00:02:58.200So I decided I'm going to actually understand this issue better and research it and see, among other things, research, is it possible to control AI that's smarter than humans?
00:03:09.640We had a paper on it that we had accepted to the biggest AI conference in the world last December,
00:03:19.020If we build machines, if we build basically a species of super intelligent robots,
00:03:25.980they're way smarter than us, faster than us, stronger than us,
00:03:28.840and can build robot factories that build more robots.
00:03:33.120It seems pretty obvious that we might lose control.
00:03:38.440You just go down to the zoo here in D.C. and look at who's in the cages.
00:03:41.720It's not the most intelligent species because intelligence kind of confers control.
00:03:46.140But our paper found with a very nerdy calculation that, yeah, the best ideas people have for controlling some such a thing don't seem to work.
00:03:55.840Your work in theoretical physics came that mathematics is just not a representation of how you understand reality.
00:04:07.640Kermit Rom, that mathematics is actually reality itself.
00:04:12.580I mean, that was your drive in what made your name in theoretical physics for years, right?
00:04:17.500You were the cutting edge of that aspect of physics.
00:04:20.360The reason they gave me tenure at MIT, I think, is because I have done a bunch of very nerdy work with experimental data and connecting it with theory to figure stuff out about our universe, how much dark matter there is, stuff like that.
00:04:34.220But the thing you mentioned about math, you know, I love math, and I think it's become more and more clear that math describes a lot, not just about how a bottle will move if I throw it, but about a lot more in our world, including information processing.
00:04:55.060And it's exactly this idea that's given us artificial intelligence.
00:04:59.200You know, people, the core idea that's given us the AI revolution is this idea that intelligence is fundamentally, you know, the capability to accomplish goals is fundamentally about information processing.
00:05:14.880And it doesn't matter whether the information is processed by carbon atoms and neurons in brains or by silicon atoms in our machines.
00:05:22.920and this is precisely what's so terrifying
00:05:28.960about the trajectory that a lot of Silicon Valley
00:12:54.700It's to show the polling when we get back from the break.
00:12:57.30085% of the American people basically agree that this is common sense and you have to do this.
00:13:03.600The oligarchs themselves, and Andreessen's not one of the biggest oligarchs, but he's one of the biggest that he and Karp and Peter Thiel and Andreessen are taking the lead on the political side.
00:13:14.960They're setting up these PACs, and they're trying to—it's very simple.
00:13:18.760Whether you have the children that are being abused or the children that are being overwhelmed and their parents are being overwhelmed by artificial intelligence,
00:13:28.000or if you have these, you know, people are not being able to monetize their intellectual property.
00:20:29.080So the godfather of the whole field of AI said in 1951, this is Alan Turing,
00:20:34.140if we build these machines that can totally outsmart us,
00:20:37.360and then we should expect them to take charge.
00:20:43.540Now, the way to keep charge as humans is, of course,
00:20:46.220to make sure that we don't let them improve themselves,
00:20:50.180that we always have a human in the loop.
00:20:54.000And for that reason, a bunch of the leading AI folks
00:20:57.220actually signed on to a thing already in 2017
00:21:01.240saying recursive self-improvement is really, really risky.
00:21:05.240It is the AI kind of what biological gain of function is biology, except you're now dealing with really intelligent things that are making themselves smarter.
00:22:18.660We're just letting the robots and the non-robotic AI systems make themselves smarter because
00:22:24.260It's going to be great, work out great for humanity.
00:22:26.820But needless to say, you know, if all the AI R&D can eventually be done by machines, they can do it a lot faster than on the human R&D timescale of a year or so.
00:22:39.900So the next version that's better, the smarter AI might come after a month or a week or an hour or five minutes.
00:22:47.440And then that one can make a better one and that one can make a better one and off it goes.
00:22:51.860And when you keep regularly improving, if you keep regularly doubling anything, you know, pretty quickly, that's what we call an explosion in physics.
00:23:03.140If you double the number of neutrons in the chain reaction rapidly, that's what we call a nuclear explosion, you know.
00:23:10.560Double the amount of intelligence over and over, that's the intelligence explosion.
00:23:14.720So clearly this is something that would be incredibly reckless to do
00:23:18.580before you've figured out how you're going to contain this thing.
00:23:22.820With nuclear, we put a lot of research making sure that our reactors don't blow up,
00:25:06.160So they – and this gets to this broader issue we talked about before with power being very much at the core of what's really driving these forces.
00:25:21.060Well, this is the point that – this is what President Trump said about the kill switch.
00:25:26.380These – whether it's Elon Musk, that we know he wants to do this, or Altman or whatever the frontier labs, and Andreessen and Karp.
00:25:35.240and CARP is building a 21st century surveillance state right now on government money.
00:26:14.260They don't believe in representative government.
00:26:16.840They don't believe in the constitutional republic.
00:26:20.040There's this whole thing about markets and just having a company.
00:26:23.160They're like going back to Italy in the Renaissance.
00:26:26.900We have basically city-states where you have Anthropic here, and you have Elon Musk here, and you have Google here.0.73
00:26:33.860They are a feudal master and everything underneath them, and they want to have as few humans in that process as possible.
00:26:40.940This is this whole concept of going from 8 billion people down to 500 million for what they call the appropriate carrying capacity of the planet.
00:26:48.500Well, if you look at past autocracies, you know, pharaonic Egypt and forward, you know, what's often gone wrong ultimately for the despot was that there were some other humans who rebelled against them.
00:27:00.660So it's much more convenient if you can replace a lot of those humans by robots, which are just programmed to be fully obedient.
00:27:08.200I think that's a little bit of the seductive.
00:27:10.140There's no doubt in your mind, as you know, these companies and the researchers and you talk to people, that is the, besides the happy talk they put up here and all it's going to be, and we're going to give you some universal basic income, we're going to give you, you know, 50,000 a year to hang around and play video games because you don't have any meaningful work.
00:27:25.960Your strong belief is that is exactly what the intent of these oligarchs is.
00:27:30.940Well, I judge people not by their words, but by their deeds, right?
00:27:35.540So you take this thing about the kill switch, right?
00:27:41.180No, we're going to play it as soon as we get back from a break.
00:27:43.160And, you know, if the U.S. government wants to shut down some nuclear reactor because it's getting close to blowing up, there is a kill switch, an emergency shutdown procedure which will do it safely.
00:30:31.320Of course, we won this one in Florida, as we'll announce here momentarily,
00:30:35.960top of the hour, but we're losing on something that's overarching to everything.
00:30:41.020If we don't get the AI right, nothing else is going to matter.
00:30:45.160Axios, talk to me about this recursive in what we're talking about,
00:30:48.560the tempo and the kill switch is that even even anthropic is saying hey this thing's moving at an
00:30:54.800accelerating rate so with recursive you could have in 90 days something would take years could come0.76
00:30:59.880out in hours yeah in short the the dream of a lot of these transhumanists is to build a digital god
00:31:07.140they're mostly atheists so they want to build their own god and then they have this idea that
00:31:10.980they're gonna make it they're gonna somehow they're gonna merge with it or or that's
00:31:18.540something you hear on war room do you believe that to the core of your being having known
00:31:22.640having met these guys and see what they're doing they say stuff like this when they're drunk to me
00:31:28.240i'm not gonna i i want to respect what people tell me privately and not reveal any names but i
00:31:34.700just hang out in san francisco for a while at the right parties and you'll see and um
00:31:40.720So coming back to the fact that you and Bernie Sanders, that I get to talk with both of you today and that you guys both agree that humans have to stay in charge, that the government has to be able to shut down hacked data centers and so on, is just a fantastic illustration of the fact that this, duh, this is the right path.
00:32:10.720And it's quite astonishing to see companies resisting, you know, why would any good faith company resist letting the elected U.S. government, the democratically elected U.S. government, shut down their data center if it gets hacked?
00:32:26.440If, you know, I maybe I'm missing something.
00:32:30.500But to me, it just seems like an urge for companies to keep control.
00:32:35.380Take the argument. Let's go. Let's go to the argument of that. President Trump and everybody's under this. We've had a Sputnik moment. The Chinese are competitive. And Jensen Wong's there with the king last night in white tie. Right. And who's an agent of influence for the CCP. What about their argument that if we don't allow these companies to have absolutely no controls whatsoever, that we will lose this race to the Chinese Communist Party?
00:33:01.420That's like saying we have to allow anyone who wants to buy hydrogen bombs in supermarkets, otherwise we would get invaded by Russia.
00:33:11.520It's like saying if the U.S. government actually, after getting all the intel from our intelligence community, comes to the conclusion that there is a particular corporate data center that's right now doing a cyber attack against the U.S. government,
00:33:27.880you know why shouldn't the u.s government have the right to shut that down why would stripping0.56
00:33:34.220that right from our very government help china in any way like duh that's that makes about as0.91
00:33:40.720much sense to me as saying that we must allow character ai to sell ai girlfriends to kids0.82
00:33:47.440because china but we have to also at this time put these oligarchs on notice we're not going to0.83
00:33:54.920allow them to build the ecosystem in which China can even be competitive.
00:33:59.540The chips, Jensen-Wang should not have the free ability to sell these advanced chips
00:36:03.880Max, good to see you through the digital framework.
00:36:07.100You look a lot better on screen than you do in person, I'll tell you that.
00:36:11.500You know, Steve, the article that you guys were talking about earlier,
00:36:15.120the Axios article, Behind the Curtain,
00:36:17.940I think it brings home the reality of the situation.
00:36:22.660You know, superintelligence is undoubtedly an undesirable outcome for all of this, but it's still theoretic.
00:36:30.600What these guys are talking about basically are just six points that can't be denied that AI is the fastest growing industry in history.
00:36:41.360one of the fastest, if not the fastest adopted technologies, that you already have systems that are dangerous enough that Anthropic, for instance, would not release it to the public,
00:36:55.060would only release it to a certain select group of corporations, and that these systems are capable of, to some extent, building themselves.
00:37:04.840I mean, I don't want to oversell that point, but undoubtedly, especially in companies like Anthropic, they're using the AIs to build the AIs.
00:37:15.480So you're you're approaching that point of recursive self-improvement.
00:37:20.380And so none of these things can be denied. You can spin it one way or you can spin it the other, but they can't be denied.
00:37:25.960And one of the points that they make, I think it's like their fifth point, is that there is a massive backlash from the public because people are becoming aware of this situation.
00:37:36.660They feel very powerless in this situation.
00:37:39.860And, you know, Altman had his home attacked.
00:37:44.960You had an official, a councilman in Indianapolis who had his home attacked.
00:37:50.140And again and again, I've had reporters ask me this.
00:37:52.980I had an editor for one of the major publications ask me this.
00:37:57.600Well, do you feel responsible for this?
00:38:01.820I think that these companies have created a situation which people are extraordinarily fearful and pointing out specific psychopathic activities is ridiculous.
00:38:12.120These companies are creating the situation is real.
00:38:19.120They have put us in a very dire circumstance and they have to be held accountable.
00:38:24.480And I think deflecting with these like random acts of violence is absurd, especially given how many children have killed themselves at the behest of chatbots or how many people have died due to the decision compression of AI systems in our military.
00:38:40.500Let's let's let's go. Somebody knows the threat. President Trump. Can we get the kill switch video? Can we play that? Then Max like you to jump.
00:38:47.760Should government have some safeguards? Should there be a kill switch for some of these AI agents?
00:38:58.500So there, I was so delighted to see our president say, yes, there should be a kill switch.
00:39:05.460In other words, if our commander in chief wants to shut down an American data center, he should have the ability to do so for the benefit of America.
