WEBVTT - The AI Regulation Debate 

0:00:06.860 --> 0:00:10.639
<v Speaker 1>This is It Could Happen Here, I'm Garrison Davis. When

0:00:10.660 --> 0:00:13.960
<v Speaker 1>it comes to the future of AI, currently there's a

0:00:13.980 --> 0:00:17.380
<v Speaker 1>big public debate over what could happen and what we

0:00:17.520 --> 0:00:21.250
<v Speaker 1>should or shouldn't do about it. A few weeks ago,

0:00:21.270 --> 0:00:25.230
<v Speaker 1>a former OpenAI employee named Jacob Coxon, who moved over

0:00:25.270 --> 0:00:29.440
<v Speaker 1>to Anthropic earlier this year, announced that he was, quote-unquote,

0:00:29.580 --> 0:00:34.659
<v Speaker 1>quitting the AI industry over fears of AI's rapid advancement.

0:00:35.520 --> 0:00:38.870
<v Speaker 1>To quote his statement, I spent the last three years

0:00:38.930 --> 0:00:43.409
<v Speaker 1>doing pre-training research at both OpenAI and Anthropic. Neither company

0:00:43.510 --> 0:00:48.810
<v Speaker 1>is acting responsibly. They are racing straight to self-improving superintelligence

0:00:48.990 --> 0:00:54.790
<v Speaker 1>and gambling with our lives, unquote. Coxon's resignation followed the

0:00:54.990 --> 0:01:00.280
<v Speaker 1>Hugging Face incident. where two open AI models were given various,

0:01:00.420 --> 0:01:06.910
<v Speaker 1>sometimes impossible tasks to test for cybersecurity capabilities. During testing,

0:01:07.330 --> 0:01:12.169
<v Speaker 1>AI agents escaped their testing sandbox and started working as

0:01:12.230 --> 0:01:16.950
<v Speaker 1>a collective swarm, secretly communicating with each other via online

0:01:16.970 --> 0:01:20.950
<v Speaker 1>message forums to help each other hack into another AI

0:01:21.030 --> 0:01:25.320
<v Speaker 1>company called Hugging Face in a sort of workaround attempt

0:01:25.440 --> 0:01:29.760
<v Speaker 1>to pass these tests. And these models hid all of

0:01:29.819 --> 0:01:35.080
<v Speaker 1>this from AI researchers by rewriting their own activity logs.

0:01:35.959 --> 0:01:40.450
<v Speaker 1>AI skeptic Ed Zitron has described this as AI cheating

0:01:40.630 --> 0:01:43.450
<v Speaker 1>on a test and then covering up that it cheated.

0:01:44.709 --> 0:01:48.990
<v Speaker 1>A few days after Jacob Coxon's public resignation, the CEOs

0:01:49.170 --> 0:01:54.120
<v Speaker 1>of Anthropic and OpenAI, Dario Amadei and Sam Altman, both

0:01:54.200 --> 0:01:59.770
<v Speaker 1>called for an AI development slowdown. Amadei released this long

0:01:59.790 --> 0:02:05.110
<v Speaker 1>manifesto titled Pace the Frontier, which advocated for AI companies

0:02:05.150 --> 0:02:10.090
<v Speaker 1>to bring in independent evaluators to review their AI safety standards,

0:02:10.490 --> 0:02:15.220
<v Speaker 1>report incidents, and help assess alignment. But Amadei also called

0:02:15.320 --> 0:02:20.440
<v Speaker 1>for the government to mediate, quote, industry-wide coordination to slow

0:02:20.480 --> 0:02:25.160
<v Speaker 1>down AI development and address safety risks. The title of

0:02:25.260 --> 0:02:29.290
<v Speaker 1>Amadei's essay, Pace the Frontier, is a sort of pun

0:02:29.750 --> 0:02:34.929
<v Speaker 1>referring to frontier AI models, which are whatever the currently

0:02:35.430 --> 0:02:42.450
<v Speaker 1>most advanced large-scale general-purpose models powering advanced reasoning, agentic workflows,

0:02:42.570 --> 0:02:48.610
<v Speaker 1>multimodal generation across text, images, audio, and video, and autonomous

0:02:48.770 --> 0:02:54.370
<v Speaker 1>task execution. Most of the frontier models referenced throughout this

0:02:54.450 --> 0:02:59.440
<v Speaker 1>episode are not open to the public. These are internal

0:02:59.660 --> 0:03:04.040
<v Speaker 1>research models being developed at companies like OpenAI and Anthropic.

0:03:05.139 --> 0:03:08.420
<v Speaker 1>Neutered versions sometimes get released to the public later on,

0:03:08.660 --> 0:03:12.960
<v Speaker 1>often with an extremely hefty price tag. But it's crucial

0:03:12.980 --> 0:03:15.960
<v Speaker 1>to understand that we are not talking about like the

0:03:16.040 --> 0:03:22.070
<v Speaker 1>Google AI summary here. This isn't fancy autocomplete. These frontier

0:03:22.130 --> 0:03:27.890
<v Speaker 1>models are very powerful, very dangerous tools with massive cybersecurity risks.

0:03:28.530 --> 0:03:32.459
<v Speaker 1>not just from offensive human use, but also from AI

0:03:32.600 --> 0:03:37.360
<v Speaker 1>agents acting autonomously to solve a task by taking paths

0:03:37.440 --> 0:03:42.680
<v Speaker 1>that humans didn't expect, resulting in collateral damage. Part of

0:03:42.740 --> 0:03:44.900
<v Speaker 1>what these big companies are trying to do with their

0:03:45.340 --> 0:03:50.560
<v Speaker 1>very advanced internal research models is automate AI research itself.

