Loading summary
Transcript59 lines
- [00:00]
A
Hey there listeners. Before we get started, a reminder that we have a live show coming up on March 4th at the Comedy Cellar in New York City. Nate Silver, Claire Malone and I will do our first ever 2028 GOP primary draft. It's also the day after the high drama Texas primaries, so we're gonna have plenty to say about that as well. It's gonna be plenty of fun. So grab a beer, grab a ticket. I'll drop a link to the tickets in today's show notes and we really hope to see you there. The moment of panic has aligned very well with all of your interests.
- [00:36]
B
Yeah, you know it's recession proof, right? That's what they say.
- [00:39]
A
I mean, until a chatbot figures out or an agent figures out exactly how to write really compelling data driven pieces about.
- [00:48]
B
Well, we can get into that.
- [00:50]
A
The Western world's deepest anxieties.
- [00:52]
B
Yeah, we can get into that because I feel like that moment may already be here.
- [01:06]
A
Hello and welcome to the GD Politics Podcast. I'm Galen Drake. If you've been enjoying your long weekend, I apologize for the potentially panic inducing content of today's episode. We seem to be in something of a freakout moment over artificial intelligence. In particular, there have been several viral posts going around on social media from people who work in AI warning of what's coming. Rinak Sharma, who was an AI safety researcher at Anthropic, quit last week, penning a letter saying, quote, the world is in peril and that we need to wise up. Zoe Hitzig, an economist at OpenAI, quit and penned an op ed in the New York Times criticizing the way that ChatGPT is implementing ads suggesting the company could use people's private motivations to manipulate them. Matt Schumer, the CEO of an AI startup, wrote a viral post on Twitter called Something Big Is Happening comparing this moment in AI to what February 2020 was like for Covid. As far as the markets are concerned, software stocks have fallen 15 to 30% over the past month in reaction to new AI developments in coding. So today we're going to talk to somebody who's been using Data to track AI's effects on the world so far, particularly when it comes to to work. John Byrne Murdoch is a columnist and chief data reporter at the Financial Times and friend of the podcast. Welcome back to the podcast, John.
- [02:27]
B
Thanks, Caitlin. Great to be here.
- [02:29]
A
It's great to have you. Also, in case AI panic isn't enough for one episode, John's also been doing a lot of work recently tracking Democratic backsliding in the US and around the world. So fittingly for President's Day, we're going to get into his research on that and perhaps we'll even ask if the two causes for panic, AI and Democratic backsliding, fit together in some way. But we're going to start with AI for now. So, John, in your estimation, why is this AI freakout happening right now?
- [03:01]
B
I think a lot of it comes down to the explosion of agentic tools. So we've obviously been talking about this since, in a big way ever since ChatGPT came out late 2022. But it was mid last year when the agentic tools like Claude Code Codex, that kind of thing really took off. I think that is just a huge step change in terms of what you experience as anyone using this stuff. In terms of you're seeing it now, just get on with stuff, just do stuff right in front of you, which obviously wasn't the case with the Chatbots. And then simultaneously, obviously the people working at these companies are now using these tools to recursively improve and build stuff again at just a much more rapid rate than what we were seeing. So I think for me it's that shift from standard LLMs to agentic LLMs, which for users and builders alike has been a real, real step change in.
- [03:58]
A
The experience does the current freakout, and I guess freakout seems to suggest almost that it's not warranted. And I don't mean to make that assessment. Perhaps it is, and maybe we'll try to figure that out through the course of this conversation. But is that current panic, if you will, more about job displacement or safety?
- [04:19]
B
That's interesting. Right? So I think there are almost separate panics going on. When you're looking at some of the resignations departures you were talking about in the intro, I think a lot of those are more to do with safety. These are people who are incredibly close to the model development. They're seeing, I think, a lot of the stuff around safety that the average person or even most heavy users are not seeing. So that stuff I think is quite safety based. There are people really stress testing what are the worst things that can happen if bad actors or just certain types of people are engaging with these tools. And I think it's what those people are seeing that is maybe prompting some of these departures. Whereas the concerns about job replacement I think is a separate thing. That's where the conversation among economists and some white collar workers is. So I'd say, yeah, two separate panics. You've got one which is looking at this edge, this extreme risk, which is maybe less likely, but extremely bad if it happens. And then you've got the broader societal socioeconomic concern about what might be about to happen to white collar work.
- [05:27]
A
If you had to choose, which is of more concern to you as somebody who's, who's relatively close to the space, could you articulate that?
- [05:36]
B
I mean, for me personally, it's the impact on white collar work. And I say that as I've been someone using Claude code very heavily in 2026. And so I'm seeing, you know, literally six inches in front of me on my screen, what these capabilities are looking like, how they're advancing, how capable they are. So for me, that's something I can palpably feel. Whereas of course, the safety stuff I'm hearing about it, I speak to people who work in that space and they're certainly very concerned, but it's just, it's harder to get my head around something which just feels a couple of steps removed from things that I'm sort of seeing and experiencing myself.
