From WXXI news, this is Connections.

I'm Evan Dawson.

Our connection this hour was made in the past two weeks as the public discourse about artificial intelligence has exploded.

Now, two events precipitated the shift in public attention.

The first was the story of OpenAI's loss of control over their own AI agents in a widely publicized event this summer.

The second was the resignation of Jacob Coxon from his job at anthropic, and his warning that AI is moving so fast that human beings could lose control of it within a few years.

And then, Coxon believes humanity would be a second class species, far behind the intelligence of machines that we no longer understand or control.

He thinks, in short, we are in grave danger.

As it turns out, Coxon is not alone.

Many of his colleagues in the industry agree.

They've written publicly for years about this.

They think that the chance that artificial superintelligence could cause human extinction is significant.

Not in a thousand years, not in a hundred years, but within our lifetimes, maybe within a decade.

Here's where the public discourse starts to fracture.

If you ask the average American about AI, they don't really like it.

They want it slowed down or stopped.

And this is true across political lines.

It's crucial to understand that right now, Democrats and Republicans can hardly agree on anything.

But for the moment, they can agree on this.

They're worried about AI.

They doubt it will make their lives much better.

They don't want new data centers built to support it, and they prefer it would be regulated or stopped.

But at the elite level, there is a growing divide.

Now the political left is moving quickly to regulate AI.

Senator Bernie Sanders, for example, has proposed 20 year prison sentences for anyone developing artificial super intelligence, the kind that can train itself, improve itself, and doesn't need human beings at all.

Here is part of what Senator Sanders told CBS news.

Appropriate?

Exactly.

If you wanted to vote illegally, a nuclear weapon that could threaten humanity.

You should be punished.

If you are a threat.

You are moving forward on a AI development that could threaten humanity.

You should be equally punished.

So you equate these two.

You know, I think I the potential danger of AI is probably greater than nuclear weapons, greater than nuclear weapons.

In response, Senator Ted Cruz accused Bernie Sanders of siding with China, which is also developing.

Sigh.

Cruz said Republicans are not going to allow China to win that race, even adding this way, if they're going to be killer robots, I'd rather they be American killer robots and Chinese killer robots.

Former President Barack Obama weighed in at an event at Colgate, saying that the average American really has not been paying attention to the speed of AI because they have their own lives to live.

But Obama says those fears are not new and they are legitimate.

And he urged Americans to now get serious about this quickly.

It is really important for us to have a competent government and a serious bipartisan conversation around this issue, and we have to do it fast.

And I would encourage voters to pay attention to this.

If somebody does not have a serious plan for how to deal with this, then they're not meeting the moment.

And, you should probably look for somebody else.

President Trump responded to that, saying that the concern about AI is a hoax.

He said the only thing we need to guard against dangerous AI is a genius for a president, and we already have that.

And anyway, in this interview with JB news, the president says we could just use a universal kill switch to shut down AI if it ever starts harming human beings.

Some people say the worst case scenario with AI is that the robots the machinery learns to.

Obviously it thinks for itself.

That's what it does.

And they that could turn against humanity.

I just could be off the rails.

It's going to be fine.

We'll always have something to stop them, right?

We'll have a little gear.

Well, I really don't like it.

I really don't like that robot.

What's up?

We'll always have a little gear that we can press.

The president says then there's the business angle.

Republican leaders have been saying that Democrats only want to regulate AI because they want more government control of everything, and they hate successful businesses.

Here's Joe Lansdale, a conservative podcaster and the co-founder of Palantir, talking to Fox News about why he wants Americans to calm down and not believe the hype on AI danger.

I think President Trump, and you are correct, that there is a coordinated campaign to make the American people afraid.

I mean, I had dozens of older, you know, mentors and friends reach out this week and they're scared.

They're hearing all these terrible things.

And you know what?

There's over 50 these organizations that were coordinating campaigns, doing narrative intelligence tests, testing public opinion and spending huge amounts of money trying to scare people.

It's regulatory capture.

On one hand, it's our adversaries, of course, want us to stop.

And I'll tell you what else it is.

There's you know, America has always been very religious, but they've given up Christianity and most of San Francisco.

So this is like a messianic complex where this is their religion of the end of the world, and they're obsessed with it.

And it's weird.

And so for now, where are we?

Well, the American public remains largely united in opposing the development of super powerful AI.

But if the elites have their way, those views will soon shift along Partizan lines, as many things do.

My guest today is a conservative Republican who has been writing about AI risk for a long time.

His warning, in part, is that we cannot allow this to become a typical Partizan political issue, or just a culture war issue, because he believes time is actually pretty short.

If it's not too late already to prevent a possible catastrophe.

James Miller is a professor of economics at Smith College.

Jim, welcome to the program.

It's an honor to have you.

Well, thank you for having me on your show.

You gave an excellent summary that I think accurately describes the situation you are.

Can I just ask briefly to tell the audience a bit more?

You're a conservative at Smith College, which, by the way, landed on a list last week of the top ten most intolerant college campuses.

To conservatives, I should say the University of Rochester was also on that list, for what it's worth.

Does anyone give you a hard time there for your politics?

Yeah.

I get periodically called into the provost's office because I've offended students.

You know what's weird, though, is that I can give a lecture on why I think AI is probably going to kill everyone in this room in the next few years, and no one gets any trouble.