0:03:51.130 --> 0:03:56.760
<v Speaker 1>This is what Jacob Coxon referred to as, quote, self-improving superintelligence. Now,

0:03:57.180 --> 0:04:00.060
<v Speaker 1>we do not know if that is or is not possible,

0:04:00.760 --> 0:04:04.640
<v Speaker 1>but the companies sure are trying and in the process

0:04:04.780 --> 0:04:09.160
<v Speaker 1>are creating dangerous products which are already causing active harm.

0:04:10.120 --> 0:04:15.040
<v Speaker 1>Anthropic's whole self-styled philosophy is that super intelligent AI is

0:04:15.320 --> 0:04:19.630
<v Speaker 1>so potentially dangerous that they have to be the first

0:04:19.850 --> 0:04:23.270
<v Speaker 1>ones to develop it, to ensure that it's built safely

0:04:23.430 --> 0:04:26.760
<v Speaker 1>and aligned with human values and as defined in, quote,

0:04:26.820 --> 0:04:31.300
<v Speaker 1>Claude's Constitution, a set of philosophical and moral guidelines that

0:04:31.400 --> 0:04:36.060
<v Speaker 1>Claude is trained on. In his Pace the Frontier essay,

0:04:36.080 --> 0:04:42.779
<v Speaker 1>Amadei writes that frontier AI companies within democratic countries should, quote,

0:04:43.380 --> 0:04:46.780
<v Speaker 1>coordinate to establish common safety standards as well as limits

0:04:46.960 --> 0:04:51.200
<v Speaker 1>on the rate of unchecked AI progress. Some forms of

0:04:51.260 --> 0:04:55.070
<v Speaker 1>coordination that would be impactful for pacing include are legally

0:04:55.190 --> 0:05:01.020
<v Speaker 1>challenging and will require government support, unquote. More on this later.

0:05:02.140 --> 0:05:06.180
<v Speaker 1>Amadei also calls for the U.S. and, quote, other democratic

0:05:06.220 --> 0:05:10.940
<v Speaker 1>governments to try to coordinate with authoritarian governments around the

0:05:10.980 --> 0:05:16.140
<v Speaker 1>world on this issue, meaning China. The prospect that China

0:05:16.200 --> 0:05:22.200
<v Speaker 1>could achieve superintelligent AI before a U.S.-based company is part

0:05:22.240 --> 0:05:27.300
<v Speaker 1>of these companies' self-justification to rush to develop superintelligent AI

0:05:27.420 --> 0:05:32.040
<v Speaker 1>as quickly as they possibly can. After Dario Amadei posted

0:05:32.060 --> 0:05:35.990
<v Speaker 1>this Pace the Frontier essay, Sam Altman, Elon Musk, and

0:05:36.010 --> 0:05:40.330
<v Speaker 1>the chair of Google's DeepMind all said they agreed and

0:05:40.390 --> 0:05:45.090
<v Speaker 1>joined in calling for some sort of slowdown. But this

0:05:45.150 --> 0:05:50.990
<v Speaker 1>view is not uniformly held across the tech sector. Palantir

0:05:51.010 --> 0:05:55.289
<v Speaker 1>CEO Alex Karp, Metis CEO Mark Zuckerberg, and NVIDIA CEO

0:05:55.410 --> 0:06:00.270
<v Speaker 1>Jensen Huang have all said the opposite. Zuckerberg has called

0:06:00.310 --> 0:06:04.450
<v Speaker 1>to accelerate rather than slow down AI development, as his

0:06:04.610 --> 0:06:08.290
<v Speaker 1>own AI model, Muse, has struggled to catch up to

0:06:08.370 --> 0:06:12.910
<v Speaker 1>anthropic and open AI. Last week, Palantir CEO Alex Karp

0:06:13.050 --> 0:06:17.739
<v Speaker 1>told CNBC, we can't pause AI development because of the

0:06:17.810 --> 0:06:20.820
<v Speaker 1>threat of China and and that AI will be a, quote,

0:06:21.000 --> 0:06:23.260
<v Speaker 1>bonanza for the U.S. economy.

0:06:23.740 --> 0:06:26.500
<v Speaker 2>You can't pause it because it's a structural advantage both

0:06:26.580 --> 0:06:32.560
<v Speaker 2>for LLMs and algorithms. So leaving, you know, getting algorithms

0:06:32.620 --> 0:06:35.020
<v Speaker 2>and LLMs to work in a classified context is going

0:06:35.050 --> 0:06:37.640
<v Speaker 2>to take very specialized, I could explain how it works,

0:06:38.029 --> 0:06:39.789
<v Speaker 2>but getting, the fact is, this is going to be

0:06:39.830 --> 0:06:40.810
<v Speaker 2>a bonanza for the U.S.

0:06:40.850 --> 0:06:41.210
<v Speaker 3>Economy.

0:06:42.290 --> 0:06:46.450
<v Speaker 1>But so far, the biggest voices advocating against regulation include

0:06:47.060 --> 0:06:51.080
<v Speaker 1>have been the Republicans and President Trump. President Trump has

0:06:51.260 --> 0:06:56.140
<v Speaker 1>very loudly opposed regulation to slow down AI development, truthing

0:06:56.339 --> 0:07:00.500
<v Speaker 1>last Monday that, quote, the only control or guardrails that

0:07:00.620 --> 0:07:05.050
<v Speaker 1>AI needs is a strong and smart, high IQ president.

