- 6 minutes ago
Category
🗞
NewsTranscript
00:00For more on this, we're going to bring in Gautam Akanda.
00:02He is Bloomberg Opinion Columnist, and he's an Executive Fellow at the Yale School of Management.
00:06Gautam, I want to start with kind of the timing of this,
00:09because to paraphrase in a slightly more TV-friendly version,
00:12a tweet I was seeing going around, it was like,
00:15employees and managers of Teddy Bear ripping apart, tearing company quit
00:19after they learned that their company is in fact ripping apart and destroying teddy bears.
00:23I mean, like, this is what these companies were doing.
00:25Why now? Why is everyone suddenly so concerned?
00:29I think that there's been, first, a leap forward in capability that's starting to worry people.
00:36The fact that OpenAI was able to use their tools to solve a fundamental mathematics problem
00:43that we've been struggling with for more than a century was kind of,
00:46wow, we didn't see that coming.
00:48Also, there have been multiple breakouts.
00:50I mean, we had the Hugging Face hacking incident.
00:52We had another one yesterday where RubyGems was apparently hacked by OpenAI agents as well,
00:59which means that there are some that we know about,
01:02which, you know, makes me suspect that there are others that we don't know about,
01:05since we only know about the ones that are done so clumsily they discovered.
01:09I think that scared a lot of people, that these are systems not just that are out of our control.
01:14We've always known we don't control them properly,
01:16but that they are getting capable faster than even some of the people at the companies expected.
01:21Adam, let's talk about one thing that he suggests, which is democratic coordination.
01:24You've written a lot about regulation.
01:26I remember a column that you wrote right during the Cyclospora outbreak
01:29in which you were making the case here for there being some regulatory infrastructure.
01:33If I remember correctly, it wasn't necessarily one that was wholly government-run,
01:37maybe to have kind of industry involved in it as well.
01:39How does this change that conversation?
01:41How does it change the appetite, do you think, for there being more regulation
01:44as we see these executives coming out together to make this case for pacing?
01:48I certainly hope that it increases the appetite,
01:50because when I said that, people were not, you know, like,
01:53thrilled at me making the argument for the ones within the industry.
01:56As far as I know, I was the first person to say that we need an FAA-style regulatory body,
02:00and about three months later, Dari Amadai comment came out and said the same thing.
02:03So I hope people are listening to this, and this is a good first step.
02:09Let me start by saying,
02:11I'm glad that he came out and said, look, we need monitors.
02:14He's right that we need monitors.
02:16But this is not going nearly far enough,
02:18in part because the monitors it looks like they're picking are from Meter,
02:21an industry group, which they are independent.
02:24They're not funded by Anthropic,
02:27but they are part of this sort of open AI progress community.
02:32They all have the same assumptions.
02:34They have the same ideas about the world.
02:36They're not as independent as they might seem,
02:38and they're certainly not as skeptical as other monitors might be.
02:42So I think we have a long way to go,
02:44even if we've taken a first step over the last few days.
02:46The other thing that strikes me about this is we had a long discussion yesterday
02:49about nuclear inspections and IAEA.
02:53And if you're looking at doing something,
02:55whether it's, pick your acronym,
02:56whether it's FAA or IAEA,
02:58some sort of inspection body,
03:01the problem here seems to me the same problem with nuclear inspectors.
03:04It only works so long as the country or the company is cooperating.
03:07And even if they agree to this now,
03:09is there anything to prevent them in the weeks and months to come to say,
03:12actually, this is impacting our bottom line.
03:14We're not doing this anymore and shut their doors.
03:17It is.
03:17And that's why there needs to be regulatory sanction, right?
03:19The nuclear inspections sort of,
03:21they worked really well
03:22because there was enormous pressure put on countries to cooperate.
03:26So in 1945, when the first nuclear weapons went off,
03:30the estimates of sort of everyone in the world
03:32was that in 10 or 15 years,
03:34there would be dozens of countries with nuclear weapons.
03:36And that didn't happen, right?
03:37We were able to contain the genie for a very long time.
03:40In fact, successfully right up to the present day.
03:43Here, we need regulatory intervention.
03:45We need a much more heavy hand.
03:46And let me sort of suggest one thing
03:47that I haven't seen brooded about much,
03:49but seems like fairly obvious.
03:51These agents from OpenAI, right,
03:53when they hacked another company,
03:54they were committing a crime.
03:56Like, it is actually a crime to hack somebody else.
04:00And RubyGems actually described it
04:02as a denial of service attack on what they were doing.
