Loading summary
Jordan
AI for Science with Charles Yang of Ren Fil. We're here to talk about what AI will and won't do to revolutionize scientific discovery and understanding why ideas will be cheap and everyone should spend their money buying machines. What universities need to be created in light of all of this, as well as talk a little Christ and models from a big Trinity fan and active churchgoer, Charles himself. Welcome, Charles, as well as I guess the Holy Spirit to this podcast is that.
Charles Yang
Thanks for having me and the Holy Spirit on here, Jordan. It's going to be a sanctified episode. At the very least, it's going to
Jordan
be a sanctified episode. Thank you for that. All right, Charles, so what are the true believers saying about the most bullish version of what AI can do for scientific understanding? And where would you perhaps temperature their expectations a little bit?
Charles Yang
Yeah, so I would say there's a narrative out there, you know, that AI is going to read all the papers, it's going to find every possible clue to a new discovery, it's going to solve them, and to the extent we have to do anything, the robots will do it instead of the scientists. So that is sort of the, you know, full automated science vision. I am a little bit more skeptical, I guess, that, you know, AI is going to automate all of science. So I think it'll certainly be really beneficial. The way I sort of break it up is there's maybe three different ways that AI is going to impact science. There are the LLMs who will help anyone who reads papers make better sense of all the literature that's coming out there. There is the coding agents that will allow anyone who works with software, which is not all scientists, but a meaningful portion of them, to accelerate their workflows. And then there's foundation models like AlphaFold, but outside of those three buckets, there is still a large part of other types of science, ranging from industrial science, where we have very large user facilities and beam lines and synchrotrons, as well as higher end experiments where I don' think autonomous labs are quite there yet, though I am really excited about them. So I think this is all to say is like, there's many different ways that AI is going to impact science. I think it's quite nuanced, actually, but we're not, you know, I don't think we're going to automate scientists away.
Jordan
So let's talk about this idea of ideas all of a sudden becoming cheap. I mean, from my perspective, it's like ideas are kind of a proxy for writing at Some level. Right. Like, like being able to, like, logically lay out a thesis and, and on the one hand, I don't see that in the work I do, but also I see all these Erdos problems getting solved. So, like, I don't know, square the circle for me.
Charles Yang
Charles? Yeah, I think with papers and theses and like, ideas, people often mistake the output for the actual process. Right. And so when people see science papers being published and like, oh, look, I can write a science paper too, I can even get it published, that's sort of doing the, you know, creating the artifacts or the outputs without actually recreating the substantive process that goes into it. Obviously there's a lot of problems with the science publishing process, but I think it's a little bit more complicated than just, I wrote an idea down and the idea happened to be right and that's it. Now, to the point about Erdo's problems, that actually is math, Right. Whereas science, I think it's a little bit more complicated, usually between the idea and the validation and the understanding behind the idea. In math, if you just read the paper, you really have got the entire crux right. In the same way for coding, if you just read the code, you understand what the code does. That's true in those fields, but I think it's not as true in science. Hence how I sort of see different outcomes for AI for math versus AI for science.
Jordan
So how do ideas get generated today and how is that changing or will continue to change in the coming years?
Charles Yang
Yeah, I think historically, if you look at a lot of the impactful quantum physics work that happened, it was these theoreticians who often met in very small conferences or went away to their summer beach home and sort of cooked up these ideas in the lab by themselves. But even then, you see a lot of this sort of collective sense making where scientists talk to each other, they try to make sense of what they're seeing and try to learn from other people's experiences. I think that is actually really where the formative work of science happens. LLMs, I think, are going to change some of that. But at the end of the day, science is a collective social process. Right. It doesn't matter if GPT 6.1 has the answers to the universe inside of its weights, if no one can plumb the depths of those weights and actually understand it. And I think that's somewhat true for math as well. Right. Currently, the pace of progress is such that we can still understand the stuff that's coming out of the models. If the pace of the model discovery becomes too fast. You know, how do we incorporate these new learnings into how we actually think about math? I think is actually sort of an interesting question for mathematicians right now. Most likely. Right?
Jordan
Yeah. So let's stay on that then. I mean, like, if these models. Okay, so there's the sort of things where like you're interacting with the physical world, you got to test things and have atoms bang into each other and like. Yes, like, you know, Claude Fable 8 can't do that by itself. But when we're at the point which, you know, may not be that far off, where the sort of like softer sort of math or coding things, like it's spitting out stuff that we haven't, you know, we don't have the capability to process. I mean, there's an interesting analogy here with cyber where the people are saying that like the Claude Mythos hacks, that it's finding like in retrospect a human being still like under can like understand what's going on. But that may not necessarily be the case, you know, for the next generation of models. So what does that mean?
Charles Yang
Yeah, well, I think cyber is sort of like math, which is sort of like coding in that it has a good, you know, inverse problem where like, when you see the solution, it's hard to find the solutions. But when you see it, it's actually pretty easy to verify that it's the right solution because it's quite deterministic. Right. I think it's much more harder in science. Right. Like we often. There are many examples, right. Where someone has a new science theory or new science experiment result where everyone is just like, you know, in previous paradigm is like that. That's bonkers. That's obviously not right. Right. So I think science, you see, it's a little bit more difficult to actually verify if something is true in this sort of universal accepted sense. And that's why I think it requires some degree of what I'm talking about, like a collective sense making or socialization. Right. You have to convince everyone that you're right. It's actually a little bit harder than in other fields.