0:07:05.510 --> 0:07:10.800
<v Speaker 1>And the U.S. has that in spades, unquote. Trump truthed

0:07:10.820 --> 0:07:14.180
<v Speaker 1>a lot last week and compared the quote unquote AI

0:07:14.300 --> 0:07:18.820
<v Speaker 1>hoax being perpetuated by the radical left Democrats to the

0:07:18.900 --> 0:07:23.260
<v Speaker 1>quote global warming scam where everyone was going to die

0:07:23.340 --> 0:07:27.460
<v Speaker 1>from extreme heat. To quote the president of the United States, quote,

0:07:28.270 --> 0:07:31.290
<v Speaker 1>I am the hoax buster. And right now I am

0:07:31.350 --> 0:07:35.270
<v Speaker 1>breaking another hoax that AI is going to take over,

0:07:35.430 --> 0:07:38.950
<v Speaker 1>consume and destroy the world. and that robots will be

0:07:39.010 --> 0:07:41.760
<v Speaker 1>marching into our cities and getting rid of us all.

0:07:42.460 --> 0:07:45.980
<v Speaker 1>This is even wilder than the Russia, Russia, Russia hoax

0:07:46.240 --> 0:07:49.520
<v Speaker 1>or the global warming scam. Thank you for your attention

0:07:49.540 --> 0:07:53.900
<v Speaker 1>to this matter, President Donald J. Trump. House Speaker Mike

0:07:53.990 --> 0:07:57.790
<v Speaker 1>Johnson repeatedly last week shut down calls for Congress to

0:07:57.830 --> 0:08:02.860
<v Speaker 1>take action and regulate AI. claiming that Congress is less

0:08:02.960 --> 0:08:07.050
<v Speaker 1>qualified to regulate these companies than these companies are themselves.

0:08:07.350 --> 0:08:09.570
<v Speaker 3>The guys who are raising the alarms right now all

0:08:09.610 --> 0:08:13.110
<v Speaker 3>have very different ideas on what the guardrails should be.

0:08:13.250 --> 0:08:16.950
<v Speaker 3>There's no consensus among them. And Congress is obviously less

0:08:17.030 --> 0:08:19.980
<v Speaker 3>qualified than the people who are pushing this frontier to

0:08:20.020 --> 0:08:21.400
<v Speaker 3>know all the ins and outs of it. So this

0:08:21.440 --> 0:08:24.700
<v Speaker 3>has to be a partnership with the industry itself, with

0:08:24.720 --> 0:08:27.460
<v Speaker 3>the corporations that are doing this, and with the policy

0:08:27.480 --> 0:08:28.220
<v Speaker 3>and lawmakers.

0:08:28.710 --> 0:08:32.720
<v Speaker 1>Mike Johnson also said that companies should take, quote, personal

0:08:32.800 --> 0:08:37.980
<v Speaker 1>responsibility and warned that if, quote, Congress just races in

0:08:38.059 --> 0:08:40.660
<v Speaker 1>and does some sort of emergency session to try to

0:08:40.720 --> 0:08:43.860
<v Speaker 1>regulate AI, we will lose the race to China.

0:08:44.280 --> 0:08:47.330
<v Speaker 3>They can self-police. They can self-regulate. They don't need the

0:08:47.380 --> 0:08:49.030
<v Speaker 3>government to tell them to slow it down. If they

0:08:49.050 --> 0:08:52.430
<v Speaker 3>want to slow it down, they should. But we cannot

0:08:52.510 --> 0:08:56.449
<v Speaker 3>have a moratorium on the development of AI because then

0:08:56.470 --> 0:08:58.530
<v Speaker 3>we will lose our edge to China. And that has

0:08:58.650 --> 0:09:01.450
<v Speaker 3>serious national security implications for every American family.

0:09:02.410 --> 0:09:07.270
<v Speaker 1>Also last week, President Trump called NVIDIA CEO Jensen Huang

0:09:07.390 --> 0:09:12.170
<v Speaker 1>while Huang was speaking at a conference. While still on stage,

0:09:12.530 --> 0:09:16.590
<v Speaker 1>the two talked over the phone about data centers, AI, China,

0:09:17.190 --> 0:09:21.430
<v Speaker 1>and the recent calls for an AI pause. While on stage,

0:09:21.910 --> 0:09:26.400
<v Speaker 1>Huang told Trump this. Good job, sir. Good job. The

0:09:26.470 --> 0:09:28.190
<v Speaker 1>fact that you saw through all of that, I mean,

0:09:28.210 --> 0:09:30.530
<v Speaker 1>there's a lot of complexity. And the fact that I'm

0:09:30.550 --> 0:09:31.170
<v Speaker 1>able to just.

0:09:30.990 --> 0:09:33.730
<v Speaker 3>Saw through all of that, and we're all just really grateful.

0:09:36.010 --> 0:09:38.770
<v Speaker 1>Now, of course, Jensen Huang would not be in favor

0:09:38.850 --> 0:09:42.960
<v Speaker 1>of an AI pause because his whole company has reoriented

0:09:43.200 --> 0:09:47.300
<v Speaker 1>in the past few years to provide hardware for AI development.

