52
  SPEAKER_00: that subtly moves them into the CCP's orbit.
• 18:56
  SPEAKER_00: This is a great deal, you should follow through on it.
• 18:58
  SPEAKER_00: Huawei's a wonderful organization, you should work with them.
• 19:00
  SPEAKER_00: What are you talking about with Tiananmen Square?
• 19:03
  SPEAKER_00: Those are just a few groups of radicals and they put it down peacefully.
• 19:07
  SPEAKER_00: That's what they want.
• 19:08
  They want a reality where not only the truth is unknown,
• 19:12
  it becomes unknowable.
• 19:14
  SPEAKER_00: And that is profoundly dangerous.
• 19:16
  We've arguably never had that sort of an existential risk to
• 19:20
  humanity in our
• 19:20
  SPEAKER_00: history as a species.
• 19:22
  SPEAKER_02: So we cannot let that happen.
• 19:23
  SPEAKER_02: Yeah, and I don't think you're overstating it.
• 19:25
  What's interesting is I think that sometimes AI is called an
• 19:30
  industry,
• 19:31
  a technology.
• 19:32
  SPEAKER_02: But as kind of we talk about in A16Z,
• 19:34
  SPEAKER_02: and I think Mark has been on record saying this,
• 19:36
  SPEAKER_02: many people have,
• 19:37
  SPEAKER_02: that AI is actually a new control layer for computing.
• 19:41
  SPEAKER_02: It's the new way that we interact with computers.
• 19:44
  SPEAKER_00: That's right.
• 19:45
  interact with them in natural language as opposed to through
• 19:48
  some sort of machine
• 19:49
  SPEAKER_02: language where we have to learn a new language to communicate with them.
• 19:52
  Now we're now communicating in the way that we normally
• 19:55
  communicate with other human beings.
• 19:58
  SPEAKER_02: But as a result, whoever controls the nature of that control layer
• 20:06
  controls,
• 20:06
  to your point,
• 20:07
  the outputs and potentially reality in how we think about
• 20:11
  things.
• 20:12
  SPEAKER_02: I sort of analogize it to when I was in the government,
• 20:16
  we were negotiating on the margins of what had been called the
• 20:21
  Budapest Convention.
• 20:22
  It was a convention on cybercrime,
• 20:24
  which I'm sure you're familiar with,
• 20:25
  which was how
• 20:27
  countries would share information in real time for cyber
• 20:31
  criminal activity.
• 20:32
  SPEAKER_02: But it was limited,
• 20:33
  and the United States worked hard to limit it to cyber crime
• 20:38
  and cyber security,
• 20:39
  SPEAKER_02: which were very narrowly defined things.
• 20:43
  the russians and the chinese at the time wanted to expand it
• 20:47
  not to be cyber
• 20:48
  security but information security and information was anything
• 20:53
  that was a threat to
• 20:54
  the party to the state um and if if that had been adopted it
• 20:58
  would have been a very
• 21:00
  very different international system that would have been
• 21:02
  created in terms of what
• 21:04
  SPEAKER_02: was being policed on the internet and it seems like
• 21:08
  SPEAKER_02: That didn't happen with the internet,
• 21:10
  but now we have this new thing that probably will become more
• 21:13
  important than the
• 21:13
  SPEAKER_02: internet and the diffusion of information.
• 21:15
  And that's what they're trying to adopt is this broader notion
• 21:18
  of information
• 21:19
  security where the state is entitled to control the types of
• 21:24
  information that
• 21:25
  SPEAKER_02: people get,
• 21:25
  SPEAKER_02: whether it's domestic consumption or international consumption.
• 21:28
  SPEAKER_00: That's exactly right.
• 21:29
  SPEAKER_00: It's a great analogy.
• 21:30
  SPEAKER_00: And it goes back to what we were saying,
• 21:32
  SPEAKER_00: kind of the top of the hour,
• 21:33
  SPEAKER_00: where for the CCP,
• 21:36
  anything about how democracy does well is a systemic challenge
• 21:40
  to their rule.
• 21:41
  SPEAKER_00: Because if you can do well in democracy, you don't need them.
• 21:44
  SPEAKER_00: And so they are absolutely terrified of that.
• 21:48
  If you look at their internal or external facing
• 21:51
  communications,
• 21:52
  they deeply care about this.
• 21:54
  And so it is absolutely essential that we don't create a world
• 21:58
  that is safe for them.
• 21:59
  SPEAKER_00: Because once they have that safe space,
• 22:00
  SPEAKER_00: they do everything they can to explore it abroad,
• 22:02
  SPEAKER_00: at least to just horrific outcomes.
• 22:04
  SPEAKER_00: I remember an unrelated investigation
• 22:07
  SPEAKER_00: If you're clever,
• 22:08
  you can find your hands on a lot of PRC procurement documents
• 22:13
  about what they're
• 22:14
  SPEAKER_00: looking for.
• 22:15
  SPEAKER_00: This is back in, I think, 2023.
• 22:17
  I still recall one of them was talking about how they were
• 22:20
  developing an AI program
• 22:23
  SPEAKER_00: that would identify people that could become a threat to the state,
• 22:26
  SPEAKER_00: which is basically this plot of Captain America the Winter Soldier.
• 22:30
  So,
• 22:31
  and it was like,
• 22:32
  they legitimately,
• 22:32
  like Hydra,
• 22:33
  were figuring out like future crime.
