Show notes
My guest this episode is Ramez Naam. Ramez has spent years thinking, writing, and investing around technological progress in energy and artificial intelligence. He is the founder and managing partner of Planetary VC, where he invests in emerging technologies across areas including clean energy, electrification, and AI. He is also the author of The Infinite Resource, More Than Human, and the Nexus science-fiction trilogy.Today’s conversation will focus on the pace of AI progress, the roadblocks that could stand in the way, and what all of this could mean for long-term economic growth.In this episode* The state of AI growth (00:38)* Roadblocks to an intelligence explosion (06:11)* AI in the short term (13:34)* AI’s economic impact (19:00)* Balancing AI safety and exploration (22:04)* What will power the AI economy? (31:04)The state of AI growth (00:38)AI capabilities are likely to be jagged for the foreseeable future. I believe in superintelligence, I believe it’s already here. We already have superintelligence playing Go. We’re crossing into superintelligence in formal math. That doesn’t mean we’ll have super capable persuaders, politicians or city planners.Pethokoukis: Where are we right now? As we’re recording this, OpenAI just released all these mathematical proofs and people are being wowed by that. Wherever you think we are at in the world of AI progress, does that result change your opinion?Naam: It’s amazing. It’s very impressive. I’m not sure I thought we would be here right now. Math is a very special domain. I think about the different types of domains that AI is good or bad at in terms of how highly verifiable they are. And AI first got really good at things like chess and then the game of Go, and now parts of math, formal math, because these have some interesting characteristics. You can generate an unlimited amount of data, and AI is very data hungry. You can get to a very firm yes or no with one-hundred percent accuracy. You can verify it as absolutely correct or not correct.But like human persuasion, politics, policy, and so on you can’t. You can do all of that at the speed of compute. You’re limited only by the speed of your computing to figure out if you won the game of Go or if the math proof is correct, which is not how it is in biological research or new material science, which are messy real world. So math is a domain that is pretty far over on the highly verifiable, symbolic, perfect training data, absolutely correct yes or no. All of it happens in software. Where something like curing aging is messy.We don’t have the training data at all. We’re going to need new instruments, new science, you need experiments, and your lag to verify anything is years, decades. And so that’s something that AI is not going to be as inherently good at, in my view.I made you leap over my first question, which is sort of where we are. There are certainly people who think that, by the end of 2027, there will be no debate that everything is changing in a profound, rapid way that changes all aspects of life. And then there are those who say “that doesn’t seem to be happening right now.” The world is a messy place.The world of carbon based life forms is a messy place. And whatever the capabilities in the lab of this technology, how it emerges into the real world is going to be a much slower and uneven process. On those two poles, where are you?Yeah, I mean I’d say look, I’m an AI optimist and I’m also an AI realist. And I think there’s two separate issues we’re dealing with in your question and the one before that. One is speed of diffusion, right? And if you look at expectations of how much impact AI will have on the economy, and you ask people in AI labs, some of them say it’s going to take us to 20 percent annual growth. I mean there’s one chart out there that says it’ll get us to 100 percent annual growth. Then you ask economistsMove the decimal point over.Yeah, just get rid of it even.You ask economists and the consensus to you is that, yeah, maybe it’s half a point of annual GDP growth. The economists have just seen too many past technologies run into friction in being absorbed into real life work. The diffusion rate, the dissemination rate, and the bottlenecking on the long pull on the last thing that can be automated.So in computer science we talk about something called Amdahl’s law, which is usually in terms of performance optimization. And if you have a piece of code and you can take 90 percent of it and speed it up dramatically, well the 10 percent you can’t speed up becomes the bottleneck. And sometimes that means you get very little actual speed up if it’s the thing that you’re waiting on.So in things that AI can do entirely on its own, I think we should expect incredible breakthroughs and incredible speed of progress. In things where the task is messy, as you said, and a human still needs to be involved where judgment is involved, where it’s a relational task, as one of Google’s post AGI economists, Alex Imas has this theory about relational tasks being a growth area for humans. In all of those I think diffusion will be slow.And that’s on top of my view that weekly held, but current view, that AI capabilities are likely to be jagged for the foreseeable future. I believe in superintelligence, I believe it’s already here. We already have