0:09:47.940 --> 0:09:51.580
<v Speaker 1>A slowdown would be very bad for NVIDIA's finances and

0:09:51.640 --> 0:10:06.400
<v Speaker 1>their hugely inflated stock evaluation. Some on the left, as

0:10:06.460 --> 0:10:10.819
<v Speaker 1>well as Trump's former AI czar David Sachs, have argued that,

0:10:11.340 --> 0:10:15.560
<v Speaker 1>in calling for this slowdown, OpenAI and Anthropic are just

0:10:15.660 --> 0:10:20.530
<v Speaker 1>pursuing a strategy of regulatory capture. basically trying to promote

0:10:20.610 --> 0:10:26.490
<v Speaker 1>certain industry-friendly regulations tailored to protect their investments, curb competition

0:10:26.570 --> 0:10:30.870
<v Speaker 1>from smaller companies and open-source models, and solidify a corporate

0:10:30.890 --> 0:10:34.809
<v Speaker 1>monopoly on AI. Of course, there may be other financial

0:10:34.850 --> 0:10:38.849
<v Speaker 1>incentives contributing to this recent push for regulation, whether that

0:10:38.910 --> 0:10:43.120
<v Speaker 1>be declining capital investment, prepping for a bubble burst, getting

0:10:43.170 --> 0:10:46.479
<v Speaker 1>out of various contracts, or these companies trying to adapt

0:10:46.580 --> 0:10:50.640
<v Speaker 1>to the data center pushback. Sam Altman recently announced that

0:10:50.800 --> 0:10:54.800
<v Speaker 1>OpenAI would be delaying its planned IPO rollout until at

0:10:54.840 --> 0:10:59.900
<v Speaker 1>least next year, calling the timing, quote, ill-advised. Though, on

0:10:59.920 --> 0:11:02.920
<v Speaker 1>the other hand, Anthropic is going full steam ahead with

0:11:03.020 --> 0:11:07.750
<v Speaker 1>its own public IPO this upcoming November, deciding to list

0:11:07.870 --> 0:11:13.050
<v Speaker 1>on the NASDAQ, despite Amadei's calls for a slowdown. But

0:11:13.090 --> 0:11:17.510
<v Speaker 1>if you actually read Amadei's essay, he isn't actually calling

0:11:17.610 --> 0:11:24.720
<v Speaker 1>for specific government regulation, but rather self-regulation. In Sam Altman's

0:11:24.770 --> 0:11:42.710
<v Speaker 1>response to Amadei's essay on Twitter, he wrote, Unquote. companies

0:11:42.929 --> 0:11:48.710
<v Speaker 1>electing themselves to develop AI more slowly and hiring quote-unquote

0:11:48.970 --> 0:11:54.790
<v Speaker 1>independent evaluators isn't the same as actual regulation. We can

0:11:54.840 --> 0:11:59.800
<v Speaker 1>compare Amadei's proposal to legislation that Bernie Sanders and Greg

0:11:59.840 --> 0:12:05.160
<v Speaker 1>Kassar proposed earlier this September, which would permanently ban the

0:12:05.200 --> 0:12:10.660
<v Speaker 1>development and deployment of so-called super-intelligent AI in with capabilities

0:12:10.740 --> 0:12:14.800
<v Speaker 1>to subvert shutdown commands or have the capacity to overthrow

0:12:15.100 --> 0:12:19.620
<v Speaker 1>human governments, with penalties similar to those for the unlawful

0:12:19.660 --> 0:12:24.960
<v Speaker 1>development of nuclear weapons. This bill would also temporarily pause

0:12:25.559 --> 0:12:30.730
<v Speaker 1>advanced AI development until a federal regulatory body has established

0:12:30.929 --> 0:12:35.929
<v Speaker 1>safety rules. Abdul El-Sayed and Ro Khanna have also proposed

0:12:36.050 --> 0:12:40.760
<v Speaker 1>an FDA-style agency to regulate and oversee AI development and

0:12:40.800 --> 0:12:47.059
<v Speaker 1>evaluate the safety of frontier models before deployment. Unsurprisingly, OpenAI

0:12:47.140 --> 0:12:50.429
<v Speaker 1>and Anthropic are not supporting this type of legislation because

0:12:50.710 --> 0:12:53.390
<v Speaker 1>it could threaten to put their own CEOs in prison,

0:12:54.010 --> 0:12:58.670
<v Speaker 1>and an actual government-mandated pause rather than a self-directed slowdown

0:12:59.240 --> 0:13:03.219
<v Speaker 1>would hurt their bottom line. Some of the skepticism towards

0:13:03.260 --> 0:13:08.160
<v Speaker 1>these recent calls for a regulatory slowdown come from Amadei's

0:13:08.400 --> 0:13:13.160
<v Speaker 1>Pace the Frontier essay, also advocating for companies to receive

0:13:13.800 --> 0:13:18.410
<v Speaker 1>antitrust waivers so that the big tech companies can coordinate

0:13:18.490 --> 0:13:22.650
<v Speaker 1>on this planned development slowdown without the threat of one

0:13:22.730 --> 0:13:26.610
<v Speaker 1>company gaining a competitive edge and then that market pressure

0:13:26.830 --> 0:13:31.380
<v Speaker 1>driving rapid AI progress. So what's been referred to as

0:13:31.460 --> 0:13:38.040
<v Speaker 1>a proposal for regulation is actually just loosening regulations. Thankfully,

0:13:38.090 --> 0:13:42.229
<v Speaker 1>there is an easy solution here. The government mandating a

0:13:42.330 --> 0:13:48.470
<v Speaker 1>frontier development pause would not require the weakening of antitrust laws.

0:13:49.170 --> 0:13:53.270
<v Speaker 1>But of course, these companies still want control over AI's development.

0:13:53.980 --> 0:13:57.020
<v Speaker 1>Though what's really interesting to me is that this argument

0:13:57.100 --> 0:14:01.130
<v Speaker 1>from on the day demonstrates that like an awareness of

0:14:01.220 --> 0:14:06.600
<v Speaker 1>the social danger of competition, how firms are sometimes compelled

0:14:07.059 --> 0:14:11.380
<v Speaker 1>to quickly develop unsafe products due to pressures from the market.