- [06:13]
A
You wrote in the Financial Times, quote, in a matter of months we've moved from, quote, no solid evidence that AI is boosting productivity, to, quote, AI is now writing fully functioning commercial software that threatens entire industries. Can you flesh out that transition that you're talking about and what it means?
- [06:34]
B
Yeah, I mean, the piece that that appeared in was one where we were essentially revisiting an analysis that had been done late in 2025, where a skeptical software engineer had said, look, I'm hearing all this stuff about AI, about LLMs helping software developers do loads more. Everyone in my sort of community is talking about this. So let's ask the question of what that would look like if it was true. So he analyzed activity on GitHub, the coding repository. He looked at new website creation, new app launches for iOS, that kind of thing, and essentially saw a series of flat lines. And what we thought we would do is come back to that a few months later and look at those same metrics which had been used then as evidence that this was all not happening, to see what they look like now. And they all now show these very, very clear upticks. So the argument that nothing's happening here, I think just. Is very clearly just no longer holds water. We're seeing all of these different indicators of activ and output shoot through the roof. Now there's obviously a separate question of does more mean better if more stuff is being produced, but it's lower quality than what we saw before, is that Actually a boost in productivity, could it even be net negative? And I mean my counterargument to that is that one of the things, one of the bits of code that would show up in this kind of analysis as something that has taken off in the agentic era is the actual creation of some of these highly successful professional software tools like Claude Cowork, which are now threatening lots of white collar work. So this is not just a case of people pushing loads of half baked ideas and scrappily coded agentic AI assisted code out there. This is real production ready tools being made using this stuff. So yeah, what I'm talking about there is, we do clearly now seem to have answered one question which is just, are these tools now helping people do more, more work? Some of which is really valuable stuff.
- [08:47]
A
Yeah, and I think what you're talking about there is the idea that Claude code created Claude cowork in 10 days all on its own. And Claude Cowork is now at least sweeping the Bay Area. And you're seeing sort of these videos online of people lining up all of these Mini Macs with a bunch of different agents doing essentially all of their work. It's almost like a sort of robot farm of AI agents doing, you know, I don't know exactly what they're doing, but they're coding. They now have access to the different apps on your computer and can make them work together. For people who are not fully familiar with this, and I'll say even until I started preparing for this conversation, I had never seen those, and this is a new development as well, those videos of the Mini Macs basically doing everyone's work, what kind of stuff are they doing? Apart from the kind of canonical example of Claude Cowork, how are these agents getting put to use?
- [09:50]
B
I mean, they do an absolutely enormous range of stuff. And I think this is one of the things that people don't appreciate, or a lot of people don't appreciate because you hear about coding and you think, well, coding, that's this thing that software developers do and it's this thing that happens in the Bay Area. But I think if you take a step back, essentially anything that happens on a computer or via a computer involves code, right? So when you put together a slide presentation or draw an illustration or write a document, all of these things, you know, ostensibly you're moving a cursor or typing on a keyboard, but ultimately the interface between that and the end result is code. So anything that can be done on a computer can be done by code. So that means a Huge range of white collar work can be done this way. So editing documents, searching the web, putting together research, checking the veracity of various things, doing calculations. All of this stuff is on a, on a very abstracted level code. And I mean I speak to lawyers who are using this stuff now, like, and trusting it, I speak to accountants, all sorts of domains where this is now in use. And I mean I've been using agentic tools in my own work at the FT as well. So yeah, I think people's concept of what code is for, I think is quite, is a lot narrower than what it actually is for. And that means the range of fields that are touched by this is a lot larger than people might think.
- [11:31]
A
Yeah, unfortunately we are an audio medium, so folks can't actually see all of the really great charts that you've put together at vft, but if folks go and check them out or I'll, I'll describe them for you. Now imagine a hockey stick, right? Sort of code being the amount of code being produced or new websites going online is pretty flat. And then all of a sudden once these agentic AI tools hit the market, there's an explosion in new, new code, new websites going online, et cetera. And I think the idea is that people who code are seeing the way that AI agents can basically make their own jobs obsolete or at least massively change them. And so they're sounding the alarm because it's happened to them. The implication being that, you know, it's going to happen to all of us too. So I guess first, are we seeing it show up in employment numbers that coders are facing displacement?