• 22:35
  SPEAKER_00: And they say,
• 22:36
  SPEAKER_00: so if someone,
• 22:37
  SPEAKER_00: if we took their land from them,
• 22:39
  SPEAKER_00: evicted,
• 22:40
  SPEAKER_00: forcibly evicted them,
• 22:42
  SPEAKER_00: and they're a member of a minority,
• 22:44
  or they engage in any sort of deleterious activity like
• 22:46
  graffiti or something,
• 22:49
  SPEAKER_00: just detain them.
• 22:50
  SPEAKER_00: There's like not, there's no, there's no due process.
• 22:53
  SPEAKER_00: It's just that person's a threat, put them away, disappear them.
• 22:57
  SPEAKER_00: And because they could pose a threat given certain metrics.
• 23:00
  So and then you can also identify them facial recognition
• 23:02
  anywhere in the city
• 23:02
  SPEAKER_00: within under three minutes.
• 23:05
  SPEAKER_00: That's what they are already exporting abroad to authoritarian regimes.
• 23:09
  SPEAKER_00: That's the future they want to create.
• 23:11
  SPEAKER_00: So again,
• 23:12
  SPEAKER_00: whatever issues you may have with our government or any other government,
• 23:17
  and there are like this is the beauty of being in a free
• 23:19
  society is you can see the
• 23:21
  SPEAKER_00: problems of any form of government warts and all.
• 23:25
  Just understand that you can't let the perfect be the end of
• 23:28
  the good here.
• 23:28
  SPEAKER_00: Like, sure, we have to work on that.
• 23:30
  Having a future where a US slash Western AI ecosystem are the
• 23:35
  default rails would
• 23:36
  SPEAKER_00: allow you to work through that and get to a more perfect union,
• 23:39
  SPEAKER_00: a more perfect form of humanity.
• 23:41
  SPEAKER_00: A totalitarian one is the exact opposite.
• 23:44
  SPEAKER_00: No one, aside from the CCP, wants that.
• 23:47
  SPEAKER_02: So let's pivot a little bit away from China to the U.S.
• 23:51
  SPEAKER_02: What does victory for us look like in this space?
• 23:55
  SPEAKER_02: What are the things that should be done?
• 23:58
  And I think perhaps most importantly,
• 24:02
  we've been at A16 proponents of open source.
• 24:06
  SPEAKER_02: What's at stake in the open source debate?
• 24:08
  SPEAKER_00: Yeah.
• 24:09
  SPEAKER_00: Again, a compound question upon that.
• 24:11
  SPEAKER_00: Yeah, no.
• 24:12
  SPEAKER_00: They're two great ones.
• 24:13
  SPEAKER_00: And so I would say,
• 24:16
  SPEAKER_00: so my own thoughts,
• 24:17
  SPEAKER_00: and I'm grabbing a lot from the Trump AI and executive orders,
• 24:22
  SPEAKER_00: previous executive orders,
• 24:23
  SPEAKER_00: different thoughts in think tanks or national security circles.
• 24:26
  SPEAKER_00: I would say there's three goals interrelated.
• 24:30
  SPEAKER_00: You have to achieve all three to get to the right outcome.
• 24:33
  SPEAKER_00: So one,
• 24:34
  you want to make sure that the United States has an
• 24:37
  unassailable lead in frontier
• 24:39
  SPEAKER_00: technologies.
• 24:40
  SPEAKER_00: Two,
• 24:41
  you want to make sure that the US slash the West AI ecosystem
• 24:45
  is adopted by the
• 24:46
  SPEAKER_00: majority of the world.
• 24:48
  SPEAKER_00: And that gets the open source issue.
• 24:50
  And then three,
• 24:51
  you want to make sure that these technologies,
• 24:53
  these systems,
• 24:55
  SPEAKER_00: lead to a flourishing,
• 24:56
  SPEAKER_00: not only of democracies in a broad GDP sense,
• 24:59
  SPEAKER_00: but of the general citizenry,
• 25:01
  SPEAKER_00: like electricity did for everyday Americans,
• 25:03
  SPEAKER_00: ultimately everyday people around the globe.
• 25:06
  If you can achieve all three of these,
• 25:09
  it would,
• 25:09
  I think,
• 25:11
  usher in a golden age for humanity.
• 25:13
  SPEAKER_00: Wouldn't be perfection,
• 25:14
  SPEAKER_00: there's no such thing as perfection,
• 25:16
  SPEAKER_00: but it would be a dramatic upscaling of our capabilities,
• 25:20
  SPEAKER_00: the quality of life,
• 25:22
  SPEAKER_00: and what we can do as humans.
• 25:24
  SPEAKER_02: And talk a little bit about how open source fits into that.
• 25:28
  And clearly,
• 25:28
  the Chinese are now at least at par with us or potentially
• 25:33
  leading us in that area.
• 25:34
  SPEAKER_02: But what is at stake there?
• 25:36
  And does it matter if you've got the best version versus one
• 25:41
  that everybody's using?
• 25:42
  SPEAKER_00: That's exactly right.
• 25:43
  SPEAKER_00: And so I was researching one out, I think, two or three days ago.
• 25:47
  SPEAKER_00: I was just reading this morning.
• 25:49
  SPEAKER_00: So our open source models are better,
• 25:53
  but they're not so qualitatively better that using a Chinese
• 25:58
  model doesn't make
• 25:59
  SPEAKER_00: sense if you're thinking just monetarily.
• 26:01
  That's because Chinese models cost,
• 26:03
  I think,
• 26:04
  1 40th as much on average to use per token.
• 26:07
  SPEAKER_00: And that's because the CCP subsidizes them.
