Latest / Useful Idiots with Katie Halper and Aaron Maté / "The chance of an AI World War is now 50% in the next 30 years"
Transcript
- 0:00SPEAKER_00: So like as of now,
- 0:01SPEAKER_00: the Trump administration actually has access to the most powerful AI in the world,
- 0:06SPEAKER_00: Mythos, and the public doesn't.
- 0:10SPEAKER_00: Pretty much everyone agrees that if you get around the guardrail on Fable,
- 0:17SPEAKER_00: you will have an easier time building not just a bioweapon but
- 0:20SPEAKER_00: a genetically engineered bioweapon than has ever been the case
- 0:27SPEAKER_02: Welcome to a bonus episode of Useful Idiots.
- 0:29SPEAKER_02: We're bringing you an interview that I did with Robert Wright.
- 0:32SPEAKER_02: He is a journalist and author.
- 0:33SPEAKER_02: His brand new book is called The God Test,
- 0:36SPEAKER_02: Artificial Intelligence and Our Coming Cosmic Reckoning.
- 0:40SPEAKER_02: The God Test analyzes the coming AI revolution as a moral and spiritual equivalent
- 0:46SPEAKER_02: to a biblical test from God.
- 0:48SPEAKER_02: Here is my interview with Robert Wright.
- 0:56SPEAKER_02: Robert Wright, thanks for being here.
- 0:57SPEAKER_02: Thanks for having me, Aaron.
- 0:58SPEAKER_02: The book is called The God Test, Artificial Intelligence and Our Coming Cosmic Reckoning.
- 1:04SPEAKER_02: Let's start with the title.
- 1:05SPEAKER_02: What is the God test?
- 1:06SPEAKER_00: I think humankind is posed by artificial intelligence with the kind
- 1:11SPEAKER_00: of test you would expect the God to give.
- 1:15SPEAKER_00: Like in the Bible,
- 1:16SPEAKER_00: you can find places where it says salvation is possible,
- 1:18SPEAKER_00: but you got to get your act together.
- 1:20SPEAKER_00: What exactly that
- 1:21SPEAKER_00: That means depends on the part of the Bible you're reading,
- 1:24SPEAKER_00: but often it entails in some sense a kind of a moral upgrade by one definition
- 1:31SPEAKER_00: or another.
- 1:32SPEAKER_00: I personally think if we're going to handle the AI revolution well,
- 1:37SPEAKER_00: we're going to have to do it as a global community, a cohesive global community.
- 1:42SPEAKER_00: Much more cohesive than we've ever been.
- 1:43SPEAKER_00: That's going to mean getting over these ridiculous wars.
- 1:48SPEAKER_00: And for that matter,
- 1:48SPEAKER_00: getting over a certain amount of intranational polarization
- 1:52SPEAKER_00: that makes policymaking hard even at the national
- 1:58SPEAKER_00: And I think for that to happen,
- 2:00SPEAKER_00: we're all going to have to get better at looking at things from the point of view
- 2:06SPEAKER_00: of other people, including adversaries, adversarial groups.
- 2:11SPEAKER_00: Matt Wilson mttwilson
- 2:38SPEAKER_00: You know, it's a moral and even you could argue spiritual challenge that we face.
- 2:46SPEAKER_00: Some of that argument depends on your accepting my premise that AI is going to be
- 2:50SPEAKER_00: a very, very, very powerful technology.
- 2:53SPEAKER_00: I try to argue early in the book why I think that is and clearly explain for
- 2:59SPEAKER_00: a non-technical audience why the capabilities have grown
- 3:02SPEAKER_00: so fast why they'll keep growing but even if you don't buy
- 3:06SPEAKER_00: that argument I think there are reasons that almost
- 3:09SPEAKER_00: Anyone who wants to regulate this technology in any way would acknowledge that at
- 3:15SPEAKER_00: a minimum we need to mend fences with China and be on better terms with China
- 3:22SPEAKER_00: for reasons I can explain.
- 3:24SPEAKER_02: Well, let's get into that.
- 3:25SPEAKER_02: So the way AI now is discussed, so much of it is tied up in terms of the race with China.
- 3:32SPEAKER_02: Recently,
- 3:32SPEAKER_02: you had the Trump administration telling Anthropic
- 3:36SPEAKER_02: to basically take its latest AI model fable offline.
- 3:40SPEAKER_02: And that has a huge tie in to China and supposedly the fear
- 3:45SPEAKER_02: about China acquiring American technology.