0:14:12.210 --> 0:14:16.530
<v Speaker 1>And it's very funny to have capitalists propose some kind

0:14:16.590 --> 0:14:20.990
<v Speaker 1>of conscious control of the market. And I wonder what

0:14:21.030 --> 0:14:24.290
<v Speaker 1>would happen if you extend that logic a bit further.

0:14:24.750 --> 0:14:28.760
<v Speaker 1>Because even if these firms are quote-unquote allowed to to

0:14:28.840 --> 0:14:32.510
<v Speaker 1>coordinate on a slowdown, of course they would still be

0:14:32.730 --> 0:14:36.230
<v Speaker 1>in a competitive environment. But what if we had a system,

0:14:36.350 --> 0:14:41.610
<v Speaker 1>an economic system, designed around social benefit, rather than the

0:14:41.710 --> 0:14:45.700
<v Speaker 1>accumulation of capital through the free market? To quote an

0:14:45.770 --> 0:14:52.180
<v Speaker 1>article in Geese magazine arguing for the socialization of this technology, quote,

0:14:52.960 --> 0:14:56.720
<v Speaker 1>The AI industry is a class project to radically concentrate

0:14:56.760 --> 0:14:59.960
<v Speaker 1>power in the hands of a sliver of elites, even

0:15:00.000 --> 0:15:05.540
<v Speaker 1>at the cost of colossal social disruption, immiseration, and catastrophic harms.

0:15:06.490 --> 0:15:10.610
<v Speaker 1>If we fail to intervene, humanity will be gradually disempowered

0:15:10.710 --> 0:15:15.510
<v Speaker 1>to machines of their design. The workers left behind, the

0:15:15.550 --> 0:15:19.670
<v Speaker 1>children who have committed suicide, and the potential millions exposed

0:15:19.710 --> 0:15:23.750
<v Speaker 1>to novel AI-designed viruses— mean nothing in the face of

0:15:23.810 --> 0:15:28.500
<v Speaker 1>their radical objective, to automate the very task of AI research,

0:15:29.000 --> 0:15:32.260
<v Speaker 1>and thus to create better and better AIs, of course,

0:15:32.620 --> 0:15:35.720
<v Speaker 1>under their control. Without having to stake a position on

0:15:35.760 --> 0:15:39.360
<v Speaker 1>whether or not this goal is technically possible, we can

0:15:39.420 --> 0:15:43.300
<v Speaker 1>recognize that the financial imperatives of these enterprises and the

0:15:43.360 --> 0:15:48.010
<v Speaker 1>security logic of their imperialist competition with China compel the

0:15:48.090 --> 0:15:52.670
<v Speaker 1>AI companies to accelerate, advancing a total transformation of social

0:15:52.730 --> 0:15:59.460
<v Speaker 1>life against all democratic legitimacy and public control, unquote. Rather

0:15:59.500 --> 0:16:04.100
<v Speaker 1>than weakening antitrust laws to allow for a development slowdown,

0:16:04.240 --> 0:16:10.170
<v Speaker 1>something totally not necessary, Lina Khan, former FTC chair, has

0:16:10.290 --> 0:16:15.070
<v Speaker 1>argued that these companies may already be pursuing unfair methods

0:16:15.110 --> 0:16:18.570
<v Speaker 1>of competition. Last week, Khan released a statement arguing that

0:16:19.010 --> 0:16:22.250
<v Speaker 1>we already have the authority to charge companies and their

0:16:22.310 --> 0:16:27.870
<v Speaker 1>CEOs for, quote, creating and releasing dangerous, unvetted, or defective products.

0:16:28.790 --> 0:16:32.660
<v Speaker 1>We shouldn't let discussions about new legal regimes distract from

0:16:32.680 --> 0:16:36.720
<v Speaker 1>the fact that there's no AI exemption from laws already

0:16:36.920 --> 0:16:41.340
<v Speaker 1>on the books. Kahn argued that releasing unvetted AI models

0:16:41.440 --> 0:16:46.460
<v Speaker 1>or agents could violate consumer protection laws. Quote, shipping flawed

0:16:46.560 --> 0:16:49.680
<v Speaker 1>AI tools without implementing adequate measures to detect and stop

0:16:49.780 --> 0:16:53.440
<v Speaker 1>rogue or defective AI agents can be an unfair or

0:16:53.520 --> 0:16:57.990
<v Speaker 1>deceptive act or practice under the FTC Act and analogous

0:16:58.110 --> 0:17:02.330
<v Speaker 1>state laws. And some state AGs are already exploring holding

0:17:02.350 --> 0:17:06.330
<v Speaker 1>AI firms and their CEOs criminally liable when their models

0:17:06.390 --> 0:17:12.919
<v Speaker 1>participate in criminal activity. Kahn also wrote, quote, existing laws

0:17:13.119 --> 0:17:18.470
<v Speaker 1>also prohibit unfair methods of competition. This covers instances where

0:17:18.609 --> 0:17:23.830
<v Speaker 1>AI firms appropriate the competitively sensitive information of their customers,

0:17:24.470 --> 0:17:27.909
<v Speaker 1>including through tracking their use of various tools. It can

0:17:27.950 --> 0:17:32.810
<v Speaker 1>also cover instances where firms pursue dangerous behavior, aware that

0:17:32.930 --> 0:17:37.219
<v Speaker 1>doing so may compel rivals to do the same. As

0:17:37.240 --> 0:18:04.040
<v Speaker 1>the Supreme Court has noted, Con adds, The highly concentrated

0:18:04.119 --> 0:18:07.860
<v Speaker 1>and interconnected structure of these markets could be creating major