- [12:26]
B
This is such a hotly debated question. So a lot of people of course, have been looking at this for the last couple years, ever since it felt like it was a possibility. You've got economists, other researchers, data journalists, all sorts of people looking at this question and the tricky thing, or there are many tricky things to really unpick whether AI is causing anything we're seeing. So one is that, so you need a hypothesis, right, which is essentially, are we seeing different employment trends in occupations that are likely to be more exposed to AI than others? So that requires you to know, like, right, which occupations would we expect this to show up in? So people have had various attempts at coming up with this. There are various indices of AI exposure. And so the basic analysis is, right, are the most exposed, Are the most AI exposed, jobs seeing worse employment trends exposed and the least exposed? And this again, the answer to this question varies. It varies depending on exactly when you look, it varies according to different countries and you get some relatively high exposure jobs that are seeing bad trends, some are seeing more positive trends. You also see different trends according to whether you look at younger or older workers. There are so many ways of cutting the data and those each yield slightly different results. The biggest smoking guns we've had on this is from work by people like Erik Brynjolsson at Stanford and a couple of other US researchers who have used really, really fine grained data and said if you drill down to occupations like software and you drill down within those to the youngest workers, then you see that the number of 22 to 26 year old software engineers, for example, has trended downwards at a time when employment in other occupations and among all the software workers has trended upwards. So that's a solid smoking gun. The problem you then run into though, is something else that is true of software as well as it being exposed to AI is it's heavily exposed to changing interest rates. And what happened a few months before the launch of all the AI tools was the sharp hike in interest rates that caused a big pullback in hiring in occupations like tech. And so you've got a competing hypothesis that would fit very neatly with that same data and it wouldn't be to do with AI. So again, people have responded to that and used more fine grained analyses. And it's Eric Brynjolz's team who I think have done the best, most rigorous work on this, where they try to control for what's happening with interest rates as well. And they still do see suggestive evidence that young software engineers are having a tough time because of AI. So yeah, there's a huge amount of attention going onto this. Different analyses give different results. I think my synthesis of it would be there are some signs that something to do with AI might be happening, but because you have to adjust for so many overlapping factors and you have to look at such small pockets of the labor market, it is still in that sort of academic debate realm. It's certainly not the case that we're seeing clear sort of macro macroeconomic signs across one or more countries that AI is taking work.
- [16:03]
A
The idea behind the impact of AI on coding is that first the goal was to make AI really good at coding so that it could help create future versions of itself and sort of cut the time in half again and again and again of how long it takes to create new AI tech. And so first it became really good at coding, but once it becomes really good at coding, it's going to Become really good at everything else. Legal research, accounting, all, all of these kinds of things. And that it's just a matter of time before whatever impact we might see on, say, the youngest coders will have an impact on the youngest accountants, the youngest lawyers, et cetera, et cetera, et cetera. Other people say the way that computer programming works is very rigid, very specific. It's like speaking its specific language. The applicability is not one to one to how you can get an AI tool to engage in that kind of a world versus one that takes nuance and judgment and finessing and taste and people skills and all that kind of thing. So that the rate of displacement in the world of computer science programming will be much faster than AI's ability to displace many other fields, even including white collar work. Which side of the equation or which side of the argument do you fall on? Like it's only a matter of time before AI gets just as good at being a lawyer as it is at being a coder, or there is something just wholly different about work outside of coding.
- [17:40]
B
Yeah, and that's a great, great way of framing it. I think, you know, the, the boring answer is somewhere between the two, I guess, the slight to sit a bit less on the fence. I'm probably a little bit closer to the second. So I do think this is clearly.
- [17:56]
A
You mean journalists are. We're in the clear.
- [17:59]
B
Don't know if anyone's in the clear. But yeah, we've maybe got a little bit longer. I mean, the fact that code either works or it doesn't, you know, in a way that is not true of really of language or various other types of knowledge work, I think that is an important distinction. You know, vibe coding. If you ask for these tools to build you an app that does something, it will build you an app that does something. Whatever is going on exactly in that code, it will do what you've asked it to do. The problem or the difference if you ask it to do some research, some quantitative research for you, for example, is there are so many different legitimate ways that a particular piece of research could be done, and there are so many ways for things to sort of quietly fail or quietly go off down the wrong path without being quote, unquote wrong in a sort of two plus two equals five sense that it is less immediately threatening in those spaces. Or it is, as you say, it just works in a very different way. So, yeah, I mean, I think it's almost like the more clearly defined and falsifiable a task, a type of work is the more it's in trouble. So that's why in law, we're seeing particular types of relatively lower grade legal work come up under threat. So just pouring over huge numbers of documents to find cases that are relevant to a particular case and then sort of noting down where in the document that thing happened. Like that is something that is quite clean and quite falsifiable, whereas other types of digital knowledge work that's less true. So, yeah, I do think it would be. I don't think it's as simple as saying, right, what we've seen for coding, we're going to see with everything else. But, but I think the neat, clean, falsifiable bits of lots of other white collar work, of which it turns out there is a lot, are in trouble. Right. So that's why we're seeing particular talk about things like accounting and auditing, which are not software, but they are still about falsifiable things, about following rules, about checking whether something is true. That kind of work, white collar work, is interrupted.
- [20:17]
A
Yeah, I mean, you talked about quantitative research. You actually, in your column is called AI Shift, you explored whether AI could replace the job of a quantitative researcher. What conclusion did you come to?