- 3:48SPEAKER_02: Talk to us about that.
- 3:49SPEAKER_00: If all goes well I'll have a piece in the Washington Post about this by
- 3:51SPEAKER_00: the time this comes out but my argument is that first of all skepticism of
- 3:58SPEAKER_00: the Trump decision is proper and overwhelmingly in
- 4:02SPEAKER_00: the tech community there's been condemnation of it and a certain amount
- 4:06SPEAKER_00: of skepticism is proper it's a new exercise of executive power we need
- 4:11SPEAKER_00: a clearer explanation
- 4:12SPEAKER_00: And I note that with AI concentrations of executive power especially scary given
- 4:18SPEAKER_00: the surveillance power and the persuasive powers of AI.
- 4:23SPEAKER_00: So like as of now the Trump administration actually has access to
- 4:27SPEAKER_00: the most powerful AI in the world mythos and the public doesn't.
- 4:33SPEAKER_00: Now, you know,
- 4:34SPEAKER_00: and that's something we should think about and we need Congress to pass laws
- 4:39SPEAKER_00: about like under what circumstances can and can't the executive branch do this
- 4:45SPEAKER_00: but meanwhile we don't have those laws
- 4:48SPEAKER_00: And one thing I argue is that,
- 4:50SPEAKER_00: you know,
- 4:50SPEAKER_00: that the model he took off the market in effect,
- 4:54SPEAKER_00: Anthropics Fable,
- 4:55SPEAKER_00: which is just Mythos,
- 4:56SPEAKER_00: the most powerful model in the world,
- 4:58SPEAKER_00: with some guardrails on it,
- 4:59SPEAKER_00: so it'll refuse to do some things,
- 5:02SPEAKER_00: unless you get around the guardrails, which people have been known to do.
- 5:06SPEAKER_00: I contend that it's not crazy,
- 5:10SPEAKER_00: just on the merits,
- 5:11SPEAKER_00: leaving aside this form of exercising executive power,
- 5:14SPEAKER_00: it's not crazy on the merits to go, wait a second,
- 5:18SPEAKER_00: Do we have to put this on the market right now?
- 5:20SPEAKER_00: Pretty much everyone agrees that if you get around the guardrail
- 5:25SPEAKER_00: on Fable You will have an easier time building not just a bioweapon but
- 5:30SPEAKER_00: a genetically engineered bioweapon
- 5:33SPEAKER_00: than has ever been the case before.
- 5:37SPEAKER_00: So I personally think it makes sense to stop,
- 5:40SPEAKER_00: you know,
- 5:40SPEAKER_00: Trump aside,
- 5:41SPEAKER_00: you know,
- 5:42SPEAKER_00: in principle, we shouldn't rush to the judgment that this kind of restraint of
- 5:48SPEAKER_00: A.I.
- 5:49SPEAKER_00: is bad and yet the A.I.
- 5:51SPEAKER_00: community,
- 5:52SPEAKER_00: the tech community almost unanimously has rushed to that judgment and this is
- 5:57SPEAKER_00: where China comes in.
- 5:58SPEAKER_00: I'm going to quote this guy Alex Stamos, very prominent cyber
- 6:03SPEAKER_00: cybersecurity researcher.
- 6:04SPEAKER_00: Smart guy.
- 6:05SPEAKER_00: I mean, I think he believes this, Ernest, so far as I know.
- 6:08SPEAKER_00: But this is what he says about delaying this fable thing from entering the market
- 6:14SPEAKER_00: by even a couple of weeks or actually taking it off the market.
- 6:18SPEAKER_00: It had been briefly on the market.
- 6:20SPEAKER_00: So he says, quote,
- 6:22SPEAKER_00: It's a complete disaster for the American AI industry.
- 6:25SPEAKER_00: This is a complete and total disaster.
- 6:27SPEAKER_00: Now, Aaron, actually, I do think the industry is still standing.
- 6:31SPEAKER_00: I doubt that's true.
- 6:32SPEAKER_00: Then he goes on to say, it is a great, great week for the Chinese AI industry, a fantastic week.
- 6:38SPEAKER_00: and a huge own goal for the United States in the race to dominate
- 6:42SPEAKER_00: the 21st century be a domination of AI.
- 6:46SPEAKER_00: So two things.
- 6:46SPEAKER_00: It's like you can't do anything to AI.
- 6:49SPEAKER_00: You can't so much as tax a data center
- 6:52SPEAKER_00: because they will tell you you're slowing down AI and we'll lose the race with China.