superintelligence playing Go. We’re crossing into superintelligence in formal math. That doesn’t mean we’ll have super capable persuaders or super capable politicians or super capable city planners or whatnot.Roadblocks to an intelligence explosion (06:11)New ideas get harder to find. Right? You pick the low hanging fruit first. So I fully believe AI is going to have more and more of a role of helping to build itself, but each step will get more challenging.Certainly there are economists. You mentioned economists who study AGI. There are certain economists who think that’s a ridiculous thing to focus your scholarship on. A science fictional reality which may never occur. And even if it should, it doesn’t need a special economics. We have economics and the tools will be perfectly fine for whatever advances we get.But certainly the technologists will say, “you ain’t seen nothing yet.” You either you don’t understand the technology or you’re willfully blinding yourself that everything is absolutely about to change. Is that what they really think? Have I described what technologists think, even the ones who are worried about “it’s going to kill us all” still think this is going to be an amazingly capable technology and even they think it’s a ten percent chance it’s going to kill us all, the ninety percent will change every aspect of our lives and will be an inflection point in human civilization.Yeah, I mean I think we should expect that AI helps us increase our rate of progress in things that we care about. There are many things that we care about where progress is much slower and much more difficult than we’d like. In pharma, we have this thing called Eroom’s Law. Eroom’s law is the opposite of Moore’s Law. Moore’s Law says computing power doubles roughly per dollar every eighteen months. Eroom’s law says that over some period of time, the amount of money you have to spend to design and develop a new drug gets ten times more costly.So we already have these steep diminishing returns to research. And so AI should help bend some of those curves and accelerate how fast we find cures to diseases, develop new materials, etc. So I see it as finally very good for humanity. But I think there’s a particular assumption in the “my gosh, the singularity is scheduled for next Tuesday” forecast or for, March third, 2027, I can tell you it’s the singularity, whatever viewpoint you have. And that is the generality of AI going to one hundred percent.That is that AI will just be able to do every single thing that humans can do better than humans. And I don’t see strong evidence of that.What we see is more like logistic curves where AI, at first can do almost nothing in a certain domain, then it gets better really fast, and it gets to where it can do fifty, seventy-five, whatever. And then it all often struggles with this long tail of stuff that humans for whatever reason just understand better. So long as that’s the case, that in some domains we care about, many of them, humans still have unique value to add, then humans remain a bottleneck. I think that prevents the fully recursive self-feedback loop from exploding with an upward asymptote to infinity.Yeah. I think to the extent that this filtered to regular people, I think the message they’re getting is these programs are really good. The thing they can right now do really well is coding. And therefore they will be very good at that. They will make themselves better. And then there’s this amazing speed of thought and speed of artificial thought flywheel, and once that kicks in, it’s probably going to kick in fairly soon. We’re talking about amazingly capable models over a very short period of time, of both human level and beyond human level capabilities.And if we get lucky, they won’t kill us. And that is the path that we’re on, and that is widely agreed, that is the path we are on. So whether it’s the end of this year, March of 2027, or April of 2028, does it really make a difference? ‘Cause that’s where we’re going, and there’s plenty of evidence that’s happening.I’d say it’s easy to do thought experiments and say if this is true and if the feedback loop really doesn’t have a lot of dampeners in it, that you can draw a plot on a curve. The science fiction writer Vernor Vinge did this and said it’s going to accelerate itself, right? That theory of the singularity is that you make an AI that is at least as good as humans at coding, and so that AI can now make a better version of itself, and the next version can make an even better version, and critically, that there’s no dampening of that loop. That each time you make a better AI, the amount of improvement it can cause is at least as big as the last step. In the real world, and this is a blog post that I had on Noah Smith’s blog two weeks ago, we see incredible dampening. We see that a better AI, currently, in conjunction with humans, as the productivity tool for humans, has incredible diminishing returns.OpenAI released this data showing that their researchers were using 124 times more tokens than they were in mid-2025. Okay? That led to seven times more lines of code being written. Well, one of those is smaller than the other. Okay? That led to 1.6 times as many experiments being run per researcher. And then what did that lead to in terms of rate of progress? It’s not clear that