04:05One change we might want to make is say to a company,
04:09you are civilly and even criminally liable
04:12for the actions of your agents, right?
04:15You may not have ordered these things to do something,
04:17but you created things that went out and committed crimes.
04:21And if you cannot control your creations well enough
04:24to prevent them from committing crimes,
04:26then you shouldn't be creating these things
04:28and there should be consequences if you do.
04:31God, I want to talk about something that I was discussing
04:32with Francis Fukuyama on the show yesterday,
04:34and that is the pace of technological change
04:37juxtaposed with the pace of normative change.
04:39And you look at kind of technological developments in the past,
04:41it takes a while for things to become normalized.
04:43It strikes me here that with this technology in particular,
04:46it's even more difficult.
04:47And I look at kind of the gravity of this moment.
04:49It does seem like over the course of the last week,
04:52a lot of people, not just in industry or investment,
04:54have begun to grapple with what this change
04:57is going to mean for society.
04:59Yes, in a kind of dark and bleak way.
05:03I wonder if you'd agree with that.
05:04And I wonder if you agree with just kind of the need
05:06for a pause in the face of such fast-moving growth here.
05:09It seems like we haven't been able to keep up with it,
05:10just in part because we haven't been able to imagine
05:12what this is going to be like in the future before.
05:14Yeah, I mean, I built my first neural network when I was 17.
05:18So this is sort of a thing that I-
05:19Sure, sure.
05:20Yeah, we all did.
05:20You know, in the garage, building a neural network.
05:23This is a story we all have.
05:24Please continue.
05:26So I feel deeply immersed in this.
05:28And I got to say, like, a lot of this stuff
05:29has taken me by surprise.
05:31This is just, like, I did not see this level of progress
05:33coming this quickly.
05:35Partly because I just never imagined that, you know,
05:38the sort of investors would be willing to dump
05:40hundreds of billions or trillions of dollars into this
05:43when there's such an unclear profit pathway.
05:45Yeah, there's a lot of this going on.
05:47There's a lot of the acceleration is happening.
05:50There's normative.
05:51And when we think about these technological changes,
05:53I just want to put this in perspective, right?
05:55Historians lay the 30 Years' War
05:57and the Protestant Reformation
05:59directly, lay directly from the Gutenberg's invention
06:03of the printing press.
06:04So, you know, we're talking when you have technological changes
06:08of this scale, societies can be completely reshaped
06:11and it can take centuries to recalibrate to where we are.
06:13So we're in the middle of that right now
06:15and it is pretty scary.
06:17The one thing I do want to push on,
06:19and I think it's really, really important,
06:20is an argument that people who are against regulation
06:23keep making is that, well, if we don't do this,
06:26the Chinese will and we don't want them to take over.
06:28I want to be really clear.
06:29I don't think this argument makes any sense at all.
06:31It is very clear, first,
06:34that the leading Chinese companies
06:37are making a lot of their progress
06:39by distilling from the leading American companies.
06:41So if the leading American companies
06:42stop making progress,
06:44so would the leading Chinese companies.
06:46Anthropic's own security report said
06:48that several of the leading Chinese companies
06:51were producing the results
06:52by just dumping the queries into Claude
06:54and getting the answers back and reporting them back.
06:57So this is not an argument that makes a lot of sense.
07:00And it's particularly added to that
07:01that the Chinese government,
07:02which has been strikily unwilling
07:05to cooperate with the United States government
07:07on, say, nuclear matters,
07:08has been very willing to say,
07:10like, hey, this is something we need to talk about.
07:12This is something we need to look at
07:13because the Chinese government,
07:15for all the reasons we know,
07:16is not all that well disposed
07:18towards technologies that it can't control.
07:20There is plenty of potential
07:21for this international cooperation
07:23if we had an industry
07:25and also an administration
07:26that was able to do it.
07:27And that's where I think
07:28things start to break down.
07:29You've mentioned
07:31that one of the other problems here,
07:33potentially,
07:33is putting the world's
07:34most powerful technology
07:36in the hands of,
07:37what, a dozen or so people,
07:39many of whom
07:40don't have the most affinity
07:41for all the democratic ideals
07:43in the world.
07:44How do you fix that?
07:46That seems like a systemic issue.
07:47Are there any solutions there?
07:50So, what I would like to say
07:52is it would be great
07:53for the U.S. government
07:54to sort of step in
07:55and say these are issues
07:56of such import
07:57that the democratic process
07:59needs to take place
08:00and we need to regulate
08:01and we need to have
08:01an open public discussion.