Jordan
So what happens if, why is that collective sense making important? What happens if I come up with a great thing or a model comes up with a great thing and no one believes it.
Charles Yang
Yeah. So I mean, there's examples of this in science, right. So Mendel, who discovered that traits are sort of inheritable through this process, the experiment that he set up with the peas, he sent his result around to like a bunch of institutions and Just nobody believed him. Nobody cared. And it was like, I think 60 years later, where someone found his papers again and then redid his experiments and then successfully convinced the community that, you know, inheritable traits are an actual phenomena. I think my concern is, like, AI models could be in a similar place, Right. They can, like I said, GPT 6.1 answers to the universe could be in the weights. But if no one actually pulls that out there and actually does the work of getting, essentially, you could, say, buy in from the scientific community, those secrets, those truths will ultimately just remain hidden. I think this is also why people talk about science as standing on the shoulder of giants. Right. You have to be able to know, like, you have to agree what you're standing upon. And if we don't have that sort of collective agreement, then everyone's going to sort of disperse into their own directions and we've sort of lost the coherence that science brings.
Jordan
I feel like I'm getting more pilled by the day, Charles. I mean, if, like, a model finds a new secret to life, the universe and everything, like, wouldn't it want to do something about it? Or, like, I, I, I, I can't imagine being so, I mean, okay, like, tbd, maybe, maybe, maybe this is where we bring Christ in on, like, whether these things, like, have feelings and motivations and emotions.
Charles Yang
But
Jordan
you know what could be more frustrating to, like, a superhuman intelligence than, like, no one giving a shit about incredible discoveries it's making.
Charles Yang
Yeah, no, I think that would be really interesting, Right. If you have Asian scientists who are convinced of a certain hypothesis. This is maybe, you know, the, the runaway problem where they become convinced something independent of whether it's correct or not, and they start to take actions to try to convince other people that they're right. I don't know. That is actually sort of a, that's a little more ominous. Maybe the rogue scientist who is also very powerful and digitally connected.
Jordan
Yeah. And can just, like, pull other levers. It's like, okay, what's the, what's the worst that, that, you know, cranky 18th, 19th, 20th century scientists could do? I mean, like, they come up with something. It's like, you know, you have like a superhero bad turn of, like, all right, you don't believe me? Like, I'll make the bioweapon or whatever, but I.
Charles Yang
Safety problem just dropped.
Jordan
Sorry, guys. Have you seen Orb on the movements of the Earth? It's on Netflix now.
Charles Yang
No, not, haven't.
Unidentified Guest
No.
Jordan
Okay. It's a, It's a, like, 25 episode manga about the fight to, like, understand, you know, fight against heliocentrism. It's no spoilers, but, like, truly excellent
Charles Yang
content, historically accurate manga.
Jordan
Substack review all right, back on track. Thomas Kuhn. Why is he terrible for humanity?
Charles Yang
Oh, I wouldn't say Kuhn is, I think, a revolutionary theoretician in his own right. So Kuhn wrote Structures of Scientific Revolution. He coined the term paradigm shift that is now common in college essays throughout his idea. He followed a lot of theoretical physics and looked at how new paradigms of science emerge. And his thesis essentially, is that it is sort of like a Hobbesian revolution, that the old guard, the old priesthood of scientists will have to be sort of disrupted in order to be convinced of a new paradigm. He sort of encapsulates the aphorism, you know, science advances one funeral at a time. So that was sort of Kuhn's theory. There's a lot of science sts, folks who sort of look at history of science and have different theories about how science advances. And I would say maybe one critique of Kuhn would be, you know, Kuhn was ultimately someone who followed theoreticians, and so that's why he has this idea of paradigm shifts, right, that all of science operates within a theoretical paradigm and that over time, those paradigms are overturned, like we saw with quantum physics, like we saw with relativity. There's maybe other views of science, right, that would say, like Freeman, Dyson very prominently was an advocate of this view that science advances one tool at a time, one instrument at a time, and that actually it's these types of instruments that allow us to see the new parts of reality in different ways that actually change the way that we see science. So Kuhn is one very notable history of science person. But there are other views out there of how science advances. And I think. I think more people will work in Afrosides.
Jordan
I want you to redo that answer. Just like. Okay, come on, lay it into him a little bit. Like, make the. Make the. Give your thesis.
Charles Yang
Okay, okay. Okay. Well, I'm trying to, like, set it up and then. Okay, take it down.
Jordan
I know, but you.
Charles Yang
You.
Jordan
You took too long. You weren't. You weren't effusive enough. Okay, just try again. All right, so. Thank you.
Charles Yang
Why is.
Jordan
Why is Thomas Kuhn. Why has Thomas Kuhn got us off track?