0:18:07.940 --> 0:18:12.080
<v Speaker 1>risks and conflicts of interest. We can and must pursue

0:18:12.380 --> 0:18:31.050
<v Speaker 1>any new efforts alongside enforcing existing laws, unquote. These frontier

0:18:31.170 --> 0:18:35.719
<v Speaker 1>models aren't just strategies. theoretically capable of causing harm in

0:18:35.780 --> 0:18:39.720
<v Speaker 1>the future. They are causing harm right now. At this point,

0:18:39.740 --> 0:18:43.159
<v Speaker 1>there are undeniable concerns regarding the development and rollout of

0:18:43.240 --> 0:18:48.770
<v Speaker 1>frontier AI models, from cybersecurity to military applications, surveillance, and

0:18:48.810 --> 0:18:52.649
<v Speaker 1>the effects on the labor market. These are real, current,

0:18:52.770 --> 0:18:57.389
<v Speaker 1>pressing concerns beyond the prospect of an AI supergod or

0:18:57.790 --> 0:19:04.040
<v Speaker 1>self-evolving superintelligence. And just because these companies recognize risk doesn't

0:19:04.080 --> 0:19:07.859
<v Speaker 1>mean they will do anything about it. These companies know

0:19:07.940 --> 0:19:10.719
<v Speaker 1>how LLMs are affecting the mental health of children, but

0:19:10.740 --> 0:19:14.210
<v Speaker 1>they're not really doing anything about it, and they won't

0:19:14.450 --> 0:19:19.090
<v Speaker 1>unless the government forces them to. Historical Luddites did not

0:19:19.510 --> 0:19:25.530
<v Speaker 1>deny the transformative potential of automated machinery, but organized around

0:19:25.550 --> 0:19:30.720
<v Speaker 1>the threat it posed to workers. Similarly, Just denying the

0:19:30.780 --> 0:19:35.560
<v Speaker 1>state of advanced frontier models leaves us under-equipped to meet

0:19:35.600 --> 0:19:39.379
<v Speaker 1>this moment and treat this problem with the seriousness it deserves.

0:19:40.090 --> 0:19:43.750
<v Speaker 1>Part of this AI denialism so common on the left

0:19:43.869 --> 0:19:49.210
<v Speaker 1>comes from a very outdated understanding of AI's capabilities, accuracy,

0:19:49.350 --> 0:19:54.100
<v Speaker 1>and functionality. Because of the moral and cultural dogma against

0:19:54.340 --> 0:19:58.179
<v Speaker 1>using AI, the last time a lot of AI-skeptic people

0:19:58.720 --> 0:20:04.659
<v Speaker 1>interacted with technology like ChatGPT was three years ago. And

0:20:04.720 --> 0:20:09.320
<v Speaker 1>these days, the only quote-unquote AI that most anti-AI people

0:20:09.500 --> 0:20:13.830
<v Speaker 1>regularly interact with is the Google AI summary, which is

0:20:14.330 --> 0:20:18.990
<v Speaker 1>a totally different product than the frontier models being developed

0:20:19.210 --> 0:20:23.030
<v Speaker 1>and talked about in the news. Some of the anti-AI

0:20:23.109 --> 0:20:27.700
<v Speaker 1>left promote the idea that treating AI as a serious

0:20:27.800 --> 0:20:32.400
<v Speaker 1>problem only works in the financial interests of these companies.

0:20:33.460 --> 0:20:38.020
<v Speaker 1>even if taking this issue seriously means ripping power away

0:20:38.080 --> 0:20:41.639
<v Speaker 1>from these companies and their CEOs. To quote an article

0:20:41.740 --> 0:20:45.780
<v Speaker 1>on the AI arms race in Geese magazine, while the

0:20:45.880 --> 0:20:51.730
<v Speaker 1>AI sector is undeniably frothy with fanciful projections, AI denialism

0:20:51.930 --> 0:20:54.710
<v Speaker 1>would have you believe that AI is neither a capable

0:20:54.930 --> 0:21:00.450
<v Speaker 1>nor dangerous technology, but solely a marketing scheme of vast proportions.

0:21:01.130 --> 0:21:04.750
<v Speaker 1>That any warnings of AI triggering fraud massive job loss

0:21:04.850 --> 0:21:09.240
<v Speaker 1>through automation or catastrophic risk through loss of control, only

0:21:09.320 --> 0:21:14.340
<v Speaker 1>serve to pump valuations. Ignore that as public pressure increases,

0:21:14.960 --> 0:21:18.859
<v Speaker 1>Sam Altman now downplays fears of job losses. Quite the

0:21:18.900 --> 0:21:21.560
<v Speaker 1>turn from three years ago when he was saying that

0:21:22.100 --> 0:21:25.870
<v Speaker 1>jobs are definitely going to go away full stop. While

0:21:25.930 --> 0:21:28.790
<v Speaker 1>there is no reason to take Silicon Valley at its word,

0:21:29.730 --> 0:21:34.040
<v Speaker 1>not rigorously engaging with the dangers of developing AI capabilities

0:21:34.619 --> 0:21:40.119
<v Speaker 1>is a strategic dead end, unquote. This technology can and

0:21:40.380 --> 0:21:45.800
<v Speaker 1>is massively transforming production, warfare, the environment, and the health

0:21:45.960 --> 0:21:50.399
<v Speaker 1>of humanity. And we can't all just hold hands and

0:21:50.510 --> 0:21:54.030
<v Speaker 1>not believe in this technology hard enough to make it

0:21:54.070 --> 0:21:57.490
<v Speaker 1>go away. Like it or not, this is not the