- [20:30]
B
Based on what we wrote and based on just that whole community has been really sort of getting into just getting its hand dirty with these tools. I think we're seeing something where the capabilities of the agentic tools are very impressive, but they do certainly still need their handheld. When people have tried to do things like just say, get an LLM, an agentic AI and tell it to just go off and do some new original research, generally it comes back with something that is sort of superficially plausible, often not incorrect, but would be considered fairly low grade in the academic field that it's been done in. Whereas when people are using these tools, sort of working hand in hand, getting them to automate or speed up large parts of their own research, then the sort of combined human plus AI output there is really, really high quality. So I think we're still in a lot of white collar work. We're still at this stage where the AIs are really impressive and really helpful, but they're not yet at the point where they could just go off and automate an entire person's sort of output in a way that maybe they can start doing in certain bits of software.
- [21:52]
A
Yeah, I think listeners to this podcast will be familiar from the segment that we do called good data, bad data, not data that just because you have, you've sort of collected Data and can say, oh, you know, this number equals this doesn't mean it's founded on a compelling argument. Like, right. There's a lot of skill that goes into what questions you ask the data, the different sources you might pursue. Like it takes a lot of judgment and knowing, well, there are advantages to pursuing this kind of data versus that kind of data, revealed preference versus expressed preference and the like. And I think it does take some, you know, as we talked about judgment or taste now some folks in the AI world say the thing that really scares them is that AI is starting to show things like that feel like human judgment and human taste. So maybe our jobs aren't so safe after all. But I guess maybe only time will tell. I don't know. Do you have thoughts on the idea of humans actually replacing their own judgment and taste with agents?
- [22:59]
B
I mean, for me that this is, this is just a very sort of hand wavy answer. But I, but I do think there's something about human taste that is different. Right. That people care more about, or at least that AI taste filtered through a human seems to be given more value by us as a society than pure AI taste, as it were. Right. I don't think we're yet seeing outputs where an AI has just completely decided what to do and done it. And people are saying, well, this is awesome. I think we're generally seeing AIs with handholding do offered really impressive things which are then sort of admired and valued. Now again, that doesn't mean that the pure AI version of that could work, but I just don't think that's what we're seeing yet. So even to take the early example where we're talking about how Claude Code wrote the code for Claude Cowork, that was not all done autonomously. Right. It was writing that code in dialogue with all of Anthropic's engineers. So I mean, look, the story of the last three years is something hasn't happened until it happens, and it might be tomorrow that this suddenly does start happening. But I think the human as a sort of filter, as a kite mark of approval, as it were, is still a really important role. And there's the question of if so much that we do as societies today, the sort of importance and value of something is attached to or at least related to the fact that it came from a certain person or certain people. So that I struggle to see completely going away. But you know, we are clearly already seeing a lot of work that has been produced, at least to a large extent by AI and which is appreciated and valued by humans.
- [24:53]
A
Yeah, listeners right in. Would you listen to the GD Politics podcast if it were hosted by an agent? I'm curious for your answers, which. And on the topic of, you mentioned vibe coding earlier, which is essentially telling an agent you want it to create code that develops an app to do this thing or that thing, or telling an agent to explore this data set to try to figure out whether or not this demographic is trending in a certain, you know, whatever. Like basically vibe coding is, is knowing what questions or what direction to give an agent. Which, which gets us to the theme of like, will AI create new jobs? Right. You know, the AI optimists are saying time and time again during technological disruption, everyone freaks out that they're going to lose their job, but people end up moving into new jobs. Do you have a perspective on whether AI is. It's really different this time because since it mimics human capabilities, there's no thing that can't be disrupted. Or hey, AI is going to create a bunch of new jobs.
- [25:57]
B
There are so many ways of coming at this one. The very, very narrow way is that, yeah, there are like the job of AI safety researcher didn't exist three years ago, right? So there's a job that's clearly been created by AI. Then you've got. There are again sort of directly related jobs. Like I think we are going to end up where there will be some jobs which are essentially checking the work of AIs. Because that for me still feels like a big. In a lot of industries, unless something has been verified by a human, just for regulatory, legal reasons, whatever, it's no good. So there's clearly going to be need for humans there in a way that may count as a new job, then you can almost go a couple of steps removed and say, well, if AI takes away a lot of white collar work, will that indirectly create a lot more care work and service work for human contact roles? Because there will simply be more demand for those types of jobs. So there are so many ways of looking at this. I don't know if it's as simple as saying, well, if AI takes various jobs, that's just going to lead in a very simple sort of zero sum sense to fewer jobs. But yeah, it's tricky. We looked at this recently to see if there was evidence of job creation. And apart from, as I say, those very narrowly defined jobs with AI in the title, it's not clear that we're seeing that yet. I guess again, the other question is, well, when productivity in a certain Occupation massively increases. That can again go in two ways. You either get less of that, like fewer people or fewer hours being put into that job because you can do more per hour, or the fact that you can suddenly be so productive in that role means there is more work going on. And we even see that show up differently in different countries, different sort of working cultures and environments. So yeah, I certainly don't think it's as simple as. As AI displaces work, there's nothing happening on the other side. But yeah, it's hard to. It is qualitatively different, as you say, to things that have taken work away in the past.
- [28:07]
A
Today's podcast is brought to you by you, the listeners. Without paid subscribers, GD Politics simply wouldn't exist. Your support means that we can make an independent podcast guided by curiosity, rigor, and a sense of humor, and that we can have conversations like the one we're having on today's episode. Paid subscribers get about twice the number of episodes and can join in the paid subscriber chat to pass along questions for us to discuss on the show. You can also connect your personal feed to your favorite podcast player so you get every episode wherever you listen to podcasts. Become a paid subscriber today@gdpolitics.com we've got plenty of great stuff in store coming your way in 2026, so it's a great time to join the crew@gdpolitics.com we hope to see you there. I want to drill down on this just for a second because I thought your work in this area was particularly interesting. You talk about latent demand, basically this idea that, hey, if I could hire an interior designer, I would. It's just that, you know, I don't make enough money to hire interior designers at their current costs. But if productivity goes up massively and just, you know, it costs less money to interior design, then a lot more people can afford that service. And you know, say in the same. Same way that everyone gets a dishwasher, everyone gets an interior designer or something like that. Can you sort of like flesh out that theory of how AI doesn't necessarily sort of take away jobs, but just creates, I guess, a massive amount more of consumption?
- [29:40]
B
Yeah, I mean, so that would essentially be saying that the amount of output people can do with a given amount of time significantly increases through AI. That means your wages effectively fill up, that hour of work go up significantly. If there is enough demand to keep up with that increased output, then essentially everyone's wages are going up without the amount of work coming down. So then you just have a richer society and people would be spending that money on something. So that's one version of looking at that. Another interesting version that a couple of people have brought up in our reporting is the idea idea that the vibe coding type uses of AI are essentially a gateway to people then wanting a more professional paid for version of their thing. They've just vibe coded. So one version of this might be people, and this is not advice for me, but people getting legal advice or talking through some legal or tax related thing with say chatgpt and then realizing, oh right, there's a thing here that I should maybe do. And then they go and get in touch with an actual accountant or lawyer. So the AI tools, you mean the.
- [30:58]
A
Thing that humans have been doing with like WebMD and their doctors for the past 15 years?
- [31:04]
B
Exactly that. Right, yeah. The healthcare one's a really interesting example. Right. Like you can now get a lot more questions answered online about health, but that hasn't led to certainly in any data or research. I've seen a drop off in healthcare utilization. If anything, some people are maybe discovering that there are more things they could go and ask their physician about. So yeah, there are lots of ways. Certainly in the current place we are with AI, there are ways in which the fact that it gets you halfway there could actually lead to increased demand for paid for human services.
- [31:38]
A
You cover AI mainly from a work perspective, as we've discussed. But I'm curious, before we move on to democracy, how do you think about it all from a safety perspective?
- [31:51]
B
Honestly, I mean, I'm not massively in the weeds of this, but I guess there are two parts even of the safety discourse here. So the sort of X risk, the extinction risk, low percentage but catastrophic outcome stuff is something. It just breaks my brain a little bit right now, to be honest. It feels so many steps removed from where we are. I understand it conceptually. We're talking about, for example, people being able to build biological weapons using these tools, that kind of thing. That's one version of it. Of course there are other flavors which are about AI just taking over the world directly. I'm just not close enough to what's going on there to understand this. But I understand it enough to know that these are really serious risks, at least if we think about the first one. Like the LLMs, helping someone, helping a bad actor make a lethal weapon, that's something we can all get our heads around. It's stuff that AI safety research labs are constantly testing and frequently showing that if you really, really, really, really try, you can get it to do these things. So I think that's absolutely something people should be concerned about. I guess a lower, a less lethal version of that is we're already, I think, by most accounts, seeing an uptick in cybercrime cyber attacks, which have been aided by using LLMs. So, yeah, I think that stuff is a very real risk, and it's great that there are a lot of very, very skilled AI safety researchers working on that stuff. Then you've got the more micro level, which is things like AI psychosis. So people who are led down quite disturbing paths by their conversations with LLMs, and people will already have seen and heard of examples of this going catastrophically badly in terms of leading to people's lives ending. So that, again, is a real risk. And again, it's something that the safety teams are looking. And both of these, the macro and the micro, are sort of very, very hard to guard against. The companies and the researchers and governments, I think, are doing as much as they can in that space. But based on the snippets that we hear from out of these communities, these are still very, very real concerns that people who think about this a lot have.
- [34:09]
A
Yeah, I'm sure there are people who will say, oh, the governments and the researchers aren't doing as much as they could in that space. I mean, certainly in the US we have a sort of. Of more laissez faire approach to AI under a Trump administration than we probably would under a Democratic administration. And I think, you know, we're not really there yet in our politics, and it's taken us forever to even sort of tackle the effects of social media on society. So I don't know how quickly we'll get to the effects of AI on society, but certainly there's a whole other podcast episode for us to have about the politics of AI regulation and whatnot. But I, you know, I wanted to say, say I know that you study a lot of social trends in addition to. I mean, this is a social trend in some ways, but, like, reading the news lately about AI gave me something of a similar instinct to the safety researcher at Anthropic who said he's moving back to the UK to disappear and write poetry. There's kind of an early Covid quality to it all where, you know, part of you, even if you try to be rational about it, like, wants to withdraw from the world. It's like, this is so overwhelming. It seems like there's so much change afoot, even just that. How do you think about that instinct? What impact might that have on society? Just fear of the amount of change that we're about to experience?