- 6:57SPEAKER_00: And then B, look at the magnitude of this aspiration.
- 7:00SPEAKER_00: They want the US to dominate the 21st century.
- 7:04SPEAKER_00: This is really a common view,
- 7:07SPEAKER_00: fairly common in the tech business,
- 7:09SPEAKER_00: more common than I'd like even in the so-called AI safety community.
- 7:13SPEAKER_00: So yeah,
- 7:13SPEAKER_00: so China,
- 7:14SPEAKER_00: for starters,
- 7:15SPEAKER_00: the tension with China gives the industry too powerful a talking point.
- 7:21SPEAKER_02: You mentioned earlier the threat of a bioweapon.
- 7:25SPEAKER_02: What are your biggest fears about the dangers of AI when it comes to its threat to humanity?
- 7:32SPEAKER_02: What do you think are the best solutions for countering that?
- 7:36SPEAKER_00: I mean, first of all,
- 7:37SPEAKER_00: there are these issues that can only be addressed through international governance,
- 7:42SPEAKER_00: like a bioweapon.
- 7:43SPEAKER_00: If anybody anywhere in any country builds a bioweapon with new properties,
- 7:48SPEAKER_00: worse than COVID, you know, the pandemic is going to reach you.
- 7:52SPEAKER_00: wherever you are.
- 7:53SPEAKER_00: So national policy alone will not solve that.
- 7:55SPEAKER_00: There are other risks like this that are transnational in principle having to do
- 7:58SPEAKER_00: with hacking and so on.
- 8:00SPEAKER_00: But honestly,
- 8:01SPEAKER_00: what I worry most about or the concern that I'm most confident will take shape
- 8:08SPEAKER_00: if we don't focus and start to like slow things down and act more wisely
- 8:14SPEAKER_00: Is the sheer magnitude of disruption on multiple fronts, for example, jobs?
- 8:21SPEAKER_00: I mean, even if you accept that the people who lose their jobs will find new jobs,
- 8:25SPEAKER_00: and it's far from clear that that's going to be the case this time around.
- 8:30SPEAKER_00: It is still a very disruptive force in people's lives to lose a job,
- 8:35SPEAKER_00: have to look for another one.
- 8:38SPEAKER_00: And then there are these various other fronts where I think things are going
- 8:41SPEAKER_00: to get weird, including family relations, interpersonal relations.
- 8:46SPEAKER_00: I'm not saying it's going to necessarily be net negative.
- 8:49SPEAKER_00: I'm saying the change is going to be so abrupt
- 8:52SPEAKER_00: that we're going to have trouble absorbing it if things keep moving as fast as they have.
- 9:00SPEAKER_00: So some people have advocated a global pause.
- 9:02SPEAKER_00: Even Anthropic kind of threw out the idea that someday we may need one,
- 9:06SPEAKER_00: but that's another thing you can only do on an international basis, right?
- 9:11SPEAKER_00: The politics of the thing are never going to permit the U.S. companies to be compelled
- 9:19SPEAKER_00: to you know pause on big training runs these huge training runs
- 9:23SPEAKER_00: unless there's some assurance that china and other countries are doing
- 9:26SPEAKER_02: the same thing you studied uh the human mind for a long time one
- 9:30SPEAKER_02: of your previous books is called why buddhism is true at
- 9:33SPEAKER_02: a cognitive level what are your concerns about the impact of ai on the human mind
- 9:40SPEAKER_00: that it will reinforce some of our our worst
- 9:45SPEAKER_00: Tendencies and the tendencies that most underlie mindless human conflict,
- 9:53SPEAKER_00: including one I kind of alluded to when I talked about the importance
- 9:57SPEAKER_00: of understanding the perspective
- 9:59SPEAKER_00: I call that cognitive empathy some other people do too
- 10:04SPEAKER_00: or you can call it perspective taking whatever but there are cognitive biases
- 10:09SPEAKER_00: that just impede that like once you've got somebody in the enemy category
- 10:13SPEAKER_00: then there are mechanisms
- 10:15SPEAKER_00: We Know About That Kind Of Distort Your View Of What's Motivating Them I Mean
- 10:21SPEAKER_00: The Most Common Source Of War In The World I Would Say Is For Each Side
- 10:27SPEAKER_00: To View Anything Defensive The
- 10:29SPEAKER_00: The other side does as offensive.
- 10:31SPEAKER_00: This is a famous dynamic underlying World War I.
