Artificial intelligence in medicine and science
- Thread starter RedFox
- Start date
I’ve seen some really interesting work with adversarial iteration until agreement is reached. Sounds like what you’re doing?
mariovitali
Senior Member (Voting Rights)
No, I wouldn't necessarily make sure that there is an agreement. Basically I start with -for example- several findings from ME/CFS studies (e.g. GWAS, replicated findings etc) and then I ask a number of LLMs (e.g. 3) to suggest a causal hypothesis.
2) Next step is to present to each LLM the hypotheses of the other two and rank it.
3) A PageRank algorithm (this is an example) gives a ranking of the hypotheses and how much each LLM agrees that another hypothesis is better.
4) An "orchestrator" LLM looks at all hypotheses, identifies where all LLMs agree, where they disagree, whether any factual discrepancies were found and whether novel information deserving further investigation has been identified.
It works surprisingly well in my opinion.
Thanks for explaining @mariovitali an interesting approach! If you’ve written more on this anywhere please do share any links.
On the topic of this thread IMHO the solution in the healthcare industry is for those in it to provide better support. Then fewer people will turn to unsuitable alternatives.
Meanwhile AI is achieving real results in STEM, here’s yet another
arstechnica.com
They’re not magic answer machines and selling or using them as such is dangerous and likely starting to hit walls in terms of financial viability https://arstechnica.com/ai/2026/06/...sers-react-to-new-usage-based-pricing-system/
But these technologies do and will continue to have uses and achieve breakthroughs in the sciences including medicine. To me a lot of discourse seems to end up in unhelpful battles over things being entirely good or bad or focus on the morals of the technology rather than the people, business practices and regulation. We need better education and better informed decision making on how we want to use the technology as it evolves.
On the topic of this thread IMHO the solution in the healthcare industry is for those in it to provide better support. Then fewer people will turn to unsuitable alternatives.
Meanwhile AI is achieving real results in STEM, here’s yet another
An OpenAI model solved a famous math problem that stumped humans for 80 years
I tried to explain OpenAI’s solution more clearly than OpenAI did.
arstechnica.com
They’re not magic answer machines and selling or using them as such is dangerous and likely starting to hit walls in terms of financial viability https://arstechnica.com/ai/2026/06/...sers-react-to-new-usage-based-pricing-system/
But these technologies do and will continue to have uses and achieve breakthroughs in the sciences including medicine. To me a lot of discourse seems to end up in unhelpful battles over things being entirely good or bad or focus on the morals of the technology rather than the people, business practices and regulation. We need better education and better informed decision making on how we want to use the technology as it evolves.
Utsikt
Senior Member (Voting Rights)
So I just learned about Lean, which is a programming language that essentially allows you to check if a math solution is logically valid or not.
Which means that for all maths problems AI tries to solve, they will be able to check if the solution they are working on actually is mathematically sound while they are working on the problem.
This makes it far easier to solve math problems than problems where you don’t have a way to check if your solution actually makes sense. The new math proofs are exciting, but still a substantially different use case compared to medicine.
Which means that for all maths problems AI tries to solve, they will be able to check if the solution they are working on actually is mathematically sound while they are working on the problem.
This makes it far easier to solve math problems than problems where you don’t have a way to check if your solution actually makes sense. The new math proofs are exciting, but still a substantially different use case compared to medicine.
Yes exactly, AI seems promising in domains where we can automatically check correctness of the final answer. In math, my understanding is it's operating as an idea generating engine, and there's a second piece of the system checking the ideas the model comes up with and stitching them together (presumably using Lean as you suggest). We should be cautious about extrapolating AI being good at advanced math to anything else.
Ironically, I am coming to think this property (which is basically falsifiability I guess) is also why humans have been able to solve such hard problems in math compared to certain other fields, where we just seem to go in circles without ever learning from our mistakes...
Ironically, I am coming to think this property (which is basically falsifiability I guess) is also why humans have been able to solve such hard problems in math compared to certain other fields, where we just seem to go in circles without ever learning from our mistakes...
Utsikt
Senior Member (Voting Rights)
This is one of those use cases where false positives and false negatives aren’t too detrimental if the flagged papers gets checked manually and people understand that not being flagged doesn’t mean that everything is correct.
It’s better to catch more mistakes than not.