- [35:29]
B
Yeah, I mean, I mean, I think it's already affecting plenty of people. You know, there are the more acute cases like that where someone literally does just sort of pull back from society and tries to hide from it all. But you know, there's been reporting recently about sort of people having low level anxiety about everything that's going on here, especially people in the software space who are experiencing this and seeing this. So I find for myself I often have to just sort of decouple almost when I really start thinking about whether it's the outlook for my own job, whether it's certainly the more extreme safety stuff, I'll think about it and then I think if I'm going to be able to go about the rest of my day, I just need to block that out. So, yeah, I do think this is something that if one really wrestles with and looks at the reports or try some of these things out, it is quite hard to then just cheerily go on with the rest of your day while that's in your head.
- [36:27]
A
As somebody who's covered American politics for basically all of my career, I understand the idea of decoupling one's personal life from one's work. Looking at a more macro level and trying to put some numbers to this. Gallup recently released polling showing that American optimism slumps to a record low. They write, quote, the percentage of U.S. adults who anticipate high quality lives in five years declined to 59.2% in 2025, the lowest level since measurement began nearly two decades ago. 60%. I guess that's not terrible. But nonetheless, I guess Americans are also known for their optimism. Since 2020, future life ratings have fallen a total of 9 percentage points, projecting to an estimated 25 million fewer people who are optimistic about the future now versus then. I don't want to sort of say this is all about AI because I don't have the evidence to say that, but we're about to transition into like democracy and sort of anxiety about the American political system. We've been talking about anxiety about our technological future. I mean, what do you make of Americans turn towards the more pessimistic side of things?
- [37:37]
B
A. I would say this is pretty common outside the US as well. Right. So something I. There's a data set I often look at at is called the Executive Approval Project and that is a collection of polls about just approval of government and leaders and stuff all around the world and pretty much everywhere is at or very close to record lows, or certainly modern lows at the moment. So I think this is all sorts of things on top of each other. The whole AI discourse, I think, is quite bifurcated. You know, there are people, as you say, who are fleeing for the hills about this, but there are a huge number of people of whom this is just not registering on their minds at all. So I would be skeptical that AI anxiety in particular is playing a big part in those numbers. Maybe playing a part, but I think you've got inflation, you've got Trump and various other political disruption in other parts of the world. You've got tariffs, you've got the very weak labor market again across most of the world. So I think it's all of that layered on top of each other at the moment, plus things being experienced through the lens of social media, which tends to amplify negatives. I suspect it's. Yeah, it's. It's a sort of all of the above situation.
- [38:55]
A
Yeah. Okay, let's talk about democracy. Specifically. You wrote a piece questioning how steep is Trump's democratic backsliding. And you wrote, while US History is hardly free from political violence or maltreatment of disfavored groups, this blitz on America's citizens, institutions, and by many estimations, the Constitution itself, ranks as arguably the most rapid episode of democratic and civil erosion in the recent history of the developed world, measured using objective criteria spanning 10 domains, including the use of state force against civilians, political prosecution, and the independence of the judiciary and civil service. I find that the US Slide during Trump's second term stands out as the most rapid in contemporary history. It outpaces the early stages of backsliding under Russia's Vladimir Putin, Turkey's Receptayyip Erdogan, and Hungary's Viktor Orban, where similar steps unfolded over several years. Okay, how did you conduct this research? I mean, that it's a. It's a bold statement or conclusion that you're coming to. Of course, as we discussed already in this podcast, you had to make a lot of judgments about which events to include, how to measure them and whatnot. So I'm curious, this. How did you get to that conclusion?
- [40:13]
B
Yeah, this was a piece I've been planning to do for a while and working on for a while. And so various organizations and researchers have developed these sort of indices of democratic health and backsliding and that kind of thing over the years. But the drawback that these all have is they tend to Be backward looking. They're basically telling you how things were going a year or more ago go. So when we're dealing with something happening very quickly right now, looking the review mirror is never going to be that helpful. So what I wanted to do was take the approach that these very well established academic efforts have taken and adapt it into something that could do this in real time. So step one was coming up with this sort of 10 items, 10 categories to include here. So this needed to be something that was methodologically legitimate in line with what are generally considered to be indicators of democratic erosion. So you've got things like the integrity of elections, of the judiciary, media freedom, that kind of thing. But I also wanted it to capture broader sort of societal and civil society problems, which pure democracy doesn't include. So things like the independence of central banks, which are obviously critical for economic health and security and their. And things like violence against civilians, that kind of thing, which aren't generally in these indices. So I put together, based on that research and a couple of my own additions, I put together this index. I then approached a few academics who are specialists in this space to get their sense of whether my list of items and specific definitions passed muster. They came back to me with a couple of tweaks, which I made, because one key thing here, when you're going to hang a strong claim off this, is it has to be really as objective as possible. It can't be something where you think, well, you know, I'll give America a score of one on this. It needs to be, did this happen or not? So there was a bit of refinement of the criteria like that. And then it was a case of just going out and essentially a lot of research, a lot of Googling, reading various think tank reports and media reports from over the years to establish if and if so, when each country had met one of these criteria, and to.