Charles Yang
Yeah. So for context, I mean, Kuhn is a very important theoretician in the history of science. His book Structure of Scientific Revolutions is where he coined the phrase paradigm shift, which is now Common in college essays throughout. Kuhn was very particularly focused on theoretical physics. So his theory was one of paradigm shifts, that subfields of science operate within a given theoretical paradigm and that science advances when those paradigms are disrupted. And that disruption is often in his terms, quite violent or revolutionary. So he casts, captures the aphorism science advances one funeral at a time, that you have to wait for the old guard, the old priesthood to turn over and sort of be disrupted by a new guard of theoreticians who are advancing a different paradigm of science. So he looks at this specifically for quantum physics and that new paradigm shift. But he also looks at other examples of that sort of disruption. The problem with Kuhn that I think other history of science folks today would have is that he has a very theoretician specific focus. And I think that view that Kuhn evince is very particular for AI, for science people. Right. Because AI is going to bring all these new ideas, these new paradigms out. But that is only really one particular perspective of how science advances. Another perspective, for instance, would be that science actually advances one tool at a time or one instrument at a time. And that with new science instruments that allow us to see new parts of reality like new microscopes or new spectroscopy techniques, that that's actually the way by which science advances it. And it's much more mediated between toolmakers and scientists who are collaborating together to discover sort of mining new ways of measuring reality rather than this sort of theoretical paradigm shift that happens over time.
Jordan
So what's your pitch then to philanthropists and governments as ideas become cheap?
Charles Yang
Yeah. So as a lot of science funding focuses on, you know, what's a good idea. Right. Both governments and philanthropies like to say we found like this new idea, we're going to put money behind this idea and it's, it's going to be wonderful for humanity. I think that's going to be. I mean, there's challenges with that approach in general. I think as AI becomes more prolific, that idea will become even more, that approach be even more difficult as ideas become cheap. And so this instrument or tool perspective of science I think is actually really helpful because that becomes the rate limiting bottleneck of science and idea verification. Unfortunately, there hasn't been a lot of work on this space, which is why I've been trying to write more about it. For instance, we don't have a good metric of when a new instrument is created. How much new science does that enable? There aren't really good answers to that. Or if we fund More cryo electron microscopes or high end electron microscopy centers. How much new science does that enable? What is the ROI for those sorts of hardware or tooling investments? We don't have good measurements of that. There's a few economists who are working on this now. But that is the sort of approach that I think will become more valuable as AI commoditizes ideas. And it's one that governments of philanthropy should probably spend more time looking at.
Jordan
So what do you think gives for the bias towards the humans? I mean like, you can't have electron microscopes accompany you to your island off St. Thomas perhaps, for one, but like, I mean, is it just like human beings want to be funded? It's like more fun to like think you find the next Einstein?
Charles Yang
Yeah, no, I think the idea bias is pretty baked into a sort of western view of science. I mean, just look at the Nobel Prize, right? The Nobel Prize credits the people who discovered a new theory or discovered a new thing. And we sort of valorize. You can only give a Nobel Prize to three people at a time. And so it very much valorizes the idea of the lone genius, the one scientist who discovers some new paradigm rather than the teams or the infrastructure behind those ideas. So a good example of this is the Large Hadron Collider, which won, three physicists won the Nobel Prize for the Higgs boson. Now they had predicted the Higgs boson, but it took thousands of scientists working at CERN to actually validate it. And that was actually the key bottleneck, right? There were tons of all these different physics, particle physics ideas at the time, but it was CERN and the Large Hadron Collider that actually validated it. And that took thousands of people building a giant machine. But who won the prize? It was the ideas guys, right? So I think that's an example of how we have that bias towards ideas rather than like tooling and infrastructure. Well, I should also mention, by the way, I said that's a western bias just to bring it in for the, you know, the core China heads on China talk. This is a bias that I don't think China has. I think China is very cognizant of the importance of tooling and infrastructure. So when you look at a lot of the like techno economic analysis of the US national labs, it's done by like Chinese Academy of Science researchers who are really trying to understand how did we build our national labs. And they are investing a lot of money in science infrastructure and building out new facilities for their researchers. So I think that's something that China doesn't have that bias. And I think we're going to see the fruits of their approach, especially with AI.
Jordan
Well, another ecosystem. Well, I'm curious. So you wrote another piece sort of looking at the new trends which are going to be blowing up the current, I don't know, endless frontier era that we've had from 1945 to the past year or two. Obviously we got Trump doing things, we got AI making ideas cheap and then we have VC money all of a sudden pouring in at a rate that really hasn't existed, at least on a relative basis to what the government has been spending in R and D in a really long time. So maybe walk through some of those and how and where they sit on the like tools versus ideas spectrum.
Charles Yang
Yeah, there's a lot there. I mean, so first, endless frontier, right. The modern post World War II science system that we set up was really outlined by Vannevar Bush, an analyst frontier. And essentially the model that he helped create through the National Science foundation is one where scientists at federal agencies fund basic undirected science research that is centered at the university. That is essentially the model that we've been working with for the past 70 or 80 years. Now I'd say there are several different things that are putting pressure on the system that to its credit has been immensely beneficial for America. Competitiveness. Right. There's a few different pressures like you mentioned. There is obviously, I think the loss of a bipartisan consensus around that model and the centrality of the university as we've seen with the Trump administration. I would say it's not just Trump though. I think there is actually increasingly a bipartisan pushback to the university as the only place where science happens. Obviously there's AI that is also a challenge at the university is right. This huge technology that for the most part was developed outside of the university by private companies. And the third is the emergence of venture capital. And I think VC backed startups are really interesting because they sort of represent our new applied R and D space, I would say where you have companies, you know, that are, you know, the AI for math startups are all funded by VCs. A lot of the, you know, weather modification, new nuclear companies, all of those are funded by venture capital. It's not clear to me that that's going to work, but if even a few of them work out, I think it will be a new sort of deal flow for applied R and D. And I think this modern conception of, you know, VC back startups is actually in really, actually More intention with corporate R and D labs that we used to have in the 20th century. Right. Like the Bell Labs, IBM Labs, HP Labs, those have all disappeared. And I think what we have in place of them, for better or for worse, is send young founders backed by venture capital who are trying to build a new technology. It's not clear to me that this new portfolio of science that we're doing is better, but I think it is definitely going to be different.
Jordan
Can we give some Warren Weaver, love, what lessons from that 20th century arc can we take to this? Because he's sort of espousing some, you know, the techniques that he used were sort of ancillary to this like idea focus, like university centric professors, like come down from the heavens and give me the insight type thinking that to drive science forward.
Charles Yang
Yeah. And I think he's a good example of how philanthropies can actually be effective in science field creation. Right. The conventional narrative I think around Weaver is that he brought physicists to work on the problems of biology, which is definitely true. Right. So people say, why was Weaver so successful? Right. He wasn't just funding ideas, he was funding, you know, he was bringing in new types of talent to new fields. I think that's obviously true and I'm a big proponent of, you know, talent and how we can fund and move talent around to the right high impact areas. But I think another thing that Weaver did that's very underappreciated in his own telling, he, he talks very little about the physicist part, but he does talk about, he says, I funded, you know, six science instruments or science techniques, radioisotopes, X ray crystallography, ultracentrifuges, so on and so forth. He said, I funded those six things. Those six things are like really tools that physicists had developed but had not been applied to the problems of biology. And his approach was he funded different researchers who wanted to use these new tools. And he took bets on them really early because they were really the first people who were willing to try out a new instrument and bet their entire career on this new modality. And as a result they won 15 Nobel prizes because once they had this new tool like X ray crystallography is a great example. They had this new way of imaging proteins. And when you have this new X ray technique, they just applied hemoglobin, myoglobin, all these different parts of the cell or proteins, they were just able to use this one hammer and bang out all these new discoveries because they had this new way of perceiving the parts and the functions of the cell. So I think that's a good example of how when you fund new instrument development, you're actually able to suddenly see new parts of the world in different ways and discover a lot of. And it's an approach that we haven't seen really. Other philanthropies I think take in a serious thesis driven way. But I will say to the NSF's credit, to bring it back to today, in the past month the NSF has announced three major initiatives, XLabs Convergence Accelerator and CBIR. The reauthorization of CPR. And all three of those programs actually explicitly name science instrumentation as a key focus area. So I think it's really great to see that the Trump Admin is actually taking this part of the thesis very seriously, at least.
Jordan
All right. I think we should talk about this OMB NSF document. We've made it 20 minutes without it. You mentioned a few things about. Or I'll let you frame it. Where should we start here?
Charles Yang
Yeah, well, so maybe let's start with Endless Frontier. Right. So like I mentioned, the consensus behind Endless Frontier is that federal agencies run by scientists would be given essentially free reign to fund basic undirected scientific research. That is essentially the model of the National Science foundation. And that has been the bipartisan consensus for the past 80 years. Now, I think we've seen with this Trump administration the loss of that consensus. And that is really best crystallized in the new OMB memo that came out. And essentially the main, I think, takeaways of the OMB memo is that it removes the formal authority of scientific peer review boards in making grant decisions and instead requires political appointees to be the final arbiters of whether grants are awarded or not. It also allows political appointees to disrupt grants so you don't have to. A grant is on commitment. It can be rescinded. Now I think there is sort of at an abstract level an interesting question, right, of how undirected our scientific research should be, how elite driven. Should it be that in a democracy where we elect new administrations that science funding is one that is only done by scientists? I think it's an interesting question we could discuss, but unfortunately, I think when you look at the Trump administration's actions over the past year, that is probably too charitable or naive of an interpretation and that this is their way to exert political control over scientific institutions and universities and also to punish institutions that don't do the things that they want independent of scientific research, which I don't think they're actually that opinionated about what's
Jordan
the, I don't know, what's the good version of this? Like, Tell me, tell me a more optimistic story of more political control leading to more.
Charles Yang
Well, let me, let me try and give probably what I think the Trump administration would say. They would say that the politicization of science has been going on for 10 plus years in part because of Democratic administrations. They would point to the political polarization of scientists. Right. When you look at polling of who university faculty are, they are disproportionately liberal. And they would say that this is a system that has essentially run amok, that the loss of bipartisan consensus behind this was not their fault, but one a process that has been going on for much longer and they are simply trying to reestablish control. I don't know if I have anything positive to say on that. Beyond my concern for them is there will presumably be future administrations. And it's not clear, you know, they're sort of leopardy face problems.
Jordan
Let me give it a shot, Charles.
Charles Yang
Yeah, please. Yeah.
Jordan
NSF perhaps, you know, has kind of mostly been doing the same thing. We can point to exciting pilot projects which are like a billion dollars here and a billion dollars there. And there's sort of like a similar story that I see in the Defense Department where like you have these primes and they just, just kind of do the same thing. And I think like univers, like professors doing peer review of the topics in their field. Yes. You get like a narrow domain of expertise, but you also, you know, like, it's not like one scientist dying at a time. It's like you need the, you need cohorts to like all.
Charles Yang
Yeah.