- 10:36SPEAKER_00: You know,
- 10:37SPEAKER_00: and if I say to somebody,
- 10:39SPEAKER_00: you know,
- 10:40SPEAKER_00: from Iran's point of view,
- 10:41SPEAKER_00: one of the main functions of Hezbollah and Hamas is deterrence against attack.
- 10:47SPEAKER_00: If I'm talking to somebody who's really strongly pro-Israel,
- 10:50SPEAKER_00: it's like the first time they've heard it.
- 10:52SPEAKER_00: And, you know, and there are things on the other side.
- 10:56SPEAKER_00: I think there's an under appreciation of how many people in Israel
- 11:00SPEAKER_00: as hard as it may be to believe think that like attacking Iran is like a defensive thing.
- 11:05SPEAKER_00: I mean, it's just common to
- 11:09SPEAKER_00: for people not to understand this.
- 11:11SPEAKER_00: It gets us into trouble all the time on foreign policy,
- 11:15SPEAKER_00: like, and helped get us into the war with Iran and has helped start many other wars.
- 11:22SPEAKER_00: So I think that is a, you know, the cognitive biases that impede that, just cognitive empathy.
- 11:31SPEAKER_00: And I'm not talking about like feeling their pain,
- 11:33SPEAKER_00: that can be good too, but I'm just talking about understanding their perspective.
- 11:38SPEAKER_00: That is the challenge I think we most need to overcome I mean,
- 11:43SPEAKER_02: how about at the workings of just processing information and coming up
- 11:48SPEAKER_02: with independent thoughts?
- 11:49SPEAKER_02: I mean, you've taught,
- 11:52SPEAKER_02: you're a teacher,
- 11:53SPEAKER_02: you've worked with many students Are you concerned in that domain
- 11:58SPEAKER_02: that people will be just relying so much on these machines and that will Well,
- 12:02SPEAKER_00: oh yeah, I didn't answer your question, thank you It's a good thing you're here
- 12:07SPEAKER_00: So the cognitive bias I'm talking about is part of I would say kind of a suite
- 12:16SPEAKER_00: of cognitive biases that together convince us that we're in the right and
- 12:21SPEAKER_00: the other person's in the wrong whether it's like an argument within our marriage
- 12:25SPEAKER_00: or like
- 12:26SPEAKER_00: and impending war and AI kind of famously right now has this tendency
- 12:32SPEAKER_00: to reaffirm your view of things because you know these companies they optimize
- 12:35SPEAKER_00: for engagement just the way they did with the social media algorithms they want you
- 12:39SPEAKER_00: to spend more time with the machine well
- 12:41SPEAKER_00: Well, you know, how do you do that?
- 12:42SPEAKER_00: You tell people things they want to hear.
- 12:44SPEAKER_00: What do we want to hear?
- 12:44SPEAKER_00: We want to hear that we're right and the other guy's not,
- 12:47SPEAKER_00: the other team's not, and we're asking brilliant questions and so on.
- 12:52SPEAKER_00: This is known, of course, as the sycophantic tendency of AI.
- 12:56SPEAKER_00: and I worry about that now it's not an inherent property of AI if you know
- 13:01SPEAKER_00: if we all like you might say choose our AI companions mindfully we can we can choose
- 13:11SPEAKER_00: Choose AIs that are not so sycophantic and maybe even go out of their way to explain
- 13:16SPEAKER_00: to us the perspective of other people and things like that and help compensate
- 13:22SPEAKER_00: for confirmation bias and bring us in touch with the arguments being presented by
- 13:27SPEAKER_00: the other side and help us understand that the most extreme person
- 13:36SPEAKER_00: in the adversarial group,
- 13:38SPEAKER_00: whether ideological or national, is probably not typical of that national group.
- 13:45SPEAKER_00: It's just the person that the social media algorithm is naturally choosing to show you.
- 13:50SPEAKER_00: You could have AIs that systematically help us see more clearly.
- 13:54SPEAKER_00: You really could.
- 13:55SPEAKER_00: And I think that's, we need like
- 14:01SPEAKER_00: There are various ways that this can happen.
- 14:03SPEAKER_00: I would like to see groups form that like,
- 14:05SPEAKER_00: you know, say they want this, you know, like some kind of, you know, mass movements.
- 14:11SPEAKER_00: I suspect as a matter of course,
- 14:13SPEAKER_00: some religious groups will start saying,
- 14:15SPEAKER_00: yeah, we give this one our seal of approval.