Another good example of a positive use case, making gene editing safer.
arstechnica.com
Precise DNA base editing using AlphaFold3-based contact modelling, 2026, Meng et al
Abstract
Achieving high specificity in biochemical transformations is crucial for research and therapeutics. This is particularly important for genome editing, where enhancing tool specificity ensures effective and precise editing outcomes. Current strategies are constrained by activity-specificity trade-offs, high labour intensity and low success rates. Here we present ContactSeek, an artificial-intelligence-driven framework that uses AlphaFold3 (AF3)-predicted contact probability5 to improve the specificity of genome editors. Using Cas9–TadA adenine base editors as a demonstration, we mapped their genome-wide off-targets and fed the off-target DNA sequences to AF3. Among AF3 outputs, we found that contact probability was more sensitive than predicted three-dimensional structures for detecting differential interactions between on- and off-target complexes. Correlating contact probability with sequencing-based off-target signals, ContactSeek identified and ranked consensus contact regions, which are neighbouring Cas residues with consistent contact changes to DNA/guide RNA, and pinpointed specificity-determining residues within them. ContactSeek can also be applied modularly and identified key residues in the TadA8e deaminase. Targeted amplicon sequencing, genome-wide profiling, R-loop assay and RNA-sequencing together confirmed the greatly enhanced specificity; our best variant, combining two mutations of Cas9 and TadA8e, outperformed several known high-fidelity adenine base editors. ContactSeek is also generalized to Cas12a-based cytosine base editors. Collectively, our framework represents an AF3-driven model tailored for specificity improvement, establishing a paradigm for improving the precision of genome editing tools through the integration of structural and functional dimensions.
Web | DOI | PDF | Nature
Team uses AlphaFold AI to redesign gene-editing proteins to make them safer
Google's AlphaFold can help ID what parts of a gene editing protein enable mistakes.
arstechnica.com
Precise DNA base editing using AlphaFold3-based contact modelling, 2026, Meng et al
Meng, Haowei; Lei, Zhixin; Yan, Yongchang; Wang, Liren; Zhang, Sihan; Rao, Xichen; Shao, Chuyun; Zhang, Xiaoting; Chen, Ke; Yang, Lei; Liu, Rongrong; Yang, Gaohui; Shen, Ruoyu; Gu, Ruichu; Wang, Xinyan; Wang, Yiya; Lu, Suiru; Lv, Zhicong; He, Bo; Wen, Han; Li, Dali; Yi, Chengqi
Abstract
Achieving high specificity in biochemical transformations is crucial for research and therapeutics. This is particularly important for genome editing, where enhancing tool specificity ensures effective and precise editing outcomes. Current strategies are constrained by activity-specificity trade-offs, high labour intensity and low success rates. Here we present ContactSeek, an artificial-intelligence-driven framework that uses AlphaFold3 (AF3)-predicted contact probability5 to improve the specificity of genome editors. Using Cas9–TadA adenine base editors as a demonstration, we mapped their genome-wide off-targets and fed the off-target DNA sequences to AF3. Among AF3 outputs, we found that contact probability was more sensitive than predicted three-dimensional structures for detecting differential interactions between on- and off-target complexes. Correlating contact probability with sequencing-based off-target signals, ContactSeek identified and ranked consensus contact regions, which are neighbouring Cas residues with consistent contact changes to DNA/guide RNA, and pinpointed specificity-determining residues within them. ContactSeek can also be applied modularly and identified key residues in the TadA8e deaminase. Targeted amplicon sequencing, genome-wide profiling, R-loop assay and RNA-sequencing together confirmed the greatly enhanced specificity; our best variant, combining two mutations of Cas9 and TadA8e, outperformed several known high-fidelity adenine base editors. ContactSeek is also generalized to Cas12a-based cytosine base editors. Collectively, our framework represents an AF3-driven model tailored for specificity improvement, establishing a paradigm for improving the precision of genome editing tools through the integration of structural and functional dimensions.
Web | DOI | PDF | Nature
Anthropic (Claude company) and other AI companies have started implementing hidden watermarks in their AI-generated text, to comply with EU regulations. It adds a slight bias in which words are picked, allowing someone to check if the text was likely written by one of these company's AIs.
Anthropic: "How Claude’s text watermark works"
Anthropic: "How Claude’s text watermark works"
This from a podcast series from the Crick presented by Brian Cox
Panel:
Legit funny that this is the limit of their thinking. Just those things? There is not a single reason why human physicians should be better at any part of health care, