Have We Reached AGI? We Talk With Immunologist Derya Unutmaz
Derya Unutmaz, an NIH-funded immunologist and professor at The Jackson Laboratory, talks to the hosts about his claim that AI and biotechnology are heading for a "bio singularity" that will transform human biology. He argues that the first level of AGI has already arrived, that current models are better scientists than he is, and that it will soon be unethical for doctors not to use AI. The hosts then debate whether his timelines are realistic and whether society is preparing for the disruption.
Transcript
Derya Unutmaz is not some Silicon Valley futurist. He's an NIH funded immunologist with 35 years of published research offering more than 100 papers. A Jackson Laboratory Professor who's received 10 and a half million dollars in federal funding. OpenAI selected him as one of 10 scientists to test GPT-5 on a real scientific experiment. He gave it a complex cancer study, and it predicted his results in minutes. But since then, he has made a series of claims that almost no one else in his field will co-sign, that all disease will be cured within a decade. And that 50% of white collar work can easily be automated this year. But predicting an experiment is not curing a disease. And so far the gap between what AI promises and what it delivers keeps growing. So is he just early with his predictions, or is he overselling what AI can actually deliver?
We are talking AI, disease, and so-called bio singularity with immunologist Derya Unutmaz.
All right, Stephen, so Derya doesn't just think AI will transform medicine, he is calling biosingularity, which is the moment when the convergence of AI and biotechnology will radically transform human biology. He's calling it the biggest discovery, really, since the beginning of time. Bigger than fire, bigger than antibiotics. Are we over blowing this?
Well, if he's right. I mean he, he makes a lot of bold predictions, right? He talks about AGI, you know, in 2026 where everybody else is talking about 2028, 2030. He's talking about SGI, super general intelligence, where computers are, are cleverer than human beings by next year or the, or the year after. So if he's right, then yes. But the question is, Laila, is he right?
I think it's just a, a really, really bold claim, especially from a scientist where I just, I think you have to approach anything, even if it is incredibly promising, with skepticism. I understand that you don't want to be left behind, so you want to kind of grab onto the opportunity. But I think saying 99%, as in I think we just have to be super rigorous.
Well, let me, let me, let me play devil's advocate on this though. Is there not though the chance that he's right? Should we not as a society put something in place if he is? Because you saw what happened to IBM share price yesterday. A paper comes out that says that software isn't going to be as necessary as it has been in the past. Their share price goes down 13%. That's billions of dollars in one day.
I know, I get that, but I also think that if he's not right, then this is just a huge overreaction. Like it's so unnecessary for IBM's stock price to be going down by 13%. That's impacting so many people. And you've got so much volatility, which, if it's driven by hype, that's a incredibly turbulent society to live in. Like, why can't we actually, like, why is there so much uncertainty where we're like, is it hype? Is it not hype? Why can't we just scrutinize this in like a scientific way?
Oh, and you have to believe that he's not pulling these numbers out of thin air, that he has to be making these bold claims and bold predictions based on something that he's, that he's really seeing, so let's hope he's right.
Right, and the other thing though, I would say to you is you can't leave it to the scientists, right? We can, we can't leave this to the scientists.
Yeah, those are fighting words for Laila.
As a society, you know, we've got to put things in place because if we don't put things in place, then, you know, then we're going to get caught out by this when it happens. So yes, science is important, but we can't leave all this to the scientists. And I think that's what we're doing right now.
But we can't leave it to, you know, the tech bros or the CEOs.
And the corporations.
The CEOs and their friends.
Right.
All right, to your side, and we want to know what you think. Is AI really a silver bullet that solves humanity's most difficult problems? Or is that exactly the kind of thinking that gets us into trouble? Tell us in the comments below, like, and subscribe. Now, here's our conversation with Derya Unutmaz.
Derya, thank you so much for your time today. Now, you've made the bold claim that if someone can survive the next five years, then they will live for the next 50. All thanks to what AI can accomplish in the medical research field. Before we dive into that claim, can you first kind of narrow down for us exactly what you're referring to when you say AI, and what will change within the next 10 years that will allow me to live for the next 50?
AI and technology in general is increasing exponentially. So, you know, our minds are very used to linear thinking. So we, we think that whatever the advances happen in the last 10 years, well, you know, next seven, eight years, we're going to have the same advance. But that's not how, how it actually works. And I think we can already see that. So the capabilities of technology and, and now with AI is literally doubling every few months. And this will actually even accelerate further. So we're going to get a point where, you know, what we sort of term as singularity, where, you know, every week or sometimes every day, there's going to be a huge, huge advances.