• 26:10
  SPEAKER_00: So right now,
• 26:10
  you have this massive export from China of compute and data
• 26:16
  through these models
• 26:18
  SPEAKER_00: that,
• 26:18
  SPEAKER_00: by regulation,
• 26:20
  SPEAKER_00: have to be pro-CCP.
• 26:22
  SPEAKER_00: Whether or not it's activated abroad yet, it's all in there by law.
• 26:26
  SPEAKER_00: And it's going out and used throughout the rest of the world.
• 26:29
  And so you can't blame the startups for using a platform that
• 26:33
  is currently legal
• 26:34
  SPEAKER_00: and is way cheaper.
• 26:35
  Startups are by default trying to change the world and defeat a
• 26:38
  Goliath in their
• 26:39
  SPEAKER_00: own native industry.
• 26:41
  But you can fault the rest of the world for letting that happen
• 26:46
  and having similar
• 26:47
  opportunities for open source startups or open source platforms
• 26:52
  within the United
• 26:53
  SPEAKER_00: States and elsewhere.
• 26:54
  SPEAKER_02: And look, our view has changed somewhat.
• 26:56
  SPEAKER_02: I think not too long ago, there was a worry about open source.
• 27:00
  SPEAKER_00: That's right.
• 27:01
  SPEAKER_02: Now,
• 27:01
  I think it's hard to draw a direct line between kind of the
• 27:04
  uncertainty around open
• 27:05
  SPEAKER_02: source that existed and kind of Chinese leadership today,
• 27:10
  SPEAKER_02: or at least on par.
• 27:11
  SPEAKER_02: But nonetheless, it's changed.
• 27:14
  I think there was more skepticism as to whether open source
• 27:17
  itself was a national
• 27:18
  SPEAKER_02: security threat.
• 27:19
  SPEAKER_02: Absolutely.
• 27:20
  SPEAKER_02: And now it seems pretty clear that actually the lack of open source...
• 27:24
  Not leading an open source is probably the bigger national
• 27:26
  security threat.
• 27:27
  SPEAKER_00: Yeah.
• 27:28
  SPEAKER_00: For these sorts of things,
• 27:29
  you should always be really careful about definitively knowing
• 27:32
  exactly what the
• 27:33
  SPEAKER_00: right answer is and knowing the future.
• 27:35
  SPEAKER_00: You always have to be like the foxes versus the hedgehogs.
• 27:38
  SPEAKER_00: You have enough of a hedgehog to know the goal and where you're going,
• 27:42
  SPEAKER_00: but enough of a fox to know they're able to try different things,
• 27:44
  SPEAKER_00: be open to different ideas.
• 27:46
  I think the mistake was made that we know for a fact open
• 27:48
  source is bad,
• 27:49
  full stop.
• 27:50
  SPEAKER_00: That really stifle a lot of innovation.
• 27:52
  SPEAKER_00: We're trying to fix that now.
• 27:53
  SPEAKER_00: So one of the things to fix it long term was to,
• 27:58
  SPEAKER_00: for instance,
• 27:59
  SPEAKER_00: DARPA is starting to put in these projects that fosters open source model
• 28:03
  SPEAKER_00: development in specific areas.
• 28:06
  SPEAKER_00: I think doing that even more broadly would be really beneficial.
• 28:09
  SPEAKER_00: I think
• 28:10
  SPEAKER_00: doing something similar to what, you know, Roosevelt did.
• 28:12
  We created the Tennessee Valley Authority,
• 28:14
  made electricity open and plentiful to all.
• 28:16
  SPEAKER_00: You could do that for AI, but you can also do that for electricity.
• 28:19
  SPEAKER_00: You know, like we are, our grid system is 60, 70 plus years old.
• 28:25
  SPEAKER_00: It is rickety.
• 28:26
  SPEAKER_00: It is inefficient.
• 28:28
  SPEAKER_00: And right now, just
• 28:30
  Changing a few things in the margins can dramatically alter the
• 28:34
  availability of new
• 28:35
  SPEAKER_00: critical infrastructure,
• 28:36
  SPEAKER_00: new data centers,
• 28:38
  SPEAKER_00: and more abundant and cheaper power.
• 28:41
  An example would be,
• 28:42
  and Ezra Klein gets full credit from his recent book,
• 28:46
  Abundance.
• 28:48
  SPEAKER_00: Canada arguably has as good if not better environmental standards to us.
• 28:53
  SPEAKER_00: To build the same thing in Canada and the United States,
• 28:55
  SPEAKER_00: in Canada it takes about two years,
• 28:57
  SPEAKER_00: here it takes seven and a half years.
• 28:59
  SPEAKER_00: That's crazy.
• 29:00
  SPEAKER_00: There is no reason for that and it's simply down to
• 29:03
  the kind of accretion of unnecessary processes put onto
• 29:07
  statutes that were made
• 29:08
  SPEAKER_00: 50-plus years ago.
• 29:09
  And these processes have nothing to do with original
• 29:11
  congressional intent or even
• 29:13
  SPEAKER_00: the ultimate intent that benefit the citizen.
• 29:15
  SPEAKER_00: So you can,
• 29:16
  SPEAKER_00: as is the case in Canada and all over the world,
• 29:18
  you can have completely safe environments while also reducing
• 29:22
  unnecessary processes
• 29:23
  SPEAKER_00: that allow for building,
• 29:25
  SPEAKER_00: that can allow for reindustrialization in a safe way rather than
• 29:28
  SPEAKER_00: deindustrialization.
• 29:29
  SPEAKER_00: Yeah.