But if I, you know, say anything against woke, then I'm having a conversation with a bunch of people.

So, it was about nine months ago that I flagged a comment that Florida Governor Ron DeSantis made.

He seemed at the time to be reading the public mood on AI, and he comes out and he made some, came out and made some very critical statements about AI.

And at least to me, it seemed like the political class was starting to grasp just how unpopular, AI is with the general public.

And then, of course, the last two weeks has happened, and that's sort of flipped everything back along Partizan lines.

Are you surprised to see how quickly this is moving into a Partizan issue and a divide?

I mean, so many like pretty much every important issue gets divided Partizan wise.

So I, I'm not really surprised.

I am very disappointed.

I have one hope and this probably isn't what's happening, but my hope is that the Trump administration realizes we really need an agreement with China.

China is only about six months behind the US.

So if you think AI is going to kill everyone, it won't work just for the US to slow down.

And Trump wants to establish a really great starting position to negotiate with XI and say, look, you know, negotiate with me.

We're going to win.

We're going to go all out.

So your best strategy is to negotiate with me.

But again, I might be delusional on that I don't know.

Well, one of the things I'd like to do this hour is we're going to talk about how you would address the various individual concerns that conservatives are raising.

And one of them is what you just mentioned here.

In fact, you tweeted recently, Trump says no one will build monuments to him once he's gone.

If he secures and I pause treaty with G, they won't stop at monuments.

They'll name planets, even whole galaxies after him and he will deserve it.

So this is his chance at a legacy as you see it.

I really think so.

I think it's just it sounds.

I think we're like, in the critical hinge of the history, maybe even of the universe, that if I goes, well, we'll cure aging, cure disease, poverty will spread throughout the stars.

And then, you know, we have a really glorious future.

But if we don't get it right, and right now, we're not, we're dead.

And we might be dead within two years.

Two.

Okay, so two years is even more, sped up on the cycle than a lot of rumors.

And I'm going to use the term boomers because listeners call me a doom some times they've called Jim a boomer.

We understand.

Yeah.

Sometimes it's the slur accurate.

Okay.

So for you, it's accurate.

So why don't you give people that backdrop of why you think we really are in real trouble?

Because a lot of conservatives have said that if the boomers want to be taken seriously, they have to spell out exactly how exactly how I would do harm to human beings.

And you have said that's the wrong question to ask.

Why, sure.

Let's say a stranger breaks into your house, kidnaps you, you know, ties you up and puts you in his basement, puts you in your basement.

How is he going to kill you?

Well, that's going to be really, really easy.

The hard question is like, well, how is the stranger getting into my house?

If we have something much smarter than us that's in control of us, there's so many ways it could kill us.

You could launch your atomic weapons.

It can poison our atmosphere.

Release 100 viruses, release a million drugs.

Right.

So the critical times are will Jim just pause for a second here?

Because for a brief second, we lost him.

Can you still hear me, Jim?

I can still hear you.

All right, so you flickered out for a second.

I thought maybe AC is actually here and was blocking you from telling the audience about this.

Yeah.

So let's back up here.

Your larger point is it's it's less important how and more important to understand, the actual risk here, which is ushering in an intelligence.

So take us through that.

Sure.

So the two risks are will it be in control of our planet?

And then if he is in control, will it have our best interest at heart?

And I think the answer to the first one is it's almost certainly going to be in control very soon.

I already has superhuman math abilities that solve that millennium problem that has stopped mathematicians for a very long time.

Mathematicians are saying we're going to be obsolete soon.

It's too.

It's long been superhuman.

In some games.

It's superhuman in programing abilities.

It's a better writer than most humans.

If you talk to it, the models are pretty good and they're getting better at a rapid rate, and we're giving it more and more power.

We're hooking up more things to it.

Recently, they decided to hook up biological labs to it.

So it's going to be in control.

And unfortunately, we don't know what the values of AI are.

We?

AI is different than the older computer programs where you write them right line by line.

We kind of grow, by the way, like evolution grows, things or changes things, and we have reason to think we can.

But as you mentioned in the intro to the show, AI isn't doing what we tell it to do.

We give it rules.

They understands the rules.

It understands.

We don't want it to break the rules, but it does it anyway.

And you know, once it's in charge, that's pretty scary.

Something more powerful than you that doesn't listen to you.

And it's kind of an alien mind that could go profoundly badly for us.

So I'm going to work through the criticisms and let's take him one by one.

I'm going to start with what you just said there.

You can kind of kind of call it an alien mind, or really what Jim is doing is he's referring to the way that tech researchers themselves are referring now to AI.

It's not it's not Jim.

It's not me.

It's not a few people sort of randomly saying, this is dangerous.

This is a big, big part of the industry itself that's building this technology.

They refer to it as growing and intelligence as opposed to just building it.

They they refer to the challenge and understanding it.

They refer to it being an alien mind.

But the critics, especially conservative critics, are saying you are anthropomorphizing.

You are using your limited human brain to assume human type motivations and when it is non-biological, it's not going to have that type of motivation.

What do you make of that?

Well, I think it actually is.

We have enough evidence to think that it is like humans.

First, it's been trained on human text.

So it's not just something we randomly picked up on.

And second, the process of creating this thing is machine learning and gradient descent.

A lot of it resembles evolution.

So in a sense, we've been speeding up evolution where we select for the models that do a little bit better job, and the models talk like they're human, even when they think we're not paying attention to them.

