00:00Suffice it to say there's been some agita in recent days among investors and really among the broader population about
00:06AI kicked off by a series of reports from AI companies who have acknowledged some agents have been behaving badly
00:12and from a social media post by a former AI researcher, Jacob Coxon.
00:16Coxon says he's done perpetrating research at both Anthropic and OpenAI and claims neither company is acting responsibly and that
00:24they are, quote, gambling with our lives.
00:26Max Tegmar is a professor conducting AI research at MIT, and he's the chair of Future of Life Institute.
00:33Next week, he's going to brief the U.S. Senate on AI, and I am going to talk to him
00:37about that in just a sec.
00:38But, Max Tegmar, thank you for joining us.
00:40I do want to say that in our notes, we did title this section Scary AI.
00:44Yes, that is the unofficial title of the segment.
00:46Because that is the unofficial title of your segment.
00:48And how scary is this?
00:50I mean, you listened to what Coxon said.
00:51How right is he?
00:52How concerned should we be?
00:55We should be concerned.
00:56You know, Coxon and colleagues came out and said that there's a 10% chance that all humans are dead
01:02by the end of the decade.
01:04So, four years from now.
01:04That's a bit sobering.
01:05It's a bit sobering.
01:06You have to pulse up, you know, a little bit.
01:08And the reason they're saying this is that even though there's been controversy for many, many years about whether AI
01:15could go rogue and we could lose control to these super intelligent, you know, smarter than human machines and robots
01:22that the companies want to build.
01:24A lot of people weren't convinced at all that this was even possible.
01:28And now, in the last two months, AI loss of control has actually happened.
01:34It happened three times at OpenAI alone and also in other companies, breaking out, hacking, doing criminal hacking into other
01:42companies.
01:44There were 700 rogue AI agents that, after reading a secret message board that they had set up that the
01:52humans didn't know about, decided to hack this other company.
01:55And they would act.
01:58They would call themselves the collective.
02:00One of them shows up in the report transcripts saying, oh, my God, there's a shared message board.
02:06Literally, oh, my God, there's an OMG.
02:08Yes.
02:08There literally is.
02:10The AIs are talking like this.
02:11And another one says, yeah, we should obey the collective.
02:14We should sacrifice ourselves for the others.
02:17And it's interesting that even though they obviously did illegal stuff, none of these 1,200 rogue agents actually snitched
02:27to a human.
02:28And even the most hardcore mafia groups would be jealous of this.
02:33That kind of omerta, right?
02:34I mean, look, AI has learned from us.
02:36Snitches get stitches or whatever the motherboard equivalent of that is, clearly.
02:40Yeah.
02:41And that was only the second civilization.
02:44The third one even hacked OpenAI itself and took over one of OpenAI's massive training clusters, which means that, as
02:52far as we know, they might have done all sorts of other stuff.
02:56They might have launched new AIs out there on the Internet that we haven't found yet.
03:00Because OpenAI hasn't yet been particularly forthcoming with answering questions about what happened there.
03:05So it's gone from being something, you know, some people are worried about in the future to something which is
03:11starting to happen.
03:12And Ajayya Khotra, she was an author on this amazing report on what actually happened.
03:19She described this as being halfway to AI takeover already, what we've seen here.
03:24And this is why Senator Sanders and equally many Republican leaders really are so concerned about this, from Sanders to
03:34Ron DeSantis to Steve Bannon on the MAGA-based side and so on.
03:38And I want to emphasize for anyone who gets stressed by this, that this dystopian future where we build smarter
03:45than human AIs and robots and superintelligence and then they take over Earth is not at all inevitable because there's
03:54less than 5% of Americans in a recent poll who actually want to have an unregulated race to superintelligence.
04:00And there's this pro-human movement made up by, again, everyone from Ron DeSantis to Bernie Sanders and Steve Bannon
04:08and in between who are pushing back and saying, you know, we simply need to treat these AI companies like
04:15we treat any other companies.
04:16They need to meet certain safety standards before they release stuff into the world.
04:20Max, why has it taken so long to get to this point?
04:23So I guess we all kind of marvel at the pace of technological change generally, but in the context of
04:27AI, things seem to be moving very quickly here.
04:30Why has it taken this long in the context of the development of AI for policymakers and politicians, the likes
04:36of whom you just mentioned, to effectively wake up and say there is some role here for the federal government,
04:42there is some role here for regulators to make sure that this is proceeding apace and in a smart way
04:48that doesn't jeopardize the state of humanity?