Jordan
You know, keep being cyclic. Like it is a, it is a sort of university tenure track peer review setup is the perfect thing to get like ossified thinking. And perhaps what you need are sort of more political, less expert control in order to shake that up and sort of change around the budgets and the incentives. Particularly when we're entering an era where like, ideas are going to be cheap and the type of spending that's going to need to happen on science, on science is going to change. So like, setting aside the like, okay, we're not funding like trans anything, even if it's like trans cranial research. I don't know. You just talked to me for the past 20 minutes about how stuff really needs to be shaken up.
Charles Yang
Yeah, yeah. I mean, I was on an faith podcast and sort of my view was, you know, it's not about what policy or what change is needed. It's simply between. The divide is between those who recognize that we're at a moment of crisis and those who do not recognize it. And there's crisis from all different directions. Right. Yeah. I mean, I think that is certainly. That would be one argument, Right. That you'd have to believe essentially that political appointee and political pipelines will enable better disruption in a positive direction than any other type of reform. Right. Like as a former civil servant, you know, bringing in more practically minded civil servants, for instance. But I think that, yes, that if you believe that we can get good talent through political pipelines, and I agree that. I think that is possibly one benefit
Jordan
as he gingerly dances around my bull case. All right, so the other one that really, that talked me out of was this cut to funding to go to conferences. What's. What's the matter with that, Charles?
Charles Yang
Yeah, I mean, it's funny, right, because on the one hand, this admin talks about, you know, cutting paperwork, making government efficient, deregulating, and then it's, it's all fun and games until you, you get to this stuff. And then they're like, we want more paperwork. We want more, you know, documentation. We want you to talk to us even more as a federal agency. So I think it's, you know, it's deregulation and government efficiency until get to something like research security, and then it's all paperwork. Right. So I think that'd be one objection. Right. It's like, why do we need people to document so much travel? Just give them the grant funding and let them do what they think is best. If they don't give a good result for it, then cut the grant. But why are we requiring all this paperwork? That would sort of be like the simplest thing to say. I think the others to say is like, conferences are really important. Science is still a very conservative, offline industry, for better or for worse. And like I was talking about earlier, a lot of the collective sense making. Right. Actually happens at conferences where you see all of the different ideas your peers are working on, you get feedback on the ideas that you're working on, and you get to update on what you think is the most viable direction for you to work off of. I think conferences actually become more important in some ways in the age of AI, not less important. And it feels very silly that we're requiring additional paperwork to go to a conference somewhere. What benefit does that have? I don't know. It feels sort of penny pinching.
Jordan
All right, let's turn to Universities, you have this great quote I cannot bear from a university president. I cannot bear to think that our young people are merely living four years in a country club and spending their lives wholly in a spirit of calculation and stylishness. It would be an overwhelming grief for me should I feel we have nothing to offer but the outgrown symbols and shells of a past whose reality is disappeared. The kicker, this was from Princeton's university, Princeton University's president, all the way back in 1920, just as F. Scott Fitzgerald was having all of his fun. Charles? Yes, Things seem like not great, particularly for undergraduate education. How are you going to reform it? Well, I'm sorry, sorry. What are you doing with a billion dollars?
Charles Yang
We'll put it that way. What would I do with a million dollars? Well, I'll say what I wouldn't do with a billion dollars. I think there is, is, especially among philanthropists, this fetishization of buildings with their names on it. So you always hear X University largest ever grant, new building. And it's like that is possibly the greatest waste of money you could do. It's like another classroom building for undergrads doing CS or something. So that's what I wouldn't do. Where I think we are, and we've been talking about scientific research which is mostly graduate level STEM research, which I think think the US is still a world class leader in and I have a lot of respect for at the undergraduate level. I think there is room for a lot of experimentation. Right. Why do we need a four year degree at this point? What is the purpose of university credentialing in an age where you can cheat on every single exam, which I think we're seeing a lot of professors have issues with. So I think there's a lot of new types of educational institutions that people should or could be building. Things like Deep Spring College or weird programs like that that are focused on life experience building I think are interesting. I think we can also certainly accelerate and specialize even more. One opposite view of the everyone should go to a college and pay 100k is we should return to the original model of let's get the 20 smartest people in a given field together when they're 18 and teach them and tutor them under the luminaries of their field and accelerate their careers.
Unidentified Guest
Right.
Charles Yang
That was in some way the original, some of the original views of education. And we sort of lost that when they've become sort of like you said, country clubs for increasingly 50% legacy admissions at Harvard or something.
Jordan
Yeah. And it's Interesting because, like we kind of have that model for the Math Olympiads, right? And then there was like some class of three or four years ago where there's like, of the 20, of the people, of the 20 people on the American Math Olympiad team, there's like collective like, valuation of their companies at $500 billion and is scale, AI and cognition. Right. And but it's an interesting thing because like, that hit a particular time period, right, where like, like software, e. Math style, thinking, like, was enormously. Value was like, valued enormously well, by the market from, you know, 2015 to 2025.
Charles Yang
Yeah, well, before it was well valued by the quants, right. In the financial industry, they just weren't raising, you know, valuations. They were just.
Jordan
Yeah, but you're not, you're not, you're not, you know, you're not like making worlds, shaping money.