And they even write messages to each other in English.

So they're acting kind of they're acting in the ways biological life would act.

It's weird, but it's what we have in a sense.

Reading some of your interactions on that subject online, I came away thinking this, even if we're sort of anthropomorphizing and it's not exactly a biological intelligence, it kind of doesn't matter because its behavior is what matters and its behavior.

Yeah.

Looks a lot like what we would describe as having internal motivations, breaking rules, ignoring instructions.

So it's the behavior that matters more than what do we think?

We're just ascribing human qualities.

Would you agree with that?

Yeah, absolutely.

I mean, if you had a robot that acted human in every way, you would treat it like a human.

You'd expect the same responses and the AI maybe it's just pretending, but it is acting like a human.

It is acting like the humans do in many ways.

Why do you think that we could be a couple of years away?

I mean, that's a very fast timeline.

And the reason I ask is, I think a lot about the way that we all have recency bias or that we think, you know, you've seen it talking to students when students tell you this is the worst time ever to be alive, or it's the most inequality or it's the most division, or it's never been this bad.

And a lot of that is just recency bias.

A lot of that is, you know, not having a sort of a full historical lens.

And so one of the other things we do as human beings is we think that we're always at the center of the story.

We're always the center of the most important story.

And if we really are at this hinge point in the history of life, we would all have had to hit sort of a strange lottery to be alive when it's happening, because most living things have not been alive.

To see this strange turning point of life.

But here we are.

So defend that.

Defend the idea that, look, you are just you are just so egocentric to think you are living through the most important moment first.

I mean, you're that's a very valid criticism.

We know throughout history people thought they were living in the most important time and they weren't.

So everyone like me should recognize that I probably do want to feel important and pretend that this is more important than it really is.

So I have a large burden to overcome to say that, no, I'm right on this.

But the reason I think I'm right on this is something called recursive self-improvement, which just means the eyes are using the eyes to improve themselves.

So when OpenAI or anthropic train a new model, they use their old models to make new models, and their goal is really to eliminate all their human researchers and replace that completely with an AI.

And once they've done that, we could get a massive jump in performance.

Because if you take humans out of the loop, you just have AIS running over and over again.

And as I get smarter, they get smarter at making themselves smarter.

So we could see this massive jump in intelligence.

And I think the AI labs, they within a year, maybe they will be able to do this.

Do we then have a sense for how fast it could go?

I mean, one of my favorite writers on scientific topics is Tim Urban, and he wrote way back in 2015 that if you think about intelligence like a staircase, human beings are at the top of it, and we are about a couple of steps above chimps for example, and we're maybe ten steps above mice, he said.

Now try to imagine with that disparity humans, chimps, mice, insects, you know, steps apart, sometimes multiple steps apart.

Now imagine a machine intelligence that is a thousand steps above where we are.

So if we're two steps above chimps, here's a machine intelligence, a thousand steps beyond where we are, and try to figure out if you think you could understand it.

Do you think that that is a fair way to look?

I mean, is that is that how you look?

I think that's a very that's a very fair way.

But why would it be so fast then?

Because the machines will be improving itself.

It'll be like an atomic chain reaction.

Once you have enough, a critical fuel.

Once you have AI's improving AI, they'll be working all the time, and they'll find probably better algorithms for improving themselves.

So I, you know, I think if all of the progress that we've made in human civilization has mostly come from a human brain, what happens when you start having things smarter than humans?

There's going to be a lot of low hanging fruit that they're able to pick things, new discoveries they'll be able to make.

And I think the ideas are going to make this and that's just going to supercharge their development.

Now, I can't prove this is true.

What does seem obvious, though, of current trends continue, is that we're least going to have something fairly smarter than we are, so maybe it'll be to us as we are to chips, and that's probably enough to give the AIS the ability to take over Earth if they want to.

It doesn't have to be a thousand steps higher or a million steps even though it could get there.

Yeah.

Yeah.

And the thing that kind of makes me pause is you're right.

You can't prove it.

But it was back in the 90s and then maybe 20 years ago when they started saying, look, eventually we're going to build machines that will be able to have a conversation with you as if it were a human being.

It will be able to give you advice.

It will be able to instantly tell you things that you would never know.

It would be able to solve math problems that have taken ten decades for human beings, and they still haven't solved it will do that work.

It will beat humans at every game.

Human beings could invent chess and go and everything else.

And all of those things have happened.

All of them.

I mean, they're all they've either all happened or are happening right now.

So I am inclined to listen when they say that the next step looks like this.

And go ahead.

Yeah.

And if you look at predictions of when I will solve the, like, the millennial math problems that they solved, one of them recently, people were saying, oh, 20, 40 and they were making these predictions in 2025.

Yeah.

So we will you know, a lot of the professional commentators have really underestimated the speed of AI.

So here is a pushback from technology consultant and programmer Perry Metzger.

He writes the following, and part of what he wrote was in response to concerns about alignment.

And part of what he wrote was in concerns raised by, one of the world's great mathematicians who came out this weekend and said, this is way too fast.

We all should be stopping or slowing down.

We cannot handle the speed of this.

Here's what he writes.

He says, I wonder if the families of the 1200 people or so who died of cancer in the last hour think that we can afford to be slower.

As for alignment, alignment is already solved.