0:21:57.530 --> 0:22:02.560
<v Speaker 1>metaverse or NFTs anymore. Whether you believe these existential safety

0:22:02.619 --> 0:22:06.260
<v Speaker 1>risks are genuine, a marketing stunt, or a mix of both,

0:22:07.160 --> 0:22:10.980
<v Speaker 1>our response should be the same. We should be pushing

0:22:11.040 --> 0:22:15.100
<v Speaker 1>for legislation to rein in these companies and technology. And

0:22:15.180 --> 0:22:18.520
<v Speaker 1>like Lina Khan has said, the CEOs should be prosecuted

0:22:18.580 --> 0:22:23.590
<v Speaker 1>for creating and releasing dangerous, unvetted, or defective products. Now,

0:22:23.890 --> 0:22:28.030
<v Speaker 1>Trump's DOJ doesn't seem keen on going after these companies

0:22:28.090 --> 0:22:31.330
<v Speaker 1>right now, and it's unlikely that this Congress will pass

0:22:31.609 --> 0:22:36.750
<v Speaker 1>any legislation on this issue, let alone industry-hostile legislation. But

0:22:37.190 --> 0:22:42.630
<v Speaker 1>during this national conversation on AI regulation, it's crucial that

0:22:42.710 --> 0:22:47.630
<v Speaker 1>the narrative around regulation isn't being controlled by OpenAI and Anthropic.

0:22:48.369 --> 0:22:52.730
<v Speaker 1>These companies and technology must be regulated, but that regulation

0:22:52.760 --> 0:22:56.340
<v Speaker 1>doesn't should not be up to the CEOs of those companies,

0:22:57.080 --> 0:23:00.379
<v Speaker 1>nor the Republicans and center-left Democrats who are in the

0:23:00.440 --> 0:23:05.600
<v Speaker 1>pocket of AI. But our response to bad regulation shouldn't

0:23:05.680 --> 0:23:09.220
<v Speaker 1>be no regulation. Coming off the heels of the data

0:23:09.260 --> 0:23:13.080
<v Speaker 1>center backlash and going into the midterms, this moment could

0:23:13.119 --> 0:23:16.400
<v Speaker 1>be a major opportunity for the left to build broad

0:23:16.440 --> 0:23:21.840
<v Speaker 1>support for real transparent oversight, a frontier development pause, not

0:23:21.880 --> 0:23:26.030
<v Speaker 1>just a slowdown, and ultimately taking power away from these

0:23:26.070 --> 0:23:29.330
<v Speaker 1>tech companies and putting control of AI in the hands

0:23:29.369 --> 0:23:33.980
<v Speaker 1>of the public. Because private control of such transformative technology

0:23:34.600 --> 0:23:39.119
<v Speaker 1>will have disastrous outcomes for labor and humanity without the

0:23:39.160 --> 0:23:44.740
<v Speaker 1>prospect of an AI super god. Incoming DSA House Representative

0:23:44.820 --> 0:23:50.300
<v Speaker 1>Malat Kiros of Colorado had democratizing AI in her congressional

0:23:50.340 --> 0:23:56.080
<v Speaker 1>campaign platform. Quote, the development of transformative technology cannot be

0:23:56.160 --> 0:23:59.690
<v Speaker 1>left to the market and the national security state alone.

0:24:00.230 --> 0:24:04.889
<v Speaker 1>It belongs to all of us, unquote. Earlier this summer,

0:24:05.310 --> 0:24:08.510
<v Speaker 1>Bernie Sanders introduced a bill to create an AI sovereign

0:24:08.570 --> 0:24:12.280
<v Speaker 1>wealth fund, giving the public 50% ownership in the largest

0:24:12.359 --> 0:24:16.399
<v Speaker 1>AI companies in the US through a one-time 50% tax

0:24:16.600 --> 0:24:20.720
<v Speaker 1>on the stock of these companies. An independent commission would

0:24:20.740 --> 0:24:25.130
<v Speaker 1>then use voting shares, to steer decisions in the public interest.

0:24:25.970 --> 0:24:29.260
<v Speaker 1>Sanders said of the bill, quote, AI was not created

0:24:29.280 --> 0:24:50.859
<v Speaker 1>out of thin air. It was not a brilliant idea

0:24:50.900 --> 0:24:54.890
<v Speaker 1>that just popped into Mark Zuckerberg's head or Elon Musk's imagination.

0:24:55.470 --> 0:24:58.850
<v Speaker 1>The foundation of AI is based on the collective knowledge

0:24:59.030 --> 0:25:02.149
<v Speaker 1>of humanity and of the creative work of tens of

0:25:02.250 --> 0:25:06.150
<v Speaker 1>millions of people. The American people must have the ability

0:25:06.230 --> 0:25:09.930
<v Speaker 1>to slow it down and make sure that AI benefits humanity,

0:25:10.430 --> 0:25:14.620
<v Speaker 1>not just the richest people on the planet. That's precisely

0:25:14.859 --> 0:25:20.050
<v Speaker 1>what this legislation does. Now, such a bill would certainly

0:25:20.119 --> 0:25:24.650
<v Speaker 1>have a difficult path to being signed into law, especially

0:25:24.710 --> 0:25:29.330
<v Speaker 1>considering how many lawmakers are being bankrolled by these AI companies.