Derya, I just want to jump in here. When you, when you are talking about AI, are you talking about LLMs or are you talking about models like AlphaFold, as in how, how are you separating them? And when you say they're advancing exponentially and they are going to lead us to this, is that LLMs that you're talking about?
So I, I, I think, you know, the, the notion of AI is currently associated with LLMs. In the next year or two, we're going to have what's called the physical intelligence. So, you know, we're going to have the world models, you know, some of the videos, for example, are generated by diffusion models and, and actually what we call LLMs are now becoming very multimodal, right? So, you know, we have the agentic systems, we have the coding abilities, we have the speech, we have the video, we have the images. And then, you know, Google actually just released sort of the, the, the world models where you can, the, the AI can understand the world in 3D. Actually, this is going to be very important for, for biomedical reasons.
So I mean, I see this as a, as a, you know, totality of intelligence, artificial intelligence, which is really becoming generalized, what we call general intelligence. So of course, you know, at the very top it's LLMs because that's, that's sort of the reasoning. I mean, you know, this is the way our brains work too, right? So we, we have a frontal cortex, we can do abstract and cognitive thinking, but then we have, you know, areas that control our hands or, you know, all, or our physical capabilities and so on and so forth.
Yeah, one of the things that really interested me listening to you, researching, coming up for, is that you're talking about AGI this year, 2026. You're talking about AGI and, and that's quite early, you know, a lot of people are talking about it in, in two years time. And my second question is, you know, there was a report, as you'll know, that came out yesterday talking about the 2028 global crisis and the stock market went crazy, and they were talking about it in 2028, and you are talking about 2026, and you're talking about AGI in 2026. My two questions are, why are you earlier than most people on this? And second of all, do you think, as a society, we're taking this seriously?
Let me, let me define what I mean by AGI, because this is the problem. You know, AGI means so many different things to so many people. So if, if you break it down, it means artificial general intelligence. So what general stands for is that, can you generalize knowledge? So for example, if you learn something on one topic, we can somehow generalize that information to learn something completely different, or understand something completely different.
What I think changed now is that why, the reason I think that we are at AGI, what I call level one, and actually I think there's going to be three, four levels of this. I think ChatGPT showed that, you know, this is the path. You know, if you, if you train enough data, if you do this and that reinforcement learning, you know, this will just keep increasing in, in power and intelligence. So it was clear that, you know, they will reach a point where, you know, most of the jobs will be literally replaced. I mean, anything that depends on using computer, or your intelligence, or your experiences or expertise could be eventually replaced by, by AI.
So I think, you know, the society, or I should say governments or organizations, I don't think anybody, you know, were prepared or, or cared for it, to honestly. I mean, you know, there, there was a lot of talk about AI is a bubble, you know? This is all hype, you know, this cannot continue to scale and, you know, at some point it's going to stop and so on and so forth. But, but I think in the last few months we start to see that this is going to have a tremendous impact. I mean, you can see like Anthropic announces that they have a new cybersecurity agent, AI model, and then suddenly $50 billion in market values lost in security companies. Yesterday they announced that, you know, they can very easily clean the, the COBOL computer language based mainframes. And then IBM loses, I don't know, billions of dollars in market value. And of course, you know, a lot of people are unfortunately going to lose their, their jobs because most of it is automated. And it will continue to accelerate in automation.
So this is, we should be very intellectually honest to say that, you know, while I believe AI is the greatest thing that will happen to, to humanity because, you know, we'll literally save billions of lives and, and reach a point of abundance, the path to that is going to be very painful. It's going to be extremely disruptive, and there is no plan, there's no, you know, preparation for, for governments or organizations, you know, how to deal with that. I mean, the, the whole education system has to change. You know, what does it, you know, what's the point of going to college just to get a job? You know, 30% of the college graduates cannot find a job in the US right now. So, so we have to rethink the whole fabric of society. I mean, this is the biggest transformation since the beginning of history. And, you know, it's, it's very disappointing that we have very little discussions. Maybe, hopefully it will start this year, but it might be a little late.
But on that note, Derya, do you think LLMs and companies like Anthropic and OpenAI are going to be the ones that take us there? Because there are a lot of critics, and I would say that your timelines are actually ahead of people, even like Demis Hassabis, who have a vested interest in, in this. But Demis doesn't think that our current LLMs are going to get us there. Richard Sutton also has kind of gone back on that and even said that people like Gary Marcus do have a point that we, we are hitting a wall. So, because a lot of people say that, look, LLMs are just regurgitating things, they're stochastic parrots, that they are telling us nothing new because obviously they've gone through all the literature and are giving that back to us. And because we communicate in language, we perceive that to be something when it's, it's not. So how, how would you respond to those people? Because there are plenty of people who have that viewpoint.
And I think I was trying to explain that. What, what is, what people are confusing, like Demis, for example, is, I mean obviously he's not confusing, but when he says AGI, what really he means is ASI. So ASI is super intelligence. Super intelligence means better than any human. So, so for example, Demis gave this example. So for, you know, if he had the, if he had cut off the, the knowledge of, of the AI models or LLMs to 1911, can they find, can they discover theory of relativity, which was discovered by Einstein? Well, that's an extraordinarily high bar. There are only like few people in the history of humanity who came up with such an incredible insights, right? So the, for the general public, you know, when I say AGI, I mean, you know, an AI system that's as good as, let's say, you know, the top 1% of humanity. And we're already there.
So, and then the other thing that people are, are confusing is that, you know, the, the, so we have the, the cognitive, the abstract intelligence, which is actually very new in evolution. It's only about 100,000 years old, since homo sapiens have evolved. The actual biological intelligence is, has nothing to do with the, the cognitive, the abstract part. You know, animals have a, an incredible intelligence, right? So most of that intelligence is actually about the physical world. So we evolve to survive. So we, we have to think temporally, we have to think, you know, in, in 3D format, and we have to think what, what actions we take will have a consequence. And, and obviously LLMs or the current AI models don't have that, but we going to get it there. So, so that's why we need the physical intelligence aspect of it, which requires embodiment like the robots as well as the, these 3D world models.
But, but that does not distract from the fact that, you know, most of humanity works on the cognitive level. I mean there, there, you know, if you are an electrician or plumber, your job is safe for, for maybe five years or maybe 10 years at most. But if you are, I don't know, like I'm a scientist, you know, my, most of my work depends on intelligence, expertise. And I can tell you that the current models like GPT-5.2 Pro is just simply better than me. You know, it can, it can come up with better ideas, it can analyze data better. And this is true for many of the physicians, many of the lawyers. So anything that depends on high level thinking and intelligence is going to be replaced fairly, fairly soon.
The, the lower, what we call lower level intelligence is actually physical intelligence, which took several hundred million years of evolution. That's kind of imprinted in our minds. We are, we're already born with that. So I don't know if, if, if that makes it a little bit more clear. So, but we're going to have couple of levels of AGI, so I think the next level is going to be having the AI models much greater memory, because this is a huge problem. You know, once the AI understands the 3D world and experiences that, then you, you, you literally get to the ASI, which is the super intelligence. And I, and I'm, I'm sure that's what Demis Hassabis mean.
But Derya, we're not there yet. So to bring this question, to bring this question back to your scientific research, you know, with, when it came to the cancer study that you gave GPT-5, your own papers are a part of the training data for AI. So when it predicted the answer to your experiment, was it really understanding your work, or was it just recognizing what you had already put in? And to that point, does that not give a, a good argument to the people who say that there will always be a place for the human component?
I, I trained as a medical doctor, I don't see patients, but, you know, I learned a lot of stuff. I was trained with, you know, 1000s of bits of information that were created by other humans. So I didn't, I didn't create any of that, and then I became a scientist and I still had to be trained with another a million bits of information. That's, that was already out there. The only thing that I, I brought there is that I could make, I could make patterns, you know, I could, I could see that, you know, if I did this experiment, I'm likely to get this output, or the output of the experiment means, oh wow, this is how the mechanism works. So that's, that's the contribution that I'm making. But I'm, I'm totally dependent on, you know, millions and millions of, of papers that have been published already. So our minds is not that different from them.
And the, furthermore, the, the models right now, and this actually started with the reasoning models, with, with all o1 model from ChatGPT, from OpenAI, that they are able to think. What that means is that they're able to truly understand the, the context of what you're giving to them. I mean, the example that I gave for ChatGPT was, this was an experiment that we had done several years ago. We couldn't figure it out. We couldn't figure out the mechanism, we didn't publish it. And so the GPT-5 point, you know, actually it was 5.0 Pro model, you know, within 12 minutes it, it already told me the, the experiments that we did actually several months later. But not only that, it came out with a mechanism that in retrospect I should have told, I should have known. So if the model is not understanding, how is it able to, to come up with that mechanism that I couldn't come up?