- [42:44]
A
Paint a picture for me of the speed of the American Democratic erosion during Trump's first year of his second term that you're talking about what kind of events charted on this sort of scale?
- [43:02]
B
Sure. So, I mean, there were some things that happened, you know, literally on the day of or within days of his inauguration back in January 2025. So there was things like the pardoning of the people convicted of political violence in January 6th. Right. I think because that has been talked about so much since January 6th several years ago, it almost feels like we're almost tired of that story now. But this is just not something that happens in normal functioning democracies. Right. Pardoning of people who have committed political violence. This is something that happens in Russia and it happens in Hungary and it happens Turkey. But it is really, really unusual. So that you had straight out of the gate, then you had Trump removing legal protections from civil servants, so facilitating sort of mass dismissals of government employees. Again, you can take each of these things and you can often say, well, you know, that's about efficiency and like, sure, maybe, but it's again, something that it really, really does not usually happen in other societies. It, it's, it's inherently sort of shaking up the system, removing expertise, that kind of thing. And this was often done with a sort of partisan sort of hint to it at the very least.
- [44:20]
A
And the goal of perhaps removing oversight and independent investigators.
- [44:25]
B
Exactly. So some of this was just, you know, large scale layoffs of people, regardless of the particular office they held. But yeah, you had removal of oversight officials specifically charged with tackling fraud and corruption. You've had had the Trump administration's frequent battles with the courts denying and refusing to comply with court orders. Even in as early as summer of last year, 2025, you had the police firing rubber bullets at protesters in Los Angeles. There was the removal of the chief of the independent Bureau of Labor Statistics. Then you had the investigation, the criminal investigation into the Fed's chair, Jay Powell, and then of course, the shootings of Renee, Nicole Good and Alex Pretty by ice. So it's a hu. Number of things that when you really just stack these up, this is just, it really is exceptional. Even when you look at what was happening in the first 12 months or so of Vladimir Putin's rise to power, or the early years of Erdogan, of Orban, or people like Chavez or Maduro in Venezuela.
- [45:30]
A
I will say I lived in Turkey in 2010 and 2011, so this is an experience I actually had myself. So it is sort of curious hearing the comparisons. I mean, do the kind of events that you're talking about happening in the United States, have they also happened in Russia, Turkey, Hungary and the like, but just more slowly, or is what's happening in America sui generis?
- [45:57]
B
Yeah, it's interesting. So some of, some of these are definitely things that, that happened and just took a little bit, a little bit longer in these other countries. But there are the, there are some differences between the types of things that are happening as well. So generally in most of these. Well, I say most, let's be more precise. So in countries like Turkey and Hungary, certainly there's less sort of explicit violence perpetrated by State actors that we've seen with, for example, ice. Now that again is perhaps something we shouldn't think of as necessarily specific to Trump, but more just sort of a magnification of how the US is already different to these other countries, like heavily armed police and government agencies, is something that is much more true of us than elsewhere. But that would be one difference. In some of the other countries you saw more overt threats to and capture of institutions and actions to immediately not just threaten, but damage the integrity of elections, which have happened elsewhere, haven't yet happened in the US Like a general pattern across all of these different events that I looked at in the five countries is that in the US you're seeing a lot more sort of unilateral actions and rhetoric from Trump and his immediate administration, generally bypassing institutions. Whereas what you were seeing in places like Russia, Turkey, Hungary was a more sort of slow moving but perhaps longer lasting capture of these institutions. So you've got Trump sort of bypassing the courts, whereas Turkey, Russia, Hungary, it was capturing the courts, Venezuela certainly as.
- [47:39]
A
Well, or at least attempting to bypass the courts and having success to greater or lesser degree. Which actually gets us to an important second piece of your research. And I'll say, by the way, in the same way I suggested those hockey stick graphs that you can see in terms of the proliferation of, of code and website generation. There's some also really good charts that are associated with this research. And you can see that while some of the democratic erosion is happening more quickly, this is not to say that the United States is currently in the same end place that Turkey, Hungary and Russia have ended up. It's just that the line drops more quickly. A cause for optimism in all of this is the resilience of, of American institutions which you started getting into. So how did you measure that piece of it?