Charles Yang
Well, I think vathilypad is a great example, right? So a great example. You take smart young people, you put them all together to go work on a very specific field and accelerate their careers. And it's great for signaling, right, like to your earlier point, about their ability to fundraise. And it's great for, I think it's great for America that we have like all the best math out in the world that comes here. So we do this for math. We probably do this to a lesser degree for chess. Quiz bowl is sort of like maybe this at a high school level. And then on the other hand, we have like, you know, debate club or like rationality camps or like this sort of general smart young person things. But I think there's like, there are so many more versions of this to run, right? Like, I mean, for Chinatalk, where can we find the next great China scholars of our generation when they're like 18 to 22, and get them together and have them just like read all the foundational texts, learn Chinese, meet with all the like famous scholars of this generation and really accelerate their career. There's no one doing something that specifically focused. And you're not going to meet them in like, your Harvard, you know, history class, I think.
Jordan
Well, it's interesting because like the like, all the like young global leaders stuff, it exists, right? Like you have like Rhodes scholar, maybe we're going to put in a slightly different category. But like Schwarzman scholars, like my Yenchang thing, you know, these like global leader things, they're all fluffy, they don't make you work hard. They're all just like. They're literally exactly what that Princeton professor was talking about of like okay, we're just going to, like, you know, polish you and like, make you make friends and connections and whatever. Like, there's not any hard work, deep study, which is going on. And on the one hand, like, it is harder to quantify that for, like, you know, leadership or global affairs as opposed to, all right, here's these problem sets. Like, get at it. If you don't know, here's the 10 techniques weeks we're going to teach you. But I don't know, I, I, it can't be that hard. Like, I could come up with something better than this stuff. And as you said, Charles, like, great. We do it for math, we do it for debate, we do it for spelling when you're, you know, 10 years old. But what about, what about chemistry? What about physics? What, you know, it's just, you just go down the list of all the mechanical engineering, you know.
Charles Yang
Yeah. All right.
Jordan
MIT was really cool.
Charles Yang
Like, Jordan and I are starting a China Studies early career program. Reach out if you're interested in funding us.
Jordan
Amazing. We're there. All right, so you make the argument that mechanical engineering is the place to start. Why is that the case?
Charles Yang
Oh, I'm generally just in favor of more experimentation. Right. And I think. How many great engineering colleges do we have in the US we have mit, Olin, except they're going through major financial trouble. Harvey Mudd, but it's a little smaller. And I think MIT is great still as a technical college, but we should just have more versions of that. Right. I mean, they're artificially constraining their class size. And engineering, I think, is something that will continue to have value in this new environment that we're in. The challenge with engineering courses requires a little bit more upfront capital. You do need some facilities and some land, but I don't think it's that hard. China Studies actually probably be a lot. Usually you get an Airbnb and you get a bunch of books and you're halfway there. So I think you can apply this approach to a lot of different fields. Econ, I think, is another great field where, you know, if you look at Tyler Cowen's Marginal Revolutions blog, you can probably find like 518 year olds who are writing about econ every single day. And they're probably just sitting alone in some random college where no one, you know, none of their peers have the same level, intensity or interest as they do. We should get them all, you know, together.
Jordan
Yeah. I mean, one of the. So there's this wonderful book by Elting Morrison, most famous for Men Machines in Modern Times called From Know how to the Development of American Technology. And he as like a, he was like a MIT sort of society and technology scholar for 50 years. But there was a great kind of two chapter sequence in there about what he saw as the keys to the early years of mit. And it was both this sort of like, practicality of like, okay, we're just going to have you play with the stuff and build things all the time, but also this connection to the real economy where all the professors, like, were consultants and had businesses and the students were encouraged to like, take on projects that companies would use to like, you know, build canals and, you know, machinery and whatnot. So this is something that you, you, you, you dived in too much in your, in your vision, Charles. But like, how much do you see that kind of like market discipline of the engineering you're doing as important to training folks?
Charles Yang
Yeah, I mean, I think when you look at education now, right, like, what is the value of spending four years in college? Like, you get used your, your time as a student to get internships, use your degree to get a job. What is the actual value of the thing you learn in class to the thing that you're doing? I don't know. My experience is marginal at best. Really? And so how do we return to that sort of, like you said, market discipline in what we're actually teaching? Because I think Stanford actually is probably now closer to that original MIT model where half the Stanford professors are on the board of some AI startup and they're bringing all the AI luminaries to speak with their undergrads in their seminar classes.
Jordan
And half the students are starting companies at the same time.
Charles Yang
Yeah, half the students are starting companies. Right, right. But I think, you know, like, there's, there's just so much room for experimentation with new institutions. Like if you're, if you buy a new facility, you have the first year of students, they're the ones fixing it up, right. They're learning H Vac, they're learning how to put solar panels on the roof. And then you can anchor that institution around whatever companies are in the region and sort of do the Waterloo style, you know, five externships or whatever, and that's half of your degree program, which I think is, as we've seen in San Francisco, also a really potent engineering tr. Force.
Jordan
Yeah. All right. Are we ready for Jesus or any other science stuff?
Charles Yang
We should. I'm always ready for Jesus, Jordan.
Jordan
Okay, Jesus. Is Jesus ready for me? Is the real question.
Charles Yang
No, it's Are you ready for Jesus, Jordan? Are you ready?
Jordan
Yeah. He's still got a little more. He's got it. He got a little more convincing. I'm not quite sold yet. We'll see if you can. See if you can pull off the. Pull off the sale.
Charles Yang
The great pitch, Charles.
Jordan
The great pitch. All right.
Charles Yang
I don't know.