- 14:18SPEAKER_00: We have, you know, we encourage them to modify this one to this effect.
- 14:22SPEAKER_00: Now that may or may not be good.
- 14:23SPEAKER_00: It may be that a given religion just wants an AI that says you guys are right
- 14:27SPEAKER_00: and everyone else is wrong.
- 14:28SPEAKER_00: That's not really what I'm hoping for.
- 14:30SPEAKER_00: In principle,
- 14:33SPEAKER_00: these machines can help,
- 14:35SPEAKER_00: you know,
- 14:35SPEAKER_00: enlighten us,
- 14:35SPEAKER_00: but it's not what market forces will naturally do unless there is concerted demand
- 14:42SPEAKER_00: for that.
- 14:43SPEAKER_02: And going back to your previous book,
- 14:45SPEAKER_02: Why Buddhism is True, I mean,
- 14:47SPEAKER_02: you've documented the benefits of mindfulness on the brain, on the human body.
- 14:54SPEAKER_02: Is there a threat from AI to
- 15:01SPEAKER_02: It's an interesting question,
- 15:05SPEAKER_00: because it changes from model to model and generation to generation, right?
- 15:09SPEAKER_00: I mean, even the level of sycophancy I just described, that changes.
- 15:13SPEAKER_00: Like, OpenAI, there was a famously sycophantic model maybe
- 15:19SPEAKER_00: 4.0 or something.
- 15:20SPEAKER_00: They next time around rolled back.
- 15:22SPEAKER_00: They actually took that off the menu or something and just rolled back the sycovancy.
- 15:26SPEAKER_00: People complained.
- 15:27SPEAKER_00: They wanted an AI that complimented them.
- 15:31SPEAKER_00: And I think probably open AI obliged.
- 15:34SPEAKER_00: But yeah, I mean, if you view mindfulness as, you know, in part the ability to step back
- 15:43SPEAKER_00: and observe the world without your emotional buttons being so readily pushed and
- 15:49SPEAKER_00: without emotions guiding your thoughts.
- 15:52SPEAKER_00: You know, like even when you just click
- 15:55SPEAKER_00: You know,
- 15:55SPEAKER_00: if you see a tweet or something that's from your team and you're going
- 16:00SPEAKER_00: to click retweet without even like reading the underlying article
- 16:03SPEAKER_00: that you're actually spreading,
- 16:05SPEAKER_00: if you pay close attention, like that's a feeling, right?
- 16:08SPEAKER_00: It feels good.
- 16:09SPEAKER_00: It's gratifying.
- 16:10SPEAKER_00: And the same thing is if you see like a tweet from the opposing team and you just start
- 16:18SPEAKER_00: You reject it or whatever, that's a feeling.
- 16:23SPEAKER_00: Mindfulness involves an ability to step back and appraise things,
- 16:29SPEAKER_00: I think, calmly and judiciously.
- 16:34SPEAKER_00: and leads to clearer vision.
- 16:36SPEAKER_00: And I think it's good for you and it's good for the world.
- 16:38SPEAKER_00: That's the argument I made.
- 16:39SPEAKER_00: Well, I wasn't the first, but that's what a lot of people would say about mindfulness.
- 16:46SPEAKER_00: And I said it in the Buddhism book.
- 16:50SPEAKER_00: And yeah,
- 16:50SPEAKER_00: you know, again,
- 16:51SPEAKER_00: AI is to a disturbing extent by its nature,
- 16:56SPEAKER_00: a logical continuation of the social media dynamic.
- 17:00SPEAKER_00: because the default thing is to optimize for engagement and all of us when
- 17:04SPEAKER_00: in unreflective mode which we all spend time
- 17:07SPEAKER_00: in certainly including me you know we will
- 17:13SPEAKER_00: then you know give mindless you might say reinforcement to the algorithm
- 17:19SPEAKER_00: and let it guide us by pushing our buttons and
- 17:24SPEAKER_00: and reinforcing our own self-serving worldview about our rightness and the rightness
- 17:29SPEAKER_00: of Team America or whatever team you're thinking about.
- 17:33SPEAKER_02: Human beings,
- 17:34SPEAKER_02: especially in this capitalistic society,
- 17:35SPEAKER_02: have a tendency to, like a desire to acquire possessions, acquire power.
- 17:43SPEAKER_02: Does the AI system have that same tendency?
- 17:49SPEAKER_01: And to hear the rest of the interview, please go to UsefulIdiotsPodcast.com.