• 29:29
  SPEAKER_02: No, I think that's fair.
• 29:30
  The one thing I might take a little issue with is I think open
• 29:33
  source especially is
• 29:34
  SPEAKER_02: hard for it to be government directed.
• 29:36
  SPEAKER_02: In some senses,
• 29:37
  SPEAKER_02: China itself,
• 29:38
  SPEAKER_02: the Chinese Communist Party itself was caught by surprise,
• 29:41
  SPEAKER_02: I think,
• 29:41
  SPEAKER_02: by the development of DeepSeek,
• 29:43
  SPEAKER_02: which actually happened over in some corner out of a hedge fund,
• 29:46
  SPEAKER_02: not in
• 29:47
  SPEAKER_02: in the areas where the state was focused on.
• 29:51
  SPEAKER_02: And that's the way open source kind of does work.
• 29:53
  And I think also an area where I think the United States is now
• 29:57
  leading is in kind of AI agents.
• 29:58
  I mean,
• 29:59
  OpenClaw,
• 29:59
  which is now the new thing making a ton of buzz in the
• 30:03
  industry.
• 30:04
  SPEAKER_02: is an open source project.
• 30:06
  SPEAKER_02: Again,
• 30:06
  SPEAKER_02: it seems to me that that is the thing we can lean into as a country,
• 30:11
  SPEAKER_02: is that we create these things in different parts of our ecosystem,
• 30:16
  SPEAKER_02: in different parts of the stack,
• 30:18
  SPEAKER_02: and that freedom and openness,
• 30:19
  SPEAKER_02: in a sense,
• 30:20
  SPEAKER_02: is what allows us to solve new problems,
• 30:22
  SPEAKER_02: whereas a state-directed effort might not be.
• 30:26
  I think when it comes to electricity and things that are
• 30:28
  clearly regulated,
• 30:29
  SPEAKER_02: that's an area where the state...
• 30:31
  SPEAKER_02: In open source,
• 30:32
  it's a little bit harder to direct energy in the way that's
• 30:35
  going to be the most
• 30:35
  SPEAKER_02: efficient.
• 30:36
  SPEAKER_00: Yeah.
• 30:36
  SPEAKER_00: To clarify,
• 30:37
  it was more like the government should be removing friction to
• 30:41
  allow the private
• 30:42
  SPEAKER_00: sector to engage in a dynamic process.
• 30:44
  SPEAKER_00: Not say you have to say the following people are great in your model,
• 30:47
  SPEAKER_00: but more just saying,
• 30:48
  SPEAKER_00: here's resources,
• 30:50
  SPEAKER_00: go forth.
• 30:51
  SPEAKER_00: If you're not productive,
• 30:52
  SPEAKER_00: then you go out of business,
• 30:52
  SPEAKER_00: that's on you,
• 30:53
  SPEAKER_00: but we're at least giving you a good start.
• 30:55
  SPEAKER_00: Right.
• 30:56
  SPEAKER_00: In a sense, that's our secret sauce.
• 30:58
  SPEAKER_00: Exactly.
• 30:59
  SPEAKER_02: In a sense, we're playing to our strengths if we do that.
• 31:02
  SPEAKER_02: Right.
• 31:02
  SPEAKER_00: Yeah, do not try to outshine a China.
• 31:03
  SPEAKER_02: That does not work.
• 31:04
  SPEAKER_02: Okay, great.
• 31:05
  SPEAKER_02: You know, you talked a moment ago about human flourishing.
• 31:07
  SPEAKER_02: And look, we live in a cynical age.
• 31:10
  SPEAKER_02: I think there are a lot of people who now,
• 31:12
  SPEAKER_02: you know,
• 31:13
  I sort of joke that if we were having the debates of the 90s
• 31:18
  around the internet
• 31:20
  SPEAKER_02: now.
• 31:20
  SPEAKER_02: Yeah.
• 31:21
  SPEAKER_02: it would be very different.
• 31:22
  SPEAKER_02: We were a very confident society in many ways back then,
• 31:25
  SPEAKER_02: you know,
• 31:26
  SPEAKER_02: coming out of the Cold War,
• 31:27
  SPEAKER_02: unipolar world,
• 31:28
  SPEAKER_02: all of that.
• 31:29
  SPEAKER_02: There's a lot more diffidence in society today.
• 31:32
  And so,
• 31:33
  you know,
• 31:34
  I think when people hear words like human flourishing,
• 31:37
  the eyes roll.
• 31:38
  SPEAKER_02: Yeah, yeah, my own eyes roll.
• 31:40
  SPEAKER_02: I mean, look, I think it's important.
• 31:44
  I tend to believe in these things and think that things like
• 31:46
  freedom of speech,
• 31:47
  SPEAKER_02: things like freedom of expression,
• 31:49
  SPEAKER_02: freedom of association are important,
• 31:52
  SPEAKER_02: in a sense,
• 31:52
  SPEAKER_02: innovations that the US and the West brought to the table.
• 31:59
  But can you give any sort of concrete recommendations or
• 32:03
  thoughts around what would
• 32:04
  policy look like that really advanced human flourishing as
• 32:08
  opposed to increase the
• 32:10
  SPEAKER_02: authority of states and authoritarian states?
• 32:13
  SPEAKER_00: Yeah, sure.
• 32:14
  So I think it's over one of the historical analogies I went
• 32:18
  back to is thinking
• 32:19
  SPEAKER_00: about the GI Bill.
• 32:21
  SPEAKER_00: So the GI Bill was one of the most successful democratizing and kind of
• 32:27
  SPEAKER_00: wealth-distributing policies in human history,
• 32:30
  SPEAKER_00: at least in American history.
• 32:31
  At the end of the day,
• 32:32
  all it was was saying,
• 32:33
  hey,
• 32:33
  you've come back for more.
• 32:34
  SPEAKER_00: Thank you for your service.
• 32:35
  SPEAKER_00: You can go and learn nearly any trade,
• 32:39
  SPEAKER_00: whatever sort of education you want,
• 32:41
  SPEAKER_00: and then go out and just go into the economy and make your way.
• 32:45
  SPEAKER_00: And that led to an incredible explosion in productivity,
• 32:51
  SPEAKER_00: innovation,
• 32:51
  SPEAKER_00: and development of the middle class,
• 32:54
  SPEAKER_00: which is absolutely essential for democracy,
• 32:56
  SPEAKER_00: not only just to thrive,
• 32:57
  SPEAKER_00: but even to survive.
• 32:58
  SPEAKER_00: So trying to make something similar to that using AI would be,
• 33:03
  SPEAKER_00: that alone,
• 33:04
  SPEAKER_00: without going into anything more complicated,
• 33:06
  SPEAKER_00: would do a tremendous amount of good.
• 33:08
  So even now,
• 33:09
  and of course,
• 33:10
  in more advanced,
• 33:10
  we're even more capable at doing this.
• 33:14
  SPEAKER_00: AI can allow for rapid reskilling, retraining.
• 33:17
  SPEAKER_00: Anyone can reach nearly a master level of education,
• 33:21
  SPEAKER_00: particularly when it's human plus AI,
• 33:23
  SPEAKER_00: like the whole Centaur,
• 33:23
  SPEAKER_00: like mixed capability.
• 33:25
  SPEAKER_00: That would not have been possible even five years ago.
• 33:28
  SPEAKER_00: So if you really kind of tune that up and push that outward,
• 33:32
  SPEAKER_00: you could allow for huge amounts of society to gain...
• 33:36
  SPEAKER_00: huge amounts of skills that are currently gated off from colleges,
• 33:39
  SPEAKER_00: graduate schools,
• 33:40
  SPEAKER_00: trade schools,
• 33:41
  SPEAKER_00: that would have maybe to be supplemented in some ways in certain cases.
• 33:44
  SPEAKER_00: If you're a plumber,
• 33:45
  you probably want to have some sort of a brief apprenticeship
• 33:48
  rather than just
• 33:48
  SPEAKER_00: getting into the pipes in my house based purely on AI.
• 33:52
  SPEAKER_00: But, you know, it would rapidly accelerate that.
• 33:54
  SPEAKER_00: And similar to the GI Bill,
• 33:56
  SPEAKER_00: my intuition at least,
• 33:57
  SPEAKER_00: is that would lead to a greater degree of human flourishing,
• 34:01
  SPEAKER_00: which perhaps is kind of a vague term,
• 34:03
  SPEAKER_00: but more concretely,
• 34:04
  SPEAKER_00: would increase opportunity for everyday people to do things that,
• 34:08
  SPEAKER_00: because of the cost of education,
• 34:10
  SPEAKER_00: because of these gating functions,
• 34:12
  SPEAKER_00: because of geographic limitations,
• 34:13
  SPEAKER_00: they just currently can't do.
• 34:15
  SPEAKER_02: Matt,
• 34:15
  I want to pivot a little bit to an area that I know you've been
• 34:19
  working on more
• 34:20
  SPEAKER_02: recently,
• 34:21
  SPEAKER_02: which is defense procurement reform.
• 34:25
  There were some pretty interesting things that happened in the
• 34:28
  recently passed NDAA.
• 34:30
  SPEAKER_02: Sorry,
• 34:30
  SPEAKER_02: I'm speaking in Washington acronyms,
• 34:32
  SPEAKER_02: but the National Defense Authorization Act,
• 34:34
  SPEAKER_02: which happens every year.
• 34:35
  SPEAKER_02: There were some important innovations in this year's
• 34:40
  SPEAKER_02: NDAA.
• 34:43
  SPEAKER_02: Can you describe a little bit what happened there?
• 34:45
  And then we'll pivot into how that could impact AI startups and
• 34:50
  kind of the ability
• 34:51
  SPEAKER_02: of AI to be incorporated in some of the things that the government does.
• 34:54
  SPEAKER_00: Sure.
• 34:54
  SPEAKER_00: So kind of the most fundamental level
• 34:57
  There's a recognition that something gone terribly wrong where
• 34:59
  we spend nearly a
• 35:00
  SPEAKER_00: trillion dollars a year on our defense,
• 35:03
  SPEAKER_00: but simultaneously we would,
• 35:06
  SPEAKER_00: public reporting,
• 35:06
  SPEAKER_00: open source calculations,
• 35:08
  would run out of critical munitions with about two weeks of
• 35:11
  major conflict against
• 35:12
  SPEAKER_00: a peer adversary.
• 35:13
  SPEAKER_00: That's completely crazy.
• 35:14
  SPEAKER_00: You're spending the most on military of any country in human history,
• 35:18
  SPEAKER_00: but you run out of things to shoot in two weeks.
• 35:20
  SPEAKER_00: Something's gone horribly wrong.
• 35:22
  So when you dove under the hood,
• 35:23
  we realized there are a few fundamental problems.
• 35:26
  SPEAKER_00: What it all came down to was
• 35:29
  SPEAKER_00: The Defense Department,
• 35:31
  SPEAKER_00: now it's the Department of War,
• 35:32
  SPEAKER_00: starting in the 1960s,
• 35:35
  SPEAKER_00: kind of created a very centralized,
• 35:38
  SPEAKER_00: regulatory heavy,
• 35:40
  SPEAKER_00: compliance heavy system that at the end of the day,
• 35:43
  SPEAKER_00: it's kind of ultimate evolution,
• 35:44
  SPEAKER_00: made it so that it was separated out from the broader economy,
• 35:48
  SPEAKER_00: broader industry.
• 35:50
  So that it was only five or so companies that had to be custom
• 35:53
  built to do business
• 35:55
  SPEAKER_00: with the Pentagon.
• 35:55
  No one else could engage in the necessary regulatory compliance
• 36:00
  hoops to do business.
• 36:02
  SPEAKER_00: That is an absurd result,
• 36:04
  SPEAKER_00: particularly where,
• 36:05
  SPEAKER_00: as is the case in our country,
• 36:07
  SPEAKER_00: the majority of efficiency,
• 36:09
  SPEAKER_00: innovation,
• 36:10
  SPEAKER_00: productivity,
• 36:11
  SPEAKER_00: and industry is outside of defense.
• 36:13
  SPEAKER_00: Right.
• 36:13
  Now,
• 36:13
  to be fair,
• 36:14
  that's a change in the structure of a kind of technology,
• 36:18
  right?
• 36:19
  Because there was a time post-World War II where the government
• 36:23
  was actually the
• 36:24
  SPEAKER_02: biggest procurer of,
• 36:27
  SPEAKER_02: and then behind the government,
• 36:29
  SPEAKER_02: probably corporations,
• 36:30
  SPEAKER_02: but the consumer was an afterthought.
• 36:33
  I think that there's even some stories from the 1950s or 60s
• 36:37
  where Digital
• 36:38
  SPEAKER_02: Equipment Corporation,
• 36:39
  SPEAKER_02: who used to create those big mainframe computers,
• 36:41
  claimed that there would only be a need for five computers in
• 36:43
  the world ever.
• 36:45
  SPEAKER_02: It's apocryphal, I think, or not.
• 36:50
  SPEAKER_02: And so there was a reason that the structure existed that way,
• 36:53
  which was that really the government was the 800-pound gorilla
• 36:56
  in the room.
• 36:57
  SPEAKER_02: But that's changed now.
• 36:59
  And in fact,
• 37:00
  consumer applications are probably far more advanced than the
• 37:03
  government.
• 37:04
  So I assume that that's why we need to reorient and rethink why
• 37:09
  commercial sort of
• 37:11
  needs should be or commercial uses should be relevant for the
• 37:15
  defense sector.
• 37:15
  SPEAKER_00: That's right.
• 37:16
  SPEAKER_00: I mean, there's I don't know the exact iPhone it was.
• 37:19
  SPEAKER_00: Something like the iPhone 12 has more computing power than an F-35.
• 37:22
  Something's gone in a really weird direction if your top-level
• 37:26
  Gen 5 fighter has
• 37:27
  SPEAKER_00: less compute power than an iPhone that came out over 10 years ago.
• 37:32
  And I'll also note,
• 37:33
  you're really correct in terms of the evolution of it.
• 37:36
  But what's interesting is that originally,
• 37:38
  the Defense Department,
• 37:39
  the Pentagon,
• 37:39
  created Silicon Valley and created it based upon what was then
• 37:44
  the incentive
• 37:44
  SPEAKER_00: structure,
• 37:45
  SPEAKER_00: which was kind of fail fast,
• 37:47
  SPEAKER_00: make things quickly,
• 37:48
  SPEAKER_00: to develop rapid prototyping,
• 37:50
  SPEAKER_00: get things to market as fast as possible,
• 37:52
  SPEAKER_00: learn from there.
• 37:53
  The Pentagon then went in a different direction under McNamara
• 37:56
  and went to this
• 37:56
  SPEAKER_00: very hyper-cautious approach.
• 37:58
  SPEAKER_00: And you've written about this.
• 37:59
  SPEAKER_00: Yes, I have.
• 38:00
  SPEAKER_00: I have, unfortunately.
• 38:01
  SPEAKER_00: Not to plug you.
• 38:01
  SPEAKER_00: Yeah.
• 38:02
  SPEAKER_00: I've written about it.
• 38:03
  SPEAKER_00: Feel free to find it online.
• 38:05
  So McNamara took the Pentagon a very different direction,
• 38:08
  actually mimicking the Soviets.
• 38:10
  SPEAKER_00: So at the time after Sputnik,
• 38:12
  SPEAKER_00: everyone thought that the Soviet model,
• 38:14
  SPEAKER_00: Central Command Authority,
• 38:16
  SPEAKER_00: was better.
• 38:17
  SPEAKER_00: So everyone went in that direction.
• 38:18
  SPEAKER_00: The U.S.
• 38:19
  SPEAKER_00: auto industry.
• 38:20
  SPEAKER_00: and then the Pentagon,
• 38:21
  the Silicon Valley went in a different direction based upon how
• 38:24
  the Pentagon used
• 38:24
  SPEAKER_00: to think.
• 38:25
  SPEAKER_00: And it continued in that way,
• 38:27
  which in part led to all the revolutions in technology in the
• 38:30
  private sector,
• 38:31
  SPEAKER_00: whereas the Pentagon just kept slowed down,
• 38:32
  SPEAKER_00: slowed down,
• 38:33
  SPEAKER_00: slowed down.
• 38:34
  Something actually very poetic about having in some ways like
• 38:37
  the child of the
• 38:37
  SPEAKER_00: Pentagon coming back to the Pentagon in a moment of crisis and saying,
• 38:41
  SPEAKER_00: you need to learn,
• 38:41
  SPEAKER_00: come back to how you used to do things.
• 38:43
  SPEAKER_00: You need to learn what you used to know.
• 38:45
  SPEAKER_00: And that's really in part what happened here.
• 38:48
  SPEAKER_00: leading to,
• 38:49
  SPEAKER_00: again,
• 38:49
  SPEAKER_00: not every reform was put through,
• 38:51
  but many ones that were really critically important to
• 38:54
  expanding the defense
• 38:55
  industrial base and increasing the capacity to not only develop
• 38:58
  munitions,
• 38:58
  SPEAKER_00: but onboard commercial technologies to assist the warfighter.
• 39:02
  SPEAKER_02: And with that in mind,
• 39:03
  SPEAKER_02: how can AI or how will AI startups,
• 39:07
  AI sort of technologies benefit from these kinds of innovations
• 39:13
  that have taken
• 39:14
  SPEAKER_02: place in the procurement space?
• 39:15
  SPEAKER_02: And more importantly,
• 39:16
  how will the US defense industrial base benefit from that
• 39:21
  potentially new avenue of
• 39:23
  SPEAKER_02: acquisition and procurement?
• 39:24
  SPEAKER_00: Sure.
• 39:24
  So the chief goal was to make it so that if you wanted to do
• 39:27
  business with the Pentagon,
• 39:29
  you don't have to add on a 20% cost because of bizarre
• 39:34
  compliance burdens that have
• 39:35
  SPEAKER_00: no actual bearing on helping the American taxpayer,
• 39:38
  SPEAKER_00: having the solution be more safe.
• 39:41
  SPEAKER_00: We've achieved a good part of that.
• 39:43
  SPEAKER_00: In doing so,
• 39:44
  that opens up the Pentagon to companies that otherwise could
• 39:48
  not afford or have no
• 39:49
  interest in putting on these massive compliance and auditing
• 39:52
  burdens that,
• 39:53
  SPEAKER_00: again,
• 39:53
  SPEAKER_00: have no actual purpose.
• 39:55
  SPEAKER_00: So based upon that process,
• 39:58
  a lot of commercial AI companies now can onboard and have
• 40:01
  access to the Pentagon.
• 40:02
  SPEAKER_00: And the Pentagon,
• 40:04
  SPEAKER_00: in addition,
• 40:05
  SPEAKER_00: is now procuring not just for a specific system,
• 40:09
  SPEAKER_00: but for a solution.
• 40:10
  SPEAKER_00: That might sound simple, but it's a huge difference.
• 40:14
  SPEAKER_00: What does that mean?
• 40:15
  SPEAKER_00: Can you put some meat on the bones here?
• 40:18
  SPEAKER_00: Yeah.
• 40:19
  So the default way I work for the Pentagon would be saying,
• 40:22
  I need the following item.
• 40:25
  SPEAKER_00: And it would give an incredibly long list that in reality would,
• 40:29
  SPEAKER_00: in many cases,
• 40:30
  SPEAKER_00: be written by a defense prime saying,
• 40:32
  SPEAKER_00: I need these things.
• 40:33
  SPEAKER_00: And then it would go to the public and saying,
• 40:34
  SPEAKER_00: who can make this exact thing that this guy told me to write?
• 40:37
  SPEAKER_00: And then the guy who told it to write would be like, I
• 40:39
  SPEAKER_00: If I can, they say, oh, great.
• 40:40
  SPEAKER_00: And then after many years of procurement hoops, they get onboarded.
• 40:44
  SPEAKER_00: If instead you have a solution oriented,
• 40:46
  SPEAKER_00: you say,
• 40:47
  SPEAKER_00: I need to find a way to keep military bases safe from drones.
• 40:51
  SPEAKER_00: Who can do that?
• 40:52
  SPEAKER_00: I don't care how.
• 40:53
  SPEAKER_00: It could be through microwave emissions.
• 40:55
  SPEAKER_00: It could be through kinetic strikes.
• 40:57
  SPEAKER_00: It could be through hacking.
• 40:58
  SPEAKER_00: Just offer me something.
• 41:00
  SPEAKER_00: That's a simple example of where all of a sudden,
• 41:03
  SPEAKER_00: it's not written for one company,
• 41:04
  SPEAKER_00: it's written for just anybody.
• 41:06
  SPEAKER_00: You have these bake-offs,
• 41:07
  people can come in and just go through a series of challenges
• 41:09
  to see if they can
• 41:10
  SPEAKER_00: offer the solution.
• 41:11
  SPEAKER_02: Interestingly enough,
• 41:12
  looping back to the beginning of our conversation when you
• 41:15
  talked about Chinese
• 41:18
  SPEAKER_02: efforts in the primate space,
• 41:20
  SPEAKER_02: That would avoid the problem of somebody defining the need.
• 41:26
  SPEAKER_02: Exactly.
• 41:26
  SPEAKER_02: And then cabining who can provide the solution to that need and say,
• 41:30
  SPEAKER_02: open it up to the world and say,
• 41:32
  SPEAKER_02: hey.
• 41:32
  SPEAKER_02: And some people may take an AI approach to something.
• 41:35
  SPEAKER_02: Others may take a hard technology approach to it.
• 41:39
  SPEAKER_02: And it would be sort of the Wright brothers versus the wingsuits.
• 41:42
  SPEAKER_02: Right.
• 41:43
  SPEAKER_02: That's exactly right.
• 41:43
  SPEAKER_02: And then somebody would win, but it wouldn't be organically determined.
• 41:46
  It'd be organically determined as opposed to being determined
• 41:49
  by the person who
• 41:50
  SPEAKER_02: has,
• 41:51
  SPEAKER_02: in a sense,
• 41:51
  SPEAKER_02: skin in the game and is putting their thumb on the scale.
• 41:53
  SPEAKER_00: That's exactly right.
• 41:54
  And it was just an infuriating,
• 41:55
  bizarre system that really was based off of central planning.
• 41:58
  SPEAKER_00: And we are thankfully, we're not fully moved away from it.
• 42:01
  There's still some areas where it happens a fair degree,
• 42:03
  but we're getting much better at it.
• 42:04
  SPEAKER_00: And again, we give huge credit.
• 42:06
  to the chairs of the Senate and House Armed Services Committee
• 42:09
  that have worked on this.
• 42:10
  And both the president and the secretary of war have worked
• 42:13
  tirelessly to push
• 42:14
  SPEAKER_00: these things through despite just incredible pushback within industry,
• 42:19
  SPEAKER_00: but even within kind of the middle layer of bureaucracy.
• 42:22
  SPEAKER_02: Great.
• 42:23
  SPEAKER_02: Well, in the last couple of minutes, let me see if I can summarize.
• 42:26
  SPEAKER_02: We've been through a lot of pretty interesting topics here.
• 42:29
  SPEAKER_02: Let me see if I can summarize at a high level where we've ended up,
• 42:31
  SPEAKER_02: and then I'd love for you to share any final thoughts,
• 42:33
  SPEAKER_02: or if I kind of got it wrong,
• 42:35
  SPEAKER_02: you can let me know.
• 42:36
  SPEAKER_02: But number one,
• 42:37
  I think that it's clear that there are very real stakes here as
• 42:43
  to whether this is
• 42:45
  SPEAKER_02: Chinese leadership—by Chinese,
• 42:47
  I mean the Chinese state-leading efforts here—or whether it is
• 42:51
  the United States
• 42:53
  SPEAKER_02: and its allies leading in this space.
• 42:56
  SPEAKER_02: Second, there are actually policy levers that we have.
• 43:00
  This isn't a sort of foreordained conflict between two
• 43:04
  geopolitical rivals,
• 43:05
  SPEAKER_02: and we're just going to watch it in sort of a Greek tragedy way play out.
• 43:09
  SPEAKER_02: But there are actually actual—
• 43:11
  policy levers that can be used that lean into the American way
• 43:18
  of doing things
• 43:19
  SPEAKER_02: through innovation,
• 43:20
  through leveraging Silicon Valley decentralization and sort of
• 43:25
  commercial leading
• 43:27
  SPEAKER_02: the way as opposed to the government leading the way.
• 43:32
  SPEAKER_02: And third,
• 43:33
  that it's really critical for the government itself and for
• 43:37
  government provision of
• 43:39
  SPEAKER_02: things like defense.
• 43:40
  SPEAKER_02: Candidly, we talked about education a little bit as well.
• 43:43
  SPEAKER_02: But areas where traditionally the government has led,
• 43:47
  there are real opportunities for private sector and for AI to
• 43:52
  really influence
• 43:53
  SPEAKER_02: education.
• 43:54
  the way American society and American values and priorities
• 43:58
  develop as well.
• 43:59
  I know it's a very high level,
• 44:00
  but is that sort of a good summary of where we've ended up?
• 44:03
  SPEAKER_00: No, that's dead on.
• 44:04
  SPEAKER_00: The only thing I would add is that those listening,
• 44:07
  SPEAKER_00: odds are you're blessed to be in a democracy.
• 44:10
  And if these things matter to you,
• 44:12
  if that's the future that you want,
• 44:14
  then get engaged.
• 44:15
  SPEAKER_00: Both at the political level,
• 44:16
  if you want to start a startup that you think could address
• 44:18
  some of these issues,
• 44:19
  SPEAKER_00: go for it.
• 44:19
  SPEAKER_00: You have the tools available to you and the freedom available to you that
• 44:24
  SPEAKER_00: almost no human in history has had.
• 44:26
  And so don't feel like you have to sit back and be just someone
• 44:31
  who's just watching.
• 44:32
  SPEAKER_00: You can actually directly engage and make a difference.
• 44:35
  SPEAKER_00: That's great.
• 44:36
  And I love ending on that note because I think we hear a lot
• 44:38
  about what we should
• 44:39
  SPEAKER_02: be worried about with AI.
• 44:43
  SPEAKER_02: But the reality is that the potential
• 44:45
  SPEAKER_02: for things like human flourishing,
• 44:48
  SPEAKER_02: for kind of releasing human spirits is enormous in this space.
• 44:53
  SPEAKER_00: Absolutely.
• 44:54
  SPEAKER_02: I want to thank you so much.
• 44:55
  SPEAKER_02: This has been a great conversation.
• 44:56
  SPEAKER_02: Super fun.
• 44:56
  SPEAKER_02: I think you were joking a moment ago.
• 44:58
  SPEAKER_02: It's kind of like all of our conversation.
• 44:59
  SPEAKER_00: Yeah, it is.
• 44:59
  SPEAKER_00: Very natural.
• 45:00
  SPEAKER_00: Absolutely.
• 45:01
  SPEAKER_02: It's always a pleasure to talk to you.
• 45:02
  SPEAKER_02: Right back at you.