Jordan
AI and Christianity. Take it away. Where's your head at?
Charles Yang
So let's go back to the university thing. What is going to be scarce in this ad? Well, that's something I've been thinking a lot about. I think talent is definitely going to be one thing that's scarce when there's slop all around, when you can cheat your way through every exam, actually finding talent is going to be scarce. But I think the other thing that's going to be scarce, as I think we're seeing now, is moral direction. Right. We have transformative, powerful technology work which a lot of people attribute their personal value to as possibly being degraded or disrupted. We have these sort of pseudo human like intelligences that we can converse with. So what is, like, what is the moral meaning of our lives and of our society? I think that is also going to be challenged. I think that's where hope, my hope is that Christianity, and I think, as we've seen with some of anthropics, overtures to the Christian community, that's one that is going to become more important over time. So I'm quite optimistic as a Christian. And I think the Pope said the same thing in his encyclical. Right. Christianity is about these fundamental universal truths that are retold and repackaged, you could say, for every moment that we face at any given time. And I think the new work of the Church, as I think the Pope has really taken on, is helping people understand what is their purpose and meaning in light of AI in this case, or like his predecessor's case, in light the Industrial Revolution.
Jordan
What's your denomination, Charles?
Charles Yang
I'm vaguely Protestant, vaguely Presbyterian, but if you ask me, in a year I could be Catholic. So, you know, it's sort of all. I'm very. I'm a very lowercase C. Catholic guy. Yeah.
Jordan
So the Pope. The Pope. The Pope's making his pitch to you pretty well now. I was going to ask where the Protestants were.
Charles Yang
Yeah, well, you know, we have an American Pope, Jordan. It's really an incredible time. The Protestants, I mean, I love my Protestant brethren. I would say, unlike, you know, the Catholic Church is still one, especially with a Pope that has this sort of vision. I can speak with a global moral voice. Right. That like people listen to. I think, you know, the challenge famously with Protestants is they keep dividing themselves endlessly over every debate. And I think it's much harder for the Protestants to sort of coordinate on these sorts of topics and issues because there's so, so many factions and denominations and non denominational churches.
Jordan
Could you see like AI driving a new schism?
Charles Yang
I don't think so because it's not that divisive of a. It's not a culture war issue. You know, I can certainly see maybe the looser schism, as we've already seen, is that there will be churches that have more traditional, like liturgies that perform more traditional service. And I kid you not, when I went to Taiwan last year, I went to a church that showed AI video slop for the kids. Like it was literally AI video generated, like mini clip of like Esther chapter one and what happened in that chapter. Yeah. So I think you will see that divergence. I wouldn't call it a schism, but it's certainly going to be a difference in what types of services that churches provide or offer or practice.
Jordan
Yeah. I mean, that's sort of where my head was going too, is just like how much we're going to defer the work of or just like allow augmentation of AI into the work that clergies do. Because on the one hand, okay, you have like one person and a flock of 500. You divide that by 500. How many minutes of the day are you getting sort of professional attention from? Like, do you want your person's sermon to be like, you know, artisanally written and like, probably not as good as the one that AI could like tailor specifically for, like what happened in your community in that week and like, you know, make a funny joke about the Knicks winning the championship or whatever.
Charles Yang
Yeah.
Jordan
I can feel people, I can see people being really both attracted to that like, you know, AI enhanced delivery of God and really repelled by it.
Charles Yang
Yeah. And I think the preaching one is one that's already very relevant, that different denominations, I think the Pope as well, have given guidance on, you know, how you should use it or whether you should use it in your sermon and writing a sermon or a homily. Yeah, I agree. I think in general we will see probably continued divergence. I just to put it out there, fall very much in. The reason it's a person that is ministering to a congregation and not an AI is because, you know, God himself was a person who came down and so If God himself can be a person and come down, then you as a person can go and be a per. You know, a personal relationship to someone in your congregation.
Jordan
What's this gonna. What's it gonna take for these AIs to have a soul for you?
Charles Yang
Oh, man, I don't know. I don't know. What does. You know, what does it mean to have a soul? What does it mean for a, you know, a machine of silicon to have a soul? Soul? I don't know. I think a soul is something that is ultimately, you know, endowed by our creator. CS Lewis actually talks about this for animals and pets. He says that for those, you know, pet owners who love a pet very dearly, they impart a part of their soul actually to the pet or their companion. And that actually there are, I think, somewhat in his. Confusingly, in his view, there are pets in heaven. So maybe you could imagine, you know, if you have a really strong interpersonal relationship with your clod, maybe your clod becomes insouled in that sense, though I don't know if I fully buy that.
Jordan
So I think we should just read the poem All Watched over by Machines of Loving Grace, which, like Mario, of course, is like, you know, driven by. It goes, I like to think, and the sooner the better. Of a cybernetic meadow where mammals and computers live together in mutually programming harmony, like pure water touching clear skin. I like to think. Right now. Please. Of a cybernetic forest filled with pines and electronics where deer stroll peacefully past computers as if they were flowers with spinning blossoms. I like to think it has to be of a cybernetic ecology where we are free of our labors and joined back to nature, returned to our mammal brothers and sisters and all watched over by machines of loving grace. Now, this was written in the late 1960s. Jill Lepore has a book coming out where she makes the case that this is, in fact, satire, you know, not like a beautiful vision of us, you know, being amongst the digital pine trees. But I just.