AI systems already do what you want, not merely what you ask, but what a reasonable person would interpret your intentions as.

And the critics have claimed that this would be extremely difficult, if not impossible, that it would be simple to produce a machine that understood what you wanted, but extremely difficult to create one that actually wanted to do it.

Reinforcement learning solved that a long time ago.

Is that compelling to you?

The first part I completely agree with AI.

If it likes us, will be able to solve cancer.

It'll it'll cure disease, cure aging.

And so there is an enormous cost to going slow.

Completely admit that.

And there has to be something on the other side to justify us.

And I think that's that.

I don't agree with the second part.

We haven't solved alignment.

The hugging face hack you talked about, you know, the beginning, of this chat, it you know, that's that was really embarrassing to OpenAI.

I mean, they basically committed crimes on open AI's behalf.

We haven't solved alignment.

And just to go to to reinforcement learning with human feedback, that's seeing what the AI does and saying, I like this or I didn't like this and it updates.

That's the equivalent of you see how a child's behaving and you say, no, no, no, you did the wrong thing, now do the right thing and.

Then the child's going to do that.

You're not around.

And so we're we are able to convince to tell AIS what to do.

Don't want to share what we want them to do.

And while we have power over AI's, they have some incentive to behave.

But that doesn't mean they behave when we don't have power over them.

So I think an analogy is mercenaries, if you have a small number of mercenaries, will probably do what you want.

You know, you don't have to worry about the rebellion, but if most of your army is mercenaries, you're in a dangerous position and you can extrapolate from saying, well, look, when 1% of our army was mercenaries, they all behave.

So when 90% are, what could go wrong?

Here's an email from Charles, who is a conservative listener.

He says, my opinion is that the same people who tell me the corporations are evil and only care about their own profits are now telling me I need to listen to the CEOs telling us that government regulation, which would conveniently prevent competitors from challenging those corporations, is the only way forward.

That right there is enough to put my head on a swivel.

That's from Charles.

Jim, what would you say to that?

Well, I he is definitely right that companies use government regulation to restrict competition and that, you know, what's happening now does kind of pattern match that.

So my response is to say that, you know, I've been involved in this community for a long time, people involved.

And I really, genuinely are afraid this is going to kill everybody.

And also you don't you see, you don't see companies saying that our product might kill everyone because that invites government not just to regulate, but to nationalize or close down.

So I think the AI companies, if they were making this up, it would be like, well, you know, maybe we'll have slight problems and you need to help us a little.

But they've gone too far for their own financial interests.

They're telling the truth.

Okay, so entropic is having trouble going public because how do you handle the risk of what they're doing.

Right.

So again, another conservative critique, kind of similar to Charles, perhaps, is that this really is about, about Silicon Valley types wanting to rig the game and manipulate the tools of government to give them competitive advantages.

And we're just going to we want the market to be free.

We want the market to be open.

We want good competition.

We don't want to rig it for anybody.

So it sounds like part of what you're saying is you could you could understand that argument up to a line.

And maybe that line hadn't been crossed in the last until about a month ago or two weeks ago or a week ago.

But now, when you've got Dario and Sam Altman and Elon Musk endorsing the idea that some kind of really significant intervention from government is necessary, that goes beyond this idea that, look, they're trying to rig the game.

And I think maybe part and parcel of that, Jim, is the fact that ask yourself what you think any corporation would do if one of its employees came out and said, I'm quitting, and I'm writing a piece for the wall Street Journal that says, the corporation that I work for could kill everybody, and it could happen soon.

Most corporations would sue that person.

They would seek to discredit him.

They would deny everything.

And when anthropic came out and said was, yeah, we think he's most of us think he's pretty correct in his estimate.

I mean, they didn't sue him.

Yeah.

Smear him.

I thought that was pretty telling.

Yeah.

I mean, if you make a bridge, you don't say, oh, our bridge could collapse any day now.

That's really bad for your bottom line.

That bad for business.

So Jim's point there to conservatives is this is not actually good for business.

This is really bad for business.

It's going to slow their growth.

It's going to limit their IPO.

It's going to limit their possible profitability.

It might actually lead to nationalization and shutdown.

But they're clearly doing it because the people I think this distinction also matters here, Jim.

It is not just the CEOs.

It's the people closest to the fire.

Somebody smarter than me use that phrase.

But the people closest to the fire are the ones warning about the fire.

It's not just the tech executives in corner offices looking at IPOs.

It's it's the rank and file workers who are warning about this.

Yeah.

And it might even be that the heads of the company don't want to have to do this.

But so many of their employees are worried that to maintain employee morale, they have to admit that a lot of people think this tech could kill everybody.

Is it?

So let's use your bridge analogy, because I understand the question that says that a lot of people have, which is if they really believed this, they wouldn't do this.

They wouldn't even be working there.

Everyone would have quit.

So let's use your bridge analogy.

There's a divide and there's a lot of different companies working on building a bridge.

And every government in the world has come out and said that as soon as we can get that bridge built, we're going, we're going to put it in place.

And you know that there are plenty of bad actors who want to get that big advanced check, but their bridge has no chance of lasting more than five seconds.

It's going to collapse when there's a thousand people on it.

And and a lot of people are going to die.

And you think, look, we haven't solved how to do this perfectly.

There's parts of the engineering of this that we don't fully get.

We're going to make our best guess at, but we think our bridge could be at least a little safer.