All right, well, we're going to continue the conversation about the use of AI when it comes to scientific research because coming up after the break, we're asking Derya why he thinks it's actually unethical for scientists not to use AI.
So Derya, when it comes to the use of AI in research, you say it's actually unethical for scientists not to adopt and utilize AI in their work. But given the ongoing issue of hallucinations, do you think it's a bit reckless to make that claim? And shouldn't this be a personal decision that each researcher comes to on their own about when to adopt AI and use it in their research methods?
Actually, I said that first for, for medical doctors, but, but it's also true for scientists. You know, I can definitely repeat that for science as well. It's, it's more sort of, the ethics is more for, from the, from the medic, physician's perspective because, you know, there you're, you're trying to treat a, a patient and you don't want to misdiagnose. So, you know, it's, it's very clear that AI, you know, from the example I just gave, is able to see those patterns and, and medicine is about pattern recognition. So, you know, when you're making a diagnosis, you, you have to think of 1000s of possibilities and then come up with a diagnosis.
So we had done, again, a study where we had, you know, millions of bits of information from a, you know, five-year-old study from a chronic fatigue syndrome subjects. And, you know, it took us about months and months of analysis of that data. In fact, we had used some machine learning at the time, and then eventually, you know, we, we were able to, to put a picture of it. So, you know, just to give you an example, I took that data and then gave it to GPT-5 Pro. And again, you know, in, in matter of 10, 15 minutes it came up with exactly the same thing, but more stuff that we have actually missed. So I am contracting months of work to literally minutes.
But I also think that the opposite is true, because there are still limitations to our current AI models, like I use GPT-5.2 Pro and it still confidently tells me things which are completely wrong, definitely not as frequently as it did, but they do still hallucinate. I mean, it's now got, the hallucination rate is much smaller, but if a doctor misdiagnosed someone due to AI, who's accountable? Isn't that equally as unethical? So I like, I don't know if we can confidently say that AI is just so infallible or even more infallible than a doctor or a scientist is at the moment. Because if you are confidently so reliant on it.
Right.
As in that, that can be quite dangerous.
That's the thing. I'd be more concerned about doctors and people in the medical community utilizing AI when it comes to my diagnosis versus in, you know, a scientific research situation.
So first of all, I mean I've, you know, I use AI day and night literally. So I, I sleep less right now because I'm, I'm so excited. I've been using GPT-5 Pro actually before it came out. I was a, I was an early tester. I have not seen a single hallucination in my own field, you know, in biomedicine, like, not, not even one. So I don't know what hallucination it, what wrong information you got. But let me, let me put it this way, I mean, we hallucinate all the time.
So I'll give you one statistics. In US alone, there are 12 million misdiagnosis every year. Only about 70% of the diseases are diagnosed correctly by, by medical professionals. And about 700,000 people die or, you know, severely become sick because of misdiagnosis, okay? So, so this is an incredible issue. So, and, and with AI, if you could improve that by 10%, you are saving hundreds of thousands of lives. And I'm, I'm betting everything that AI will improve that, you know, maybe, you know, instead of 12 million misdiagnosis, we're going to have maybe few hundred thousand misdiagnosis.
What we have right now is the worst it will ever be. You know, there's going to be new models coming out that are going to be better and better, 10 times, 100 times, 1,000 times. So, so there is, there is no limit to that. So there is a point where, you know, as I said, I mean, it's not just unethical, it's going to be malpractice not to use AI model. You know, you can write me, write down this and, and, you know, we can talk maybe in few months or in a year, because, you know, if, if an AI model is able to diagnose something better than any physician in the world, then not using that would be like, okay, I don't want to use a stethoscope, I'm just going to put my ear on the patient because, you know, I trust myself, my own intuitions.
I do think it is, it is extremely dependent on the context that you feed it, the memory it has, and the way that you are doing the scaffolding around it, as opposed to kind of just the raw model. I agree the raw model is incredibly powerful, but depending on the, on the context that you have, entire code base is being completely erased when you don't, don't want them to, so they can go off piste. And so I think it is critical that we know how to, I guess, supplement them or scaffold them correctly because there, there is risk there if they, that's not done properly.
I mean there's, there's very little risk, but as I said, I mean the risk of humans is much, much greater even now. And it's going to be much greater very soon. You know, I work on a very, very specific, very complex topic, you know, biology is very complex.
All right, I like that we're talking about this in utilizing AI in the medical field, because that brings me to my first of our rapid fire questions, and that is where do you think people will draw the line at bringing AI into their own lives?