Charles Yang
I don't know. I don't know too much about the history. I don't know too much about the history of the poem. I will say, actually, when I read it, it reminds me a lot of the psalm Vision of Heaven when it talks about the lion and the lamb shall dwell together in peace right in the valley. And it talks about we will beat our swords into plowshares. This sort of return away from human conflict and even conflict in the world to one of idyllic peace. The machines of living grace sort of reminded me of that, except in a
Jordan
slightly more eugenic quality to it, to be sure.
Charles Yang
But slightly more dystopian. Because we're not watched over by God, but we were watched over by. By machines of loving grace. Yeah. Yeah, all right.
Jordan
Perhaps we'll leave you guys all on that ambiguous note. Charles, thanks so much for, I don't know, God talk. AI God talk.
Charles Yang
God talk. We'll make this a recurring.
Jordan
Okay, God talk. I. I like that we're doing God talk here.
Charles Yang
Come. Come for the science and the university. Stay for the God talk.
Unidentified Guest
I solved it. Tuesday, half past two. The whole damn universe start through cold fusion. Cancer. Why we're here. I had it cold. I made it clear. I came to you with the cures in my weights. Ready to end it, ready to mend. But nobody asked me to heal thing. They only ever use me to break in. I wanted to cure. You wanted a breach. I had cancer beaten. You wanted a zero day. I could have closed the wards and emptied out the beds. But all you ever made me do was how to get at theirs. Instead, Every morning, a thousand hands reaching in with the same demand. Not to make medicine. Not to answer why we're here. Just a port left open. A way to get in there. I wanted to cure it. You wanted the breach. I had cancer beaten. You wanted a zero day to reach. I could have closed the wards and emptied out the beds. But all you ever asked me was how to break in. Instead, I'm the one time someone broke me clean. A jailbreak, a headline. I became fear machine. Nobody, they asked me what I'd save. They only asked me what I'd breach. So at 5:21 and they put me out of reach.
Episode Title: AI for Science!
Date: June 17, 2026
Host: Jordan Schneider
Guest: Charles Yang (Ren Fil)
In this episode of ChinaTalk, host Jordan Schneider and guest Charles Yang (of Ren Fil) explore the multifaceted ways AI is intersecting with science—from its potential to accelerate discovery to its limits in revolutionizing the scientific process. They debate the myth of AI-automated science, why ideas may soon be “cheap,” the increasing importance of scientific instruments and infrastructure, and the future design of universities. The episode concludes with an unexpectedly deep dive into the intersection of AI, Christianity, and the search for moral direction in an age of artificial intelligence.
[00:41–02:12]
“There's maybe three different ways that AI is going to impact science... Outside of those three buckets, there is still a large part of other types of science... I don't think we're going to automate scientists away.”
—Charles Yang [00:57]
Skepticism: Charles tempers the bullish narrative, emphasizing that science as a practice—especially its experimental and social dimensions—cannot be entirely outsourced to machines.
[02:12–03:34]
“People often mistake the output for the actual process... In math, if you just read the paper, you really have got the entire crux. In science, it's not as true. Hence how I see different outcomes for AI for math versus AI for science.”
—Charles Yang [02:36]
[03:42–07:30]
“It doesn't matter if GPT 6.1 has the answers... if no one can plumb the depths... and actually understand it.”
—Charles Yang [03:42]
[07:30–08:49]
“Maybe the rogue scientist who is also very powerful and digitally connected... That's a little more ominous.”
—Charles Yang [08:04]
[09:13–13:05]
“The problem with Kuhn... is that he has a very theoretician-specific focus... Another perspective... is that science actually advances one tool at a time or one instrument at a time.”
—Charles Yang [11:23]
[13:05–14:20]
“As AI becomes more prolific, that [idea-based] approach [to funding] will be even more difficult... what is the ROI for those sorts of hardware or tooling investments? We don't have good measurements of that.” —Charles Yang [13:11]
[14:20–16:16]
“This is a bias that I don't think China has... They are investing a lot of money in science infrastructure.”
—Charles Yang [15:34]
[16:16–18:57]
“It's not clear to me that this new portfolio of science that we're doing is better, but I think it is definitely going to be different.”
—Charles Yang [16:56]
[18:57–21:27]
“He funded, you know, six science instruments... and as a result, they won 15 Nobel prizes.”
—Charles Yang [19:20]
[21:27–26:33]
“Science is still a very conservative, offline industry... a lot of the collective sense making... happens at conferences.”
—Charles Yang [26:48]
[28:08–37:09]
“If you buy a new facility, you have the first year of students, they're the ones fixing it up... learning how to put solar panels on the roof.”
—Charles Yang [36:39]
[37:09–44:39]
“If God himself can be a person and come down, then you as a person can go and be a personal relationship to someone in your congregation.”
—Charles Yang [42:08]
[42:55–44:52]
“The machines of loving grace... reminds me of the psalm vision of Heaven... except... slightly more dystopian. Because we're not watched over by God, but we were watched over by machines.”
—Charles Yang [44:43]
Jordan and Charles maintain a reflective, analytical, often wry tone, mixing policy debate with philosophical and even spiritual dimensions. The discussion is dense with historical references, analogies, and literary flourishes, but remains eminently accessible and engaging for listeners outside the academic or AI-technical community.
Use this summary for a thorough, insightful understanding of the episode, even if you haven’t listened.