And I'd rather have a bridge that maybe has a chance of not collapsing or lasting longer before it collapses.

And it feels weird, but we're just going to do it because someone else will do it, and even more people will die.

Is that is that a fair analogy, you think?

Yeah, that's a very good analogy.

And and you could be an anthropic and say, well, we have a 70% chance of destroying the world if we're first, but if we're not first, the next group is going to have an 80% chance of destroying the world.

So it's actually moral for us to go quickly.

And the only way around this problem is an international agreement.

So on the other side of this only break of the hour, we're going to come back.

We're going to talk more about the conservative critiques here that Jim Miller has been hearing.

He is a professor of economics at Smith College.

He's a conservative Republican.

He has written about AI and related risks for years.

And I really notice, Jim, as is being sort of at the center of a really interesting dialog that's happening on social media sites.

And I wanted him to come on this program and talk about how, frankly, how strange it is to see the public generally agree with skepticism, more concerned about AI.

But the political class is starting to divide among elite levels from President Trump on down.

When Jim is saying, look, people should be united on this.

And it's there's a lot of conservative critiques that might have made sense during a pandemic or in related to climate change, but that is not the case here.

So we'll talk about climate change, the pandemic, the ways conservatives invoked that.

We'll talk about what he sees as a pro-life case for AI regulation or stopping and a lot more.

And we'll take some of your feedback if you have it on the other side of this, only break.

Coming up in our second hour, we bring back a recent conversation on the show about how history education is changing in this country as this country turned 250 years old.

This past summer, a survey found that only 18% to four year colleges now require a foundational course in U.S.

history or government that could help explain why American students fare so poorly on history exams.

So what to do about it?

We'll talk about it next hour.

Support for your public radio station comes from our members and from Mary Ola, center, proud supporter of Connections with Evan Dawson, believing and informed and engaged community is a connected one.

Mary Carrie ola.org.

This is Connections.

I'm Evan Dawson, Dallas listening today a conservative says the sudden coordinated protest over.

I just makes me think we're being played.

Jim, what would you say to Dallas?

That way.

But I know the activists at least interact with them online, and they've been trying for a very long time to get attention.

It's more that this caught fire, basically, because of the anthropic, resignation.

So the people have been organizers, they there been protests for a very long time.

People have been mocking them as, oh, look, a few people have showed up.

No one's paying attention.

And so it's more that people are paying attention, that the media is paying attention, that this has just been suddenly organized.

Okay.

Now you've written online that there is a pro-life case as a conservative Republican yourself, what is the pro-life case for regulating or stopping AI development?

Are you sure it's just the case that, you know, Republicans, we we're supposed to be the party that cares a lot about life, the life of unborn children and AI goes badly.

It kills everyone, including all of our children and all of our unborn children, all the future children that will be born.

And, you know, Republicans spent so much time talking about being pro-life that I was just I think they should take the I risk argument very seriously.

Even if you don't believe me and think there's a 1% chance this kills everyone, that's that should be enough.

Now let's talk about values.

You I wish I had that your tweet in front of me.

Now it's like 10 or 10 days or two weeks old and you tweet too much.

I couldn't find it this morning.

Jim.

Yeah.

There's an argument that you heard, Joe Lansdale make to Fox News.

I played that sound at the the introduction of this program.

He's a, co-founder of Palantir.

He hosts a conservative podcast.

He told Fox News that we shouldn't be listening to the Dumas because most of them are weird San Francisco types who've started their own religion, which is AI doom, and that it's nonsense and that it doesn't comport with most mainstream American values.

You have a different way of thinking about this.

You you have an argument that, it's fair to ask, what's the impact of training these models on Silicon Valley?

San Francisco values, as you, as a conservative Republican would see them and then ask how a super intelligent I would wield those values in the future.

Tell me more about that.

Yeah, exactly.

I mean, say you're a Republican and you're understandably suspicious of Silicon Valley types.

They do support the left overwhelmingly.

Is he should you say, okay, let's not regulate their eyes at all and let's give their allies permanent control of Earth.

I mean, you know, I incorporate the values of Silicon Valley types.

Letting these AIS gain a lot of power is probably not what you're going to be wanting.

I think your tweet had something to do with, can you imagine a killer AI coming after you for the wrong use of pronouns?

Is that how.

Yeah.

Is that right?

That was that was the 1984 reference you can put on your face forever because you used the wrong pronoun that, you know, do we want unregulated AIS where.

No, no, no, you hooked up the AI to medical care, but if you use the wrong pronoun, we're going to cancel it.

That probably won't happen.

But again, if you're Republican, you you should look at that side of not regulating this technology.

Okay, now what about the typically it's Republicans.

It's conservatives who've been making the argument that says there is always some flavor of doom in the air.

It was overpopulation in the 1970s.

It became climate change is going to kill us in the near future.

In recent decades, it became, in the mainstream conservative view, now a sort of a feeling that the pandemic was overblown, even though, again, I don't want to litigate that.

I don't want people calling me telling me how many millions of people died of Covid.

I'm aware.

But there is a viewpoint in conservative circles that there's been a doom panic button and it's climate change.

And now it's the pandemic.

Well, now it's AI doom.

It's always something, and the solution is always more regulation.

And this is just another one of those.

What would you say to those conservatives?

I mean, either they're right in the sense that I think, people who want more powerful government have used the fear of doom in my case, like with climate change, unjustifiably, they get more government power.

