Chris Hayes on thinking intelligently about AI
"The biggest threat is that it just makes everyone stupider, which obviously is a genuine worry."
PN is supported by paid subscribers. Become one ⬇️
Media commentary about artificial intelligence often veers into doomerism or grifty hype. Chris Hayes is trying for something more thoughtful.
Hayes — who you might know from his MS NOW show, exquisite Bluesky posts, and/or his “Why Is This Happening?” podcast — recently debuted a new podcast series called “The AI Endgame” that tackles some of the big questions around AI.
As new episodes drop each week, we connected with Hayes to pick his brain about AI and its societal impact, how he sees the rapidly evolving technology impacting his kids’ education, the ways AI is useful (and not useful) for his work, and much more.
“The information aspect, which I don’t think we’ve even really started to touch, is front of mind for me,” Hayes told us. “We have one very clear example, where it seems someone at xAI, possibly Elon Musk himself, added a new layer into the model about ‘kill the Boer’ and the persecution of white Afrikaners. All of a sudden, Grok was just spewing this stuff at every opportunity.”
“In the same way that a generation of people would go to Google if they were wondering who’s running for office or something like that, now people ask Anthropic or ChatGPT or Gemini. The power that comes from putting a finger on the scale of those models to manipulate the political information people consume is so insane to think about.”
A full transcript of the conversation between Hayes and Public Notice contributor Thor Benson, lightly edited for length and clarity, follows.
Thor Benson
I’ve seen you grappling with big questions surrounding AI on social media. When you decided to launch this podcast series, was it because you feel that you’re starting to get a handle on some of them?
Chris Hayes
No. I want to get a handle on it, and this is the way I do that — interviewing people, talking to them, thinking through something.
I found myself recoiling from the topic, partly because it feels like it’s being shoved down our throats. When I open my computer, it’s like, “Do you want to use AI on Zoom?” And no, I don’t. I just want to talk to my team. But at the same time, this does seem to be a pretty important development in the general lay of the land of American society, economics, and politics. I needed to get out of this defensive crouch and try to learn. That’s really the impetus, and what I’ve discovered is that a lot of people feel the same way.
Every discussion of this makes me want to run away from it. But I do feel like I should get my arms around it a little bit, because clearly it’s coming for us.
Thor Benson
How did you think about picking who to talk to on the show? If you only invited people in the industry, you’d get one perspective, but economists might have a very different one.
Chris Hayes
We were trying to vary it up. There’s no shortage of people in the industry talking to each other on podcasts. That is the majority of AI-related content. In some ways, I was trying to take a step back from that to be with people who are trying to think deeply about it, either as critics or as researchers or philosophers.
To me, the most interesting thing are the implications and what it means for a bunch of things that I’ve thought about for a long time, like the mind-body problem in philosophy. I also wanted to have some critical voices, including extremely critical ones.
Thor Benson
There are all sorts of extreme takes out there — everything from people saying AI is going to be the end of the world, to others saying it’s nothing but hype. Where does your view fall?
Chris Hayes
I try to be radically open to the full spectrum of views, and I still think, because it’s so uncertain, that outcomes anywhere along that spectrum remain possible.
The one thing I would say for sure is that I don’t think the technology is bullshit. It just very clearly is not. I do, however, think it’s possible the financing and economic investment are going to collapse in on themselves. That seems very plausible to me.
When we were coming out of World War II and had dropped two atomic bombs on Japan and were in the midst of what was then called the nuclear age, there was a notion that the transformative technology of the century was nuclear power. That made a certain sense. Nuclear science had obviously created the most powerful weapon, one that could destroy all human life, and it also seemed destined to solve the energy problem once and for all. But I don’t think anyone really looks back on the 20th century now and thinks nuclear power was the defining technology.
That isn’t to say that nuclear power was a fraud. But I think there’s a possibility AI ends up being kind of like that. It’s clearly incredibly powerful, but maybe it’s so powerful that its containment necessitates a way of dealing with it that makes it impossible for it to be deployed at scale.
I don’t think it’s a fake technology. By every possible metric, it’s very real. How large a role it’s going to play in our lives, how transformative it will be, whether it makes sense financially, I’m still pretty radically open about all of that.
Thor Benson
As someone who’s been writing about AI for a long time, I find it very unsatisfying to hear comparisons to fax machines or spreadsheets, because those things weren’t trying to replace our brains. What’s your view on how unique AI is?
Chris Hayes
There’s a difference in degree versus difference in kind question that we keep circling around on the show, and I am more in the difference in kind group. But I keep coming back to what it would mean to think of AI as a “normal” technology.
What are the most radically transformational technologies of our time? The car’s got to be up there. It basically changed everything about how modern life works. But because AI is trying to replicate the thing that’s most internal to us, it feels the most threatening, the most scary, and also potentially the most transformative. There’s no way to escape that. The other thing is, it’s amazing how adaptable we are, how much we will adjust to a computer being able to do a thing that it couldn’t do before, and just keep living our lives.
Thor Benson
Do you have a kind of hierarchy of concerns with AI? Are you more worried about how it’ll affect the economy, democracy, society, or some combination thereof?
Chris Hayes
The proximate threat right now is massive job dislocation and economic disruption. That could have a seismic effect on the economy and political coalitions. There’s also so many safety concerns, particularly around cybersecurity and warfare. Adversarial systems working against each other is pretty scary.
The information aspect, which I don’t think we’ve even really started to touch, is front of mind for me. We have one very clear example, where it seems someone at xAI, possibly Elon Musk himself, added a new layer into the model about “kill the Boer” and the persecution of white Afrikaners. All of a sudden, Grok was just spewing this stuff at every opportunity.
In the same way that a generation of people would go to Google if they were wondering who’s running for office or something like that, now people will ask Anthropic or ChatGPT or Gemini. The power that comes from putting a finger on the scale of those models to manipulate the political information people consume is so insane to think about.
Thor Benson
Obviously, you spend a lot of time covering politics. AI is extremely unpopular in the polls, and people don’t like the tech oligarchs and data centers. How do you see the politics of this playing out?
Chris Hayes
It’s going to be more and more salient. The stuff that’s happening at the grassroots level is fascinating. It’s really a political jump ball, and we’ve just lost the muscle memory as a society to think through how to regulate this stuff and bring it under democratic control.
We obviously constructed that for the entire nuclear system. We have an FDA. We have a Federal Reserve bank. We’ve got all sorts of institutions we’ve created over the years to finds ways to combine democratic control with technical expertise, so you don’t have plebiscites or congressional votes on interest rates or new drugs, but you also aren’t just like, “Well, you guys can just do whatever you want to do.”
It’s very clear to me that whatever framework we call it for AI is going to have to be in that middle space. It’s been a while since we’ve shown an ability to pull that off.
Thor Benson
With your research about AI for the show, what have been some common misconceptions you’ve been thinking about, or new things you’ve learned?
Chris Hayes
It’s interesting to trace through the ascent of the complexity of what it’s able to do and how quickly it’s getting better. The fact AI keeps crossing thresholds of tasks is pretty stunning to me.
The most mind-blowing thing I keep coming back to, which is actually from Gideon Lewis-Krauss’s piece in The New Yorker and something he said on my podcast before this series, is to think about how we had the steam engine before we had the laws of thermodynamics — which is to say we had the practice before theory. We were able to build an internal combustion engine before we understood why it worked. I find that parallel pretty fascinating.
We’re using this technology we don’t quite understand that is reliably churning out results that are pretty unsettling and uncanny in how well they are able to mimic sophisticated, intricate chains of genuine, causal reasoning. That’s fascinating to me.
Thor Benson
On a personal level, you’re someone who likes to write and you also have kids. How are you thinking about how it’ll impact your work and the future your children will live in?
Chris Hayes
I think about it all the time in terms of what their college education is going to be like, about the fate of research papers and essays, and I think about how learning to write is learning how to think. It does seem like it’s obviously going to transform big chunks of how kids learn. The biggest threat is that it just makes everyone stupider, which obviously is a genuine worry.
Thor Benson
How do you think about using AI in your personal life and for your kids?
Chris Hayes
I don’t think my kids are using it at all. For my own life, there are tasks where I don’t want it to do the thinking for me, but I do use it to do what I consider more rote stuff, research. Checking stuff.
I distinguish between thinking and generating and grunt work. I would never be like, “Hey, what do you think is a good outline for this book?” That’s my job. But I will be like, “Hey, what year was this guy born? Can you search all my files for something? I think it’s in one of the PDFs.” Maybe that’s a fake distinction and I’m getting stupid, too. Right now, it definitely has the feeling of a bicycle for the mind that Steve Jobs described.
That’s it for today
No podcast this afternoon, but we’ll be back tomorrow with a special Saturday edition of the newsletter. If you appreciate today’s PN, please do your part to keep us free by signing up for a paid subscription.
Thanks for reading, and have a great weekend.
The “makes everyone stupider” line is the right worry aimed at the wrong target. AI does not lower intelligence. It removes the friction that built intelligence in the first place. Learning to write is learning to think, Hayes says, and he is correct, but the mechanism matters. You do not get the thinking from the finished essay. You get it from the struggle of producing it. Outsource the struggle and you keep the output while quietly losing the faculty. The danger is not a dumber person. It is a person who cannot tell the difference, because the results still look fine.
Hayes feels this himself. He draws a clean line between thinking and grunt work, then immediately admits the line might be fake. It is fake, or at least it moves. “What year was this guy born” seems like pure retrieval, until you notice that the act of hunting through your own files is how you stay intimate with your own material. Every offloaded task feels trivial in isolation. The cost is only visible in aggregate, years later, as a capacity that no longer answers when you call it. That is the actual endgame worth a podcast. Not whether the machine can think, but whether we will still bother to.
Johan 🐌
What bothers me is the intent of AI’s tech bros masters: How can we trust them to act humanely vs. financially? The Anthropic mess pits moral v. amoral; and those tech bros prefer profit first.