- [48:31]
B
So this was essentially taking each of the items, actions, things on the index, as it were. So this was 10 items and then three levels of severity for each item. So taking each of these actions and saying, right, was this committed by just a unilateral executive action which bypassed institutions and therefore was perhaps very severe, but could theoretically be quite quickly reversed if there were then a change of power, for example, or reversed by the courts? Or was this something which used the institutions or took over the institutions in such a way that makes it harder to overturn and perhaps just permanently tips the balance in someone else's favour? So you've had things like very early on in, under Hugo Chavez, in Venezuela, there were changes to the constitution and electoral system, which just made it much harder for Chavez and his party to then be removed from power later on. You've seen similar things in Hungary and certainly Russia, whereas in the US you're seeing a lot of fire and fury. But thus far, if midterms were held tomorrow, for example, based on all the polling we have, it would not be a good result for Trump. And similarly, you mentioned the courts. There are a huge number of ongoing court cases against the Trump administration. And there have been several things over Trump's the first year of his second term where an initial action was found to be unlawful by the courts and was then reversed. Look at Kilmar Abrego Garcia, for example. So the US Is different in the sense that most of the damage, the vandalism, whatever you want to call it that has been done has been done in such a way that is not going to be as enduring, at least so far, compared to what we've seen elsewhere.
- [50:28]
A
You know, we like to talk about the United States on this podcast, but the US Wasn't the only subject of your investigation into the resilience of liberal democracy in the 21st century. And you asked sort of is liberal democracy in decline everywhere? What was your conclusion?
- [50:48]
B
Yeah, well, I mean, everywhere is certainly a strong word. And every time we talk about democracy and liberals and Trump, you have to remember Canada, where Trump is almost single handedly sort of propelling the liberals there and perhaps democracy forward. So it's not necessarily the same story everywhere, but there are various indicators. Whether you want to look at some of these academic indices of liberal democratic health that I talked about earlier, whether you want to look at vote shares for sort of mainstream classical liberal parties, whether you want to look at trust in government or even your optimism index earlier, we are seeing a lot of evidence in a lot of the developed world that however you want to define sort of liberal democratic, societal, socioeconomic health, as it were, things have not been in a great way over the last 10, 15 years. So yeah, I think this for me is just something such a huge trend that we're seeing in so many places. I talked earlier about approval of governments and leaders just being steeply down to sort of decade long lows in so many countries that yeah, there's clearly something happening here that I think requires a lot of thought as to what's causing this and therefore how we might be able to reverse that.
- [52:10]
A
Yeah, I mean, one of the culprits that you point to to is declining economic prospects and that as the rising tide lifts, all boats there's more trust in institution, trust of outsiders that is, you know, conducive to stronger liberal democracies and that as economic prosperity falls, the reverse becomes the case. I think that's, I think the sort of first part of that equation is well documented during the 20th century, and I think we're learning, learning to what extent the second part of the equation will be documented in the 21st century. Just to bring this all full, full circle, when we inject AI into all of this, I think a lot of people think that it's going to concentrate more wealth in fewer people's hands. Does that sort of exacerbate this liberal democratic decline and doom spiral?
- [53:11]
B
I mean, I think unless there's very careful policy work is done to redistribute that wealth, I think almost certainly. Right. The concerns about wealth inequality and especially billionaires, I think have really risen over the last decade or so. Not just post financial crisis, but especially recently. And certainly when people don't feel that their own lives are improving economically, financially, and that everything's expensive, of which obviously we've had with the bout of inflation, they do start looking at those who have a lot more money than them. So, yeah, if the increasing adoption of AI, increasing impacts of AI do increase wealth concentration, and I think that has to be the base case here, then yeah, certainly I think we could start seeing even more discontent, even more concern about inequality, and really sort of quite strong and rising public backlash. Especially given that one way of looking at what AI is doing is that it's sort of increasing the share of the population who can perceive of themselves as losing out in some way. Right. So the bottom half of the socioeconomic distribution, as it were, take like non college grants as a proxy, for example, already feel in some way that they are not society's winners. But if you now start giving college graduates, white collar workers a reason to think that they're losing out as well, then I think just the level of, of social and political discontent you're going to get is going to be really quite something.
- [54:39]
A
All right, well, with that, happy President's Day. No, I'm kidding. If you want to sound a note of optimism at the end, you're more than welcome to. Otherwise, I think we've had a really good conversation here.
- [54:49]
B
I mean, my optimistic thing is I wrote last year about whether we've passed peak social media, and I do think that heavy use of social media can be and has been corrosive to things like public trust and it has amplified the things that the genuine things that have been going wrong in the world. And I think if people do start spending less time on these platforms, I hold out some hope for a bit more social cohesion and perhaps a little bit more optimism to return.
- [55:21]
A
All right, well, we're going to leave things there for today. Thank you so much for joining me. John von Murdock.
- [55:26]
B
Thank you.
- [55:27]
A
My name is Galen Daruk. Remember to be become a subscriber to this podcast@gdpolitics.com and wherever you listen to podcasts. Paid subscribers get about twice the number of episodes and can join in our paid subscriber chats and pass along questions for us to discuss on the show. Most importantly, you ensure that we can keep making a podcast that prioritizes curiosity, rigor and a sense of humor. Also be a friend of the POD and go give us a five star rating. We're every whenever you listen to podcasts, maybe even tell a friend about us. Thanks for listening and we will see you soon.