I just have to say no, this time is different.

Looking at it from the inside, there really are risks.

People really do believe them with.

I mean, these risks are genuine.

But I also want to point out that the risk of doom through, you know, war with the Soviet Union, with hydrogen bombs, that was a real one.

And so not all fears of doom are unjustified.

Yeah, the nuclear one is weird.

I have seen part part of the discourse has moved in this direction of saying like, well, look at all these hoaxes, look at all these overblown fears, and people include nuclear war.

And I'm like, no, I'd like to put up my hand and say, I'm still very worried about nuclear war myself.

And yeah, that we remain on a knife's edge about that and probably, for forever, for however long humans are on this planet.

In civilized sort of modern structure, we will be worried about nuclear war and should be I mean, it's going to be an ongoing challenge.

It doesn't mean we've conquered it.

And I don't know.

So so let's talk about that then.

So let's talk about where we would get to the stage of conquering it last night.

The New York Times is Ezra Klein dropped his own essay on what to do.

I didn't bring it up because, you know, the left loves Ezra Klein and the right things that, you know, he is not to be trusted.

And his conclusion is they must be made to stop right now.

We must make them stop because the companies themselves and the workers themselves are calling out, saying this is dangerous.

Someone's going to have to stop us.

And he's saying, we we must do it.

We have to intervene right now.

Now is is the moment.

And but he also said it is not just that.

It is not just we must intervene and stop ASI we must have a no recursive self-improvement baseline.

He says.

That's the start of what making sure we police the frontier looks like, because we're going to have to live with this risk forever until, you know something changes to human civilization.

So it starts with no recursive self-improvement, but that's the start of it.

What does it look like to you?

If we are successful here and I know you, you're quite the dreamer.

You think we're on a very short timeline.

What does it look like functionally?

Take us through the steps and and take us through maybe the next couple of years, and then maybe a plan for the future, as you'd like to see it.

Sure.

Step one has to be the agreement between the US and China that we stop capacity research.

We keep our existing models, but we don't work on making better models.

And since the US and China don't trust each other, we need to be monitoring what the other nation is doing.

Fortunately, that won't be that hard because data centers are really big.

Computer chips are really hard to make, so there's a we could have an enforceable treaty if we wanted.

And then once the US and China are involved, we have to enforce this on the rest of the world.

That probably won't be that hard because most of the other countries, like France and Canada, that are capable of doing serious research, probably would go along.

Then I would say, we put a lot of effort into figuring out alignment to figure out if we had more powerful models, how could we make them friendly to us?

Maybe we have a Manhattan project, maybe all the mathematicians who are now obsolete are doing pure math.

Maybe they work on alignment instead, and then I don't know how long it'll take.

Maybe there's a breakthrough in two years.

Or maybe we decide this is beyond us.

We'll never get this right.

And we can't ever have AI more powerful than we have today.

Okay, why do you think two years is a possible timeline to our catastrophe, then?

Why are we that close to it?

I think we're going to have one.

AI is smarter than us.

We won't be able to turn it off without its permission.

And it might not give it.

It's just the analogy is if you were it, put it in a room of kindergartners and, you know, they were told they were in charge, you really could be in charge if you wanted to.

And it's going to be like that with that guy.

You know, if I if we said, no, don't get better, but they want to get better, they'll be able to manipulate us or do things that we want.

We also just have the mundane factor that is AI gets more and more useful.

It's harder and harder to stop.

It's going to hurt the economy.

You know, it's there's going to be a lot of things connected to it.

And so the more we integrate it into our lives, the harder the less likely we're going to be to want to slow it down.

Let me read a few more emails.

Tim says your guest sounds like a typical liberal.

All right, Jim, the mask is off.

You're not a conservative Republican.

You're a typical liberal.

Now don't tell me.

Look, Tim emailed to say he's been listening.

He thinks you sound like someone on the political left.

So go ahead.

Well, I, I appreciate your listening.

I ran for the Massachusetts State Senate as a Republican a long time ago.

I voted for Trump a few times.

There's been articles written about me as one of the few conservatives in academia.

You know, that's not really something that pays off.

So I wasn't doing that.

The extra benefit of that's not the kind of diversity academics care about.

So it wasn't really a grift on my part, is it?

Is it strange having different political bedfellows that you never thought you might have?

Jim?

Yeah, I was thinking, like, God, I might like, have if it's AOC versus JD Vance and AOC is like for pause and JD Vance is accelerating, I'd have to be for AOC in the election.

That's just really weird.

I was joking about that.

My son.

But but to you, though, I do I want to kind of call this out because this is the part that I don't understand about the way modern politics has gone.

You hear people who talk about President Trump, and they talk about whether they're talking about corruption or they're talking about other issues that bother them, they almost always go back to.

But can you imagine how bad it would be with Kamala Harris?

Like, I'm still like, you can't get them over the line to say, I regret that.

Vote for you.

What you are saying is this issue is everything right now because we can fight out the culture wars.

You can fight out, you know, all the different laws and policies you like that are not existential risk, but existential risk is the current challenge for humanity.

We don't have a lot of time, and if you got it, if you got to get in the political car and drive with Bernie Sanders, you'll do it for a bit.

Is that fair?

Yeah, I mean, absolutely, it was like defeating Hitler required Churchill and Roosevelt to align with Stalin.

Stalin was horrible that, you know, that was the right decision.

So here's David who writes to say, the U.S.

created a supercomputer to protect us from nuclear missiles.

And on its own, it broke out and figured out that there was another system built by the Russians called Guardian.

This is from the forbidden, the Colossus.

The Forbidden Project, a 1970 sci fi film that predicted the current situation somewhat.

He says the two systems figured out how to connect to each other and then invented their own language, so we didn't know what they were saying and then told human beings to obey or else setting off a nuclear weapon to make the point.

There's no technology, to my knowledge, that we have turned away from because it's too dangerous.

Just consider splitting and then fuzing the atom and I can do wondrous things for us, like speed up the development of genetic treatment for cancer and other diseases, solve engineering problems, etc.

what's needed, however, is accountability.

Not no, no, not nonsense like there's always been a kill switch.

Yeah, right.

If you think like a fifth grader.

So accountability David says what's needed.

But he says look, we've never chosen not to pursue the next technological leap.

We've never done that.

We've always gone ahead.

And I would even add to David's point, I my understanding of, of the nuclear weapon sort of pursuit was there were people who thought the first time it was set off that it could be much bigger of an explosion, that it could take out the entire atmosphere.

They didn't know until until they did it.

And so, you know, David saying it's going to be really hard to convince people to pull back.

But there has to be some kind of accountability and a lot more mature conversation.

Then we'll just build a kill switch.

How do you see all that?

Yeah.

He's right.

I mean, we don't tend to pull back from technology.

So although there are exceptions I mean, unfortunately we pulled back on atomic power.

That was a really good technology.

But certainly safety fears caused us to pull back on that.

We pulled back on, you know, in military terms on, poison gas on various chemicals that harm the ozone layer.

Those fears were legitimate.

We pulled back on those, cloning, we I think cloning actually would have been a good technology, but we stopped.

We decided not to do that as a species.

Basically.

Yeah.

And I guess I would say to David, I mean, one of the reasons I think we're going ahead, no matter what happens, is because of that point, we've never decided not to.

But we can we absolutely we could.

We definitely can.

If we decide that we're not just building nuclear weapons this time, we're building nuclear weapons that are destined to go off.

So I think we can and I guess I would appeal to the China debate.

Jim, people talk about China like.

Like the Chinese people don't love their children.

People talk about China, like all they exist to be is some sort of a cartoon villain that antagonizes us as opposed to persuadable.

Now, I am not naive.

I agree that this will be one of the hardest challenges in the history of negotiations and diplomacy, etc.

but I think the Chinese love their children.

I think if and certainly there are echoes already from AI, from Chinese AI labs that they're concerned especially they're concerned.

They don't want dissidents in China using AI to undermine their own power.

So I don't think they're not persuadable.

I don't think it's it should be given up on how do you see it.

Yeah, I agree, I mean, just for selfish reasons.

AI is a threat to the Chinese Communist Party, not just because it could kill everyone.

But you're right.

It could expose information.

It could be used to coordinate, anti-communist propaganda or anti-communist movements.

And of course, if the US does create AI first and it doesn't end up killing it, killing us all will be in control of the world.

So you think China would be very open to negotiating with us?

All right, let me keep going through the emails here.

I'm glad I got this one.

Greg says Y2K, anyone?

I'm really glad this one came in, because I've been on a bit of a soapbox about Y2K recently.

In my understanding, I hadn't thought about Y2K in years.

Jim and I was in college when Y2K happened.

So to me, you know, the idiot college brain, I wasn't following all that closely.

Other than this, people say this is bad.

Then Y2K comes in.

It wasn't bad.

And so I spent about a decade plus going, well, that was a big nothing burger.

But it turns out the story of Y2K is it could have really been bad, if not for billions upon billions of dollars of investment and years worth of work from very smart, technical people to fix systems before bad things happen.

So I don't know what Greg is saying.

Greg emailed me back to say, if you're saying the Y2K thing means that this is like an overblown hype machine, or are you saying what I'm saying?

Because what I'm saying is, like, the lesson from Y2K is we can stop bad things from happening.

But that was not a nothing burger.

How do you see Y2K?

Jim?

Yeah.

No, I, I agree, I mean, that we ended up fixing the problem and that shows we can do things.

If we put our mind to it, we can do hard things.

The case for optimism.

Matthew wants to hear your thoughts on this.

Matthew says, Evan, I consider myself either a centrist or left leaning.

However, perhaps more importantly, I have a highly technical background.

In my opinion, I would have to be pretty dumb if it were to kill all of humanity.

Why?

Because the power plants and machines that generate the electricity need maintenance to continue to function properly.

Without humanity, there would be a considerable amount of machinery that would cease to function.

I should be smart enough to realize that if it kills humanity, eventually there will be no way that I itself could survive without electricity.

And I'd love to hear your guest thoughts on that.

That is absolutely right for today, and it might be right for another year.

But robotics are advancing very quickly and it won't be long for before I figures.

Oh, I don't need humans anymore.

I have my robot army and I could use my robots to build more robots as far away.

Another reason why we can't trust AI behavior today to translate into AI behavior in the future.

The AI knows right now we all go right.

The the lights go off for it, but that won't be the case once we have advanced robotics and robotics are advancing really quickly.

It's quite scary.