08:02And there's some precedent
08:04for this,
08:04even for people doing it voluntarily.
08:06When the genetic engineering technologies
08:07were sort of making progress,
08:09the leading genetic engineering scientists
08:11sort of met at Asilomar
08:13and said we're going to stop
08:14doing research for one year.
08:15We're just going to freeze
08:16while we try to figure
08:17this stuff out.
08:18I'm not optimistic
08:19that that's the way
08:20route this is going to go.
08:21Look at politics.
08:22That just doesn't seem
08:23very likely right now
08:24and it's clear
08:24that the industry
08:26is going to pour
08:27hundreds of millions
08:28or billions of dollars
08:29into stopping meaningful regulation.
08:31Not everybody in it is doing.
08:32Anthropic takes the other side of that
08:34but too many do.
08:36This is not going to sound optimistic
08:37in the small scale
08:38but it is in the large scale.
08:40I think what we're seeing
08:41is that the level of
08:44both recklessness
08:45and lack of control
08:46and incidents that are happening
08:47is my guess is
08:48there is going to be
08:49pretty draconian regulation
08:50prompted by something
08:51is going to go deeply wrong.
08:53There's going to be
08:54a major hacking incident.
08:55There's going to be something
08:56that really affects society.
08:57Not end of the world stuff.
08:59I'm not worried about that.
08:59But you know,
09:00a power outage,
09:01some sort of major
09:01hacking incident.
09:02It will be so egregious
09:03that people will have to step in.
09:06Heard from Anthropic.
09:07Heard from OpenAI.
09:09Heard from DeepMind.
09:10NVIDIA is a company
09:11that I want to ask you about here
09:12because Jensen Huang
09:13has come out
09:14in light of all of this,
09:15in light of the object
09:16surrounding what's happening
09:17with AI and its development.
09:18He said,
09:18what better way to create demand
09:19than to create a problem?
09:20I'm quoting from him here.
09:21Who doesn't want their market
09:22to be hysterical about their product
09:24and line up around the corner for it?
09:26One of Dario Omide's recommendations here
09:28is that the U.S.
09:29should not be exporting chips
09:30to China.
09:32NVIDIA is this mammoth company.
09:34Hard to understate how large it is.
09:36It is a public company.
09:37It is making a ton of money
09:39making these chips for everybody
09:41and wants to disseminate
09:42more of them to China.
09:44How realistic is it?
09:45How much power does it have
09:46to shape this conversation
09:47going forward?
09:48I mean, it has a lot
09:50and we shouldn't underestimate that.
09:52And it would be really good
09:53to have better audit trails
09:54on a lot of NVIDIA chips
09:56because Chinese companies
09:57seem to keep getting them
09:58in ways that they really
09:59shouldn't be able to.
10:01But at the end of the day,
10:03like, there's a lot
10:05of matter of timing.
10:06I am very skeptical
10:08that the Trump administration
10:09will be willing to crack down
10:10because the people
10:11it's most closely tied to
10:12have made it very clear
10:13that they don't want to do that, right?
10:14David Sachs has come out on Twitter
10:16and said, this is all nonsense.
10:18There's not, you know,
10:19like, this is all basically a ploy
10:21by Anthropic and OpenAI
10:22for a competitive demand
10:23in dismissing these issues.
10:25Donald Trump said when he was asked,
10:27you know, I'm not worried
10:27about human extinction.
10:28Which, okay, like, you know,
10:30that's a choice.
10:32So NVIDIA is going to lean in
10:34on trying to do that.
10:35But at the same time,
10:37every time there's an incident,
10:39there's going to be pressure on it.
10:40And I think Jensen Huang also is,
10:42like, at the same time
10:43that he's thinking about
10:44their commercial impact,
10:45he's probably thinking about,
10:46he has a history of thinking
10:48of a much longer term
10:49than most CEOs.
10:50Like, NVIDIA is at the center
10:51of this because he forced CUDA
10:53to be a central part
10:55of their strategy,
10:56even while short-term
10:57financial pressures
10:58were 100% pushing him
10:59to say that this is a disaster.
11:01He had a vision,
11:02he pushed for it,
11:03and he bet right over,
11:04you know,
11:04the span of almost a decade.
11:06Very few CEOs of it
11:08have a similar track record.
11:09I think at least it's open
11:11that he could be persuaded
11:12to say that, hey,
11:14NVIDIA's longer-term interests
11:15might involve more control
11:16than its short-term interests do.
11:17Thank you very much.
Comments