I think once people understand the true power of this, I don't think there would be a line. Why would you draw a line when you are getting such an incredible benefit for your, for your daily life, for your health? So, but, you know, of course I'm sure there will be some people who will, who will refuse to use AI. I mean, you know, there are people in the world who don't use internet or, or phones, but, but I think for general population, this is, this is going to be fully integrated into their lives.
Once the trust is there, yeah, I agree.
So Derya, my question is what is something that's universally accepted in your specific field that you actually disagree with? Which I imagine is quite a lot.
Yeah. Quite a bit. I, I'm not very conventional, so I mean, I guess the, the, the most important one would be that aging is, is a natural phenomena and it can never be cured or reversed. So, so I 100% disagree with that.
Derya, my, my question is that, and, and, you know, you are, you are, you know, as you say, thinking somewhat differently to a lot of other people, but what can you see 18 months out or whatever that, that we are not talking about at the moment?
Well, one thing that's, that's for sure is that there's going to be a tremendous disruption. You know, we, we might have to question the way we do science. Like, is, is there any point in publishing papers right now? You know, you send it, you send a publication, it takes six months to a year to publish something when the world is moving in an hourly, daily basis, when AI models can discover drugs in, in matter of minutes. So that, that brings us to a completely different level. So, you know, something that, that maybe people are not talking enough, again, you asked that question, is the, is the shock that we are going to experience in 18 months.
Derya, thank you so much for your time today. We really appreciate your insights and you taking the time to speak with us.
All right, so really fascinating conversation with Derya. Did he make the argument that convinced both of you that the train is actually moving as quickly as he says it is?
I mean, I, I like using AI, I use it quite a lot, but I, I'm, especially in science, I'm still very, very skeptical of how fast that's going to move. As in, I wish we could have pushed him a little bit more on the, the aging and the curing side, side of...
Right.
You know, curing all diseases because as in it, it's not, it's not that straightforward. Like if AI was really, really that good, it would just be making, you'd send it out into the world and be making a million trillion dollars in the stock market, but why is it not, you know, challenging hedge funds? Why is it not, like there are more bottlenecks along the way in the real world than I think were discussed?
Well, so, so I'm not a scientist, so, you know, I can't argue, you know, which model of AI is doing this or doing that, whether we're going to get AGI one or two or three or four, I'm, I'm not a scientist. I think the point is that wherever it's at and the speed that it's going is going to cause, as Derya has said, massive disruption to society. And there's no point sitting here sort of saying, "Oh, science is not perfect." And losing, you know, and not taking into effect the, the effect it's having on society, and the effects it's having on society are now, you know, people are losing their jobs. People are finding it hard to get jobs. People are making millions of dollars on the stock market using AI, you know?
The pros.
And people are losing millions of dollars on the stock market because of AI. So, so, so I don't know whether he is right or he is wrong. I understand that he's a little bit maybe further out than others, but I think it's missing the point because I think the point is, what are we doing about it? And the answer is nothing.
And he did agree with you that no one is taking this seriously enough right now.
Right, and that's, so I, that's what I agree with him on. So I don't know if he's right about science, right? I mean, what would I know.
He is right about the fact that no one's paying close enough attention.
Well, I think so.
I, I agree that we shouldn't just sit back and let it happen, but on the other side, we shouldn't just...
But we are.
But we shouldn't sit back and be like, "Oh, well, AI will just solve everything." Right, that like...
Right.
We should just use AI to teach everyone. We should use AI to do all the science. We should use AI to, you know, it's, it's, I don't think it's a silver bullet for anything. So I think both extremes are just not valid.
So what are you going to say to people who don't have a job? What's your, what, what's your advice to people who don't have, "Don't worry. The AI's not quite as good as it thinks it is."
No, I don't, I don't think, I wouldn't say that.
What would you say?
I would, but I, I don't know.
Figure out a way to make AI work for you so that you can keep your job versus being replaced by AI.
Okay.
Now.
Because yeah, that, that's a top level thing that we do need to, need to resolve. But I think also just scaring everyone from just saying that, "Look, there's nothing you can do. It's coming for you," is equally as unproductive. Where it's like, don't even bother. Don't even bother doing anything, like it, it's going to be better than you, don't even bother thinking. Don't even bother trying to come up with a solution, which I think is equally as dangerous.
Right, well, our huge thank you to Derya Unutmaz for joining us. If you think that AI will cure disease within a decade, or think that it's just a dangerous thing for a scientist to claim, well then we want to hear from you. Drop your comments below, subscribe, and make sure to like. We will see you next time right here on Agents of Tech.