00:40:43.100So we're talking now about the fact that it's completely legal for any AI company right now to make something more powerful than Mythos and just release it to the public.
00:41:44.280Listen, as I said, most of our audience are not wild about many aspects of the U.S. government.
00:41:51.720But this is a different question. The oligarchs that run these five or six biggest companies all see themselves superseding that government where you have no say-so at all in this.
00:42:04.320That's why I call it techno-feudalism. They want to go back. And this is why Elon keeps talking about the they'll pay you.
00:42:11.720Now it's not universal basic income. It's universal high income. There's always going to be a tip. Joe Allen, you got something?
00:42:17.880Yeah, there's I mean, there are solutions being proposed. There's two pieces of legislation right now, you know, Trump America, AI Act from Marsha Blackburn, and then the the act being put forward soon by Jay Obernolte and Ted Lieu.
00:42:35.160In both of those, you have, you know, the call for a federal agency to oversee these companies.
00:42:42.640What's interesting, though, and I'm not trying to throw shade on Obernolte and Lou,
00:42:47.700but in their American Leadership in AI Act, it would appear they're going to position Casey,
00:42:53.300the Center for AI Standards and Innovation, as the key player in this or a key player in this.
00:42:59.720assuming that that's the case you also have open ai recommending that right you have sam altman
00:43:06.520recommending that and um you know i wonder if we won't end up in a place if we're not careful
00:43:11.840where you end up with an agency that has already been captured by these companies or at the very
00:43:18.800least influenced um anyway just to to add some sunshine to an already uh sunny day over at the
00:43:26.260war room well no it's it's uh in the agency to me the last thing we need is another federal
00:43:32.560federal regulatory apparatus that doesn't work it's got to to me be like the atomic energy
00:43:39.280commission was at the beginning that kept uh that kept oppenheimer and these guys on the straight
00:43:44.460now anyway short commercial break uh max will stick with us for another segment we're going
00:43:48.760to go to tallahassee at the top of the hour get an update on all this
00:43:52.560The American health care system is broken, and for most Americans, nothing changes.
00:44:03.740There's still delays, denials, high-cost insurance roadblocks.
00:44:07.820So when I find people doing things differently, I talk about it.
00:44:12.880All-family pharmacy is not your typical big-chain pharmacy.
00:44:16.300This is an independent, family-owned pharmacy that gives you access to over 400 medications delivered straight to your door.
00:44:24.880They've got ivermectin, antibiotics, antivirals, NAD+, even your daily maintenance medications, and so much more.
00:44:34.860If you already have a prescription, your doctor can send it directly.
00:44:38.740If you don't, their doctors handle it.
00:44:41.300As long as there is a medical necessity, they'll take care of you.
00:44:44.940and i'll tell you this the feedback from people listening to this show and watching
00:44:49.140has been incredibly strong people are using it it's working for them and they're sticking with
00:44:54.800it that's because it cuts out the delays the middlemen and all the usual nonsense this is
00:45:01.540about being ready before you need it go to allfamilypharmacy.com that's all one word
00:46:26.980You heard an example of it right there.
00:46:28.960Or, you know, whatever politician people listening to this likes or whatever ones they dislike, you know, nothing is as bad as that.
00:46:43.280You know, on my more cynical days, I tell myself that no matter how bad things are, it can always get worse.1.00
00:46:52.080This transhumanist vision is worse.1.00
00:46:55.340Someone not sure whether humans should continue to exist or not.
00:46:58.420And let's be clear, this is not limited to Peter Thiel.
00:47:05.780It's very, very popular in a lot of my people, among a lot of folks who work in these companies.
00:47:15.260You can go look up Sam Altman Merge, and you can see an article the CEO of OpenAI wrote saying that humans are going to build their own replacement.
00:51:53.820This should be catnip to any politicians because if you want to be on the right side of history, you want to be on the right side of the voters.0.93
00:52:00.200The transhumanists, the accelerationists, very loud, but they're very few.0.98