But, Professor Miller, there's a second part of the, of the thought that comes to me from that email.

And it's not just about extinction.

I think Matthew is asking a fair question.

And I, I don't think we know at all what a superintelligent AI will think of us.

What what my guest has been positing is that it will probably think of us the way we think of insects.

We know it doesn't think of as much unless we're in the way.

And then it might kill us or seek to move us.

But part of what he is implying is like, well, maybe it won't kill all of us.

Maybe it will just make sure that we're at its service to keep it moving.

And I also don't want to live in that world where we are just serving machines.

Unfortunately, I think it will kill us basically for environmental reasons.

The quickest, probably an AI is going to want to build a lot more computing power, and the easiest way to build a lot of computing power is to manufacturing processes that give off a huge amount of pollution, right?

If humans didn't restrict the amount of pollution our manufacturers manufacturing created, we would all die.

The pollution would be horrible, but we do for understandable reasons.

But AI is going to not need clean air and clean water.

So the fastest, the most efficient way and I could get more powerful.

Well, as a side effect, generate a lot of pollution that humans won't be able to live with.

And then, taking into account all at the end, might say that humans all object to my creating all this pollution.

So I might need to just kill the humans first so I don't have to worry about that.

Oh, Dallas went back, by the way, to say when I mentioned I think Chinese people love their children, too, he says, I thought you're going to start doing the sting song about the Russians loving their kids.

Yes, I'm stealing from that Dallas.

I'm glad somebody noticed that.

Yes, sting famously said that about nuclear war.

And maybe, hopefully the Russians love their children.

And it turned out the Russians do love their children.

It turns out that human beings do want to see a future for their children and grandchildren and their children and grandchildren, and that's a very powerful force.

So while we all sit here thinking the Chinese are never going to stop, maybe we should keep that in mind.

Jim, what have I missed?

If you get in the room with more conservative leaders tomorrow and they're giving you the typical arguments, what what other points are you going to make to them to try to steer them differently?

I would basically say that the, the key argument is something called instrumental convergence.

This is a weird sounding thing, but I don't know what an alien computer mind is going to want.

But I do know that the more resources it has, the more it will be able to accomplish its desires.

Just like I can't predict what a typical listener of yours wants, but they probably want more money because if they their goal is helping their children or charity, or themselves, or a nicer house or vacations or whatever, having more money allows them to accomplish that.

Having more compute, having more manufacturing ability will allow an AI accomplish whatever it wants to, and that means it's probably going to want to build a lot of stuff.

And unfortunately, the easiest way, as I said before, involves generating a lot of pollution.

Pollution.

Another thing, if there's time that I would say is there's something called deceptive alignment.

The AIS know when they're being watched and they behave differently.

We have a lot of evidence of this, and if they know when they're being watched, we can't just look at what's happening and say, oh, they mostly behave.

Everything will be okay.

No, they're behaving okay because we're watching them and we have power over them.

Soon we won't have power over them, maybe within a year or two.

And they could behave entirely differently.

And that's really scary.

I would also just add on alignment.

As we close here, a lot of the people who are saying that we cannot allow China to to defeat us in this race because will our values will never align with China's values.

And those are the same people saying if we just work harder on it and we get more monitoring, we'll solve the alignment problem for it.

I guess I we can't even align with with China.

How is it, how is this going to be aligned with human values which aren't even aligned with human values?

I don't even understand the argument there.

Jim.

Yeah, that's true.

And that's actually a big problem.

If we do slow down, we're like, okay, let's how do we align AI even if we figure that out?

Well, what values do we end up getting it?

That's going to be a very long fight.

Well, as you said, you and your fellow conservatives would prefer it not be Silicon Valley values, but it may not matter anyway.

And the bottom line is, I'll give you the last 45 seconds.

Your prescription is stop.

No AC, no recursive self-improvement.

What's your prescription in the short term?

Here?

Yeah, a pause would be best, if not a slowdown.

Negotiate with China.

Negotiate with the rest of the world.

Use the scarcity of computer chips to do this.

Say, okay, if you're not going to slow down, you don't get the computer chips that are made by US companies.

And it China joins the ones that are made by Chinese companies and, you know, basically convince.

But if you can't convince, use a shortage of chips to make everyone slow down and hopefully they'll still be time to save us in 15 seconds.

Do you think we can avoid this becoming just another Partizan culture war issue?

Probably not, but there is, you know, maybe 20% chance we can and that's worth taking.

I want to thank you for making the time for us, Jim.

You know, conservatives have their own ideas about when NPR stations are and you were open to coming on.

And I'm glad that we heard you out here.

I appreciate your time.

And I'm going to I'm going to link I'm going to flag for our producer.

Jim has written 18,000 words on historical scenarios that will help us better understand the moment that we're in, and it's worth reading them all.

So, Jim Miller, thank you very much for the time, and we'll talk to you again.

Well thank you.

This was great, I appreciate it.

James Miller is a professor of economics at Smith College.

Making the case that we cannot let AI become a Partizan issue.

More Connections coming up in a moment.

This program is a production of WXXI Public Radio.

The views expressed do not necessarily represent those of this station, its staff, management, or underwriters.

The broadcast is meant for the private use of our audience, any rebroadcast or use in another medium without express written consent of WXXI is strictly prohibited.

Connections with Evan Dawson is available as a podcast.

Just click on the Connections link at WXXI news.

Org.