accelerated the rate of progress at all.OpenAI and Anthropic will both off the cuff, in front of the press, in loose situations, make claims about AI starting to build itself. Anthropic’s actual system cards that they put out with every release say things like, “We don’t see signs of strong uplift.” Anthropic said, when Mythos came out, we think it’s increasing the productivity of our team by about a factor of four. We think you would have to do ten times more than that. A forty times productivity increase to double the rate of AI progress. And then on their next card they said it’s clear to us that the productivity gain that we get from using AI has been key to maintaining the current pace of progress in all these domains, whether it’s drug discovery or nuclear fusion or AI.New ideas get harder to find. Right? You pick the low hanging fruit first. So I fully believe AI is going to have more and more of a role of helping to build itself, but each step will get more challenging.AI in the short term (13:34)I think in a few domains the world will be totally different. Jobs will be totally different in the future. Jobs have already changed for software developers. You almost never write a line of code yourself now, but everyone’s an architect.Assuming that what you’re saying is right, does the world of 2030 look utterly comprehensible to us? Or is it something different? Because it’s still this new technology, very powerful. It is improving. You figure businesses will get better at using it. Is that enough? And I guess when I was thinking about this question, I’m like, is all the money going into data centers? Is the progress you’re describing, is that good enough to justify that investment?I’d say there’s wide variance and I’m giving you my views, but let’s all update on data. And so I think there’s a lot of data we can actually ask the labs to give us that will help us understand this phenomenon. And that’s something I mentioned in my blog post. The Elasticity Institute has a set of requests. A paper that came out last week that had some of the godfathers of AI on it also had a set of requests for the labs to release data and that would help us ground all of this more in reality or empiricism, right?Without additional data, the best data we have is what OpenAI put out a couple of weeks ago. On the basis of that, I think in a few domains the world will be totally different. Jobs will be totally different in the future. Jobs have already changed for software developers. You almost never write a line of code yourself now, but everyone’s an architect. Everyone now tells the AI what they want done and reviews what came back and changes what the AI is doing and translates from messy human requirements of what the customer wants into something more structured that the AI can understand.So the job has changed, but software development jobs have not gone down. They’ve gone up. Mathematicians are suddenly feeling the angst that let’s say graphic artists felt when DALL-E came out. “Am I still useful?” And math is changing as we speak at a pace that is freaking out mathematicians, but it’s fundamentally probably really good for the world to give us greater understanding. But how much is this going to change the job or the life of most people by 2030? Not much. I mean you will have everyone who will have a halfway decent, maybe better, personal assistant. If you’ve never had an executive assistant, it’s no longer just a secretary, executive assistant in a corporate setting, it just does all kinds of stuff and helps you cognitively offload in a good way. Complexities like, yeah, like get me that flight and that hotel room and make sure I get there in time to have dinner with John etc.We’re all going to have assistance like that you kind of can have right now, but I don’t think the world’s going to be radically transformed. In terms of your other question of all this capex, like are we in a bubble? Can it be justified? The world’s total spending on software and IT, just securing your network and managing it inside offices and so on, is a couple trillion dollars a year. It’s one to two trillion in software development and another trillion or more in IT.So that alone gets you really pretty far. And I think one thing we’re seeing is that, in a lot of places there’s high demand elasticity. If you make it easier to write software, that 1.4 trillion, let’s say, that the world spends on software today isn’t an upper bound. It’s just what we spend. When the cost of making software plunges, the world might have more appetite to make software for all sorts of niche scenarios. I showed Replit, which is an easy way to make apps with just a few prompts, like a four or five sentence prompt, to a friend of mine who runs a home inspection business.And her eyes lit up with all the things she could make to help run her own business where it’s not a big enough market for there to be big software companies going after like home inspection software. She would never pay a developer to do it. But now she or like she and one of her more technical employees can go make this stuff for themselves and they’re paying a little bit of money to an AI company, and suddenly their productivity in the real world has gone up. They spend less of t…
Full show notes at the publisher 
Our TOPPODCAST Picks 
Stay Connected