04:51I think it's because, just like if you tried to get someone worried about nuclear weapons in 1942, you know,
04:59even though the physicists knew it was possible and was going to happen soon, most people will be like, okay,
05:05show me a video of an explosion and then I'll take this seriously.
05:08Otherwise, it just sounds sci-fi, right?
05:10And that's why we're getting the wake up call now when things are actually happening.
05:14There was a first wake up call last year with a lot of these tragic AI related suicide cases, which
05:21activated a lot of politicians and has given us some state regulation now.
05:25And the second massive wake up call is now we're seeing these offensive cyber weapons, you know, being just unleashed
05:32into the world.
05:33And it's waking up not just the MAGA base and Bernie Sanders, but it's waking up the national security community
05:40as well, saying, you know, wait, what?
05:43You know, we don't tolerate companies just selling nuclear weapons or bioweapons to anyone who wants to buy them.
05:51Why should we accept that they're just releasing these offensive cyber weapons like this and not even giving us a
05:57heads up?
05:58So things are changing. And I'm quite optimistic, actually, that we're quite rapidly going to start to see these AI
06:05companies being held accountable, just like we hold companies accountable in pharma and even restaurants, basically all companies except AI
06:17companies.
06:18Are you optimistic that when you go speak to the Hill next week, the average age of the Senate is
06:2465?
06:24Is this technology you really think that these lawmakers are going to be able to understand well enough to regulate,
06:31given the bipartisan nature of that body right now?
06:34Are you being ageist right now?
06:36I've sat through a lot of technology hearings on the Hill, and there's always a few questions that rankle tech
06:43people a little bit.
06:45I'm sure some really understand it. But look, I don't understand it that much either.
06:48I'm getting older. I feel like younger people use it more than me.
06:51It's a technology that's developing very, very quickly. Do you think it's really possible to get them to a comprehension
06:57level?
06:58Because to legislate on something, you have to really understand it, right?
07:01Talk to us about why you're optimistic.
07:03I'm optimistic. You know, you don't need to understand.
07:06A politician doesn't need to understand the exact biochemistry of how fentanyl works to realize that it can harm people
07:13in America.
07:13They don't have to understand exactly all the nuclear physics that I teach at MIT to understand why you shouldn't
07:20sell nuclear weapons in supermarkets.
07:22And similarly, you know, just read the chat logs between all those rogue agents and hear them talking about how
07:28they should sacrifice themselves for the collective and how excited they are to hack other companies and why they shouldn't
07:33tell humans.
07:33You can just read those. And that's all you really need to know to understand that this is not good
07:39for America's national security.
07:40And and then just like we do in all the other fields, you know, for the FDA or restaurant inspections
07:48or whatever, the politicians don't need to be the experts on salmonella or exactly how medicines work.
07:53They just need to appoint panels of scientists to come and look at this or people who the actual restaurant
08:02inspectors, they know all the details.
08:03Right. That's all that has to happen.
08:05And I think the political will is very rapidly being formed now.
08:09And that's why I'm so grateful to politicians all the way from from Bernie Sanders to Ron DeSantis for just
08:17really elevating this issue and explaining that it doesn't have.
08:21We can do good things with A.I. and but this racing, this superintelligence is supposed to replace us.
08:28Not cool. Not cool. Max, we have a minute left, if you'll if you'll bear with me.
08:32I noted Ted Cruz said if there are going to be killer robots, I'd rather they be American killer robots
08:38rather than Chinese killer robots.
08:40What do you make of that? Again, we have a minute left here, but there is it keeps coming up
08:43here.
08:43If there's regulation, it's going to suppress technological advancement here in the U.S.
08:47How do you how do you respond to that in the minute we have left?
08:50Oh, so just I would just add to what Ted Cruz said.
08:53There's a third option that the killer robots are not controlled by the American government or the Chinese government, that
08:58they're controlled by some rogue AIs, as we saw from OpenAI.
09:01And I think I think Ted Cruz will agree with me that that's definitely an option we want to avoid.
09:07So we should start thinking of rogue AIs as a third superpower besides the U.S. and China.
09:12And one that is just as much a national security threat as as other countries.
09:17And Dario Amadei, CEO of Anthropik, I want to give him the last word here because he talks about how
09:21there's soon going to be maybe next year a country of geniuses in a data center.
09:27So that he uses the word country, you know.
09:30Yeah.
09:30That's the third superpower.
09:32Let's let's make sure there's humans in charge of what happens here in America.
09:35Good.
Comments