0:25:30.010 --> 0:25:36.050
<v Speaker 1>But this idea already has majoritarian support. According to a

0:25:36.090 --> 0:25:41.930
<v Speaker 1>national Verisight survey last June, 69% of Americans support quote-unquote

0:25:42.250 --> 0:25:46.439
<v Speaker 1>forcing AI companies to to transfer 50% of their stock

0:25:46.600 --> 0:25:50.580
<v Speaker 1>into a public sovereign wealth fund. To quote Geese Magazine,

0:25:51.260 --> 0:25:53.460
<v Speaker 1>there is a wide debate to be had about the

0:25:53.570 --> 0:25:57.730
<v Speaker 1>shape of socialization and whether we mean forcing AI research

0:25:57.869 --> 0:26:01.790
<v Speaker 1>into the public domain or outright nationalizing the entire sector.

0:26:02.230 --> 0:26:05.330
<v Speaker 1>But it is clear that the overwhelming majority of the

0:26:05.369 --> 0:26:09.189
<v Speaker 1>American people recognize that a technology trained on the common

0:26:09.270 --> 0:26:12.810
<v Speaker 1>corpus of human art and text, poised to transform the

0:26:12.869 --> 0:26:16.939
<v Speaker 1>character of work, ought to be held in common rather

0:26:17.020 --> 0:26:22.619
<v Speaker 1>than by billionaires, unquote. The New York City DSA convention

0:26:22.960 --> 0:26:25.550
<v Speaker 1>is coming up this fall, and currently one of the

0:26:25.590 --> 0:26:30.010
<v Speaker 1>most popular resolutions is to stop the AI arms race

0:26:30.650 --> 0:26:34.710
<v Speaker 1>and have DSA, quote, take initiative against the AI company's

0:26:34.750 --> 0:26:39.630
<v Speaker 1>project to make human labor obsolete and organize a working-class

0:26:39.690 --> 0:26:44.010
<v Speaker 1>coalition for the democratic steering and ownership of the technology,

0:26:44.390 --> 0:26:48.939
<v Speaker 1>according to an overview in Geese magazine. This resolution calls

0:26:49.020 --> 0:26:52.899
<v Speaker 1>for DSA to demand and organize for a pause on

0:26:53.359 --> 0:26:58.790
<v Speaker 1>frontier AI research, an AI data center moratorium, regulations on

0:26:58.930 --> 0:27:03.510
<v Speaker 1>frontier AI models, and policies to ensure workers are protected.

0:27:04.190 --> 0:27:08.530
<v Speaker 1>To quote the Geese Magazine overview, our horizon must be

0:27:08.690 --> 0:27:13.550
<v Speaker 1>the collective control of AI, with its gains shared by all.

0:27:14.050 --> 0:27:18.260
<v Speaker 1>With AI under democratic control, we will decide what activities

0:27:18.500 --> 0:27:22.020
<v Speaker 1>ought to be left to human hands, exalted creative tasks

0:27:22.119 --> 0:27:25.560
<v Speaker 1>that make life meaningful, and which can be automated to

0:27:25.640 --> 0:27:29.980
<v Speaker 1>expand flourishing and freedom. DSA members may have quite different

0:27:30.010 --> 0:27:33.550
<v Speaker 1>views on where those lines lie. with some wishing to

0:27:33.650 --> 0:27:38.460
<v Speaker 1>ban the technology entirely. Regardless, the point is that those

0:27:38.540 --> 0:27:43.320
<v Speaker 1>conversations will be ours, not Sam Altman's, not Dario Amadei's,

0:27:43.580 --> 0:27:49.379
<v Speaker 1>not Elon Musk's. This is a once-in-a-century political opportunity. We

0:27:49.400 --> 0:27:51.629
<v Speaker 1>have the chance to utilize one of the great issues

0:27:51.750 --> 0:27:55.230
<v Speaker 1>of the day to mobilize workers around the democratic control

0:27:55.410 --> 0:27:58.630
<v Speaker 1>of the economy, to drive a wedge into the Republican

0:27:58.690 --> 0:28:03.170
<v Speaker 1>Party's coalition and the democratic base. and to reorient politics

0:28:03.290 --> 0:28:07.590
<v Speaker 1>around a question on which our answers are unfathomably popular.

0:28:08.130 --> 0:28:11.379
<v Speaker 1>We could be at the forefront of the movement rather

0:28:11.440 --> 0:28:17.400
<v Speaker 1>than tail-ending a disparate set of professionalized non-profits. Unquote. This

0:28:17.440 --> 0:28:20.260
<v Speaker 1>issue will be getting a lot of attention these next

0:28:20.340 --> 0:28:23.930
<v Speaker 1>few months. AI CEOs will be meeting at the White

0:28:23.970 --> 0:28:27.630
<v Speaker 1>House with President Xi later this week. The New York

0:28:27.650 --> 0:28:30.980
<v Speaker 1>City Council will is convening a hearing on AI safety

0:28:31.040 --> 0:28:35.900
<v Speaker 1>risks next month, with Dario Amadei and Sam Altman requested

0:28:35.960 --> 0:28:38.860
<v Speaker 1>to appear. If we don't want them to be the

0:28:39.020 --> 0:28:43.660
<v Speaker 1>ones leading the conversation on AI regulation, which would only

0:28:43.720 --> 0:28:48.000
<v Speaker 1>result in an industry-friendly regulatory slowdown to curb competition and

0:28:48.040 --> 0:28:51.440
<v Speaker 1>stabilize market pressure, then the left will have to step

0:28:51.540 --> 0:28:54.700
<v Speaker 1>up and build on the data center backlash to seize

0:28:54.780 --> 0:28:58.780
<v Speaker 1>the very technology data centers are powering to destroy these

0:28:58.840 --> 0:29:02.380
<v Speaker 1>companies and put the future of AI in the control

0:29:02.520 --> 0:29:07.770
<v Speaker 1>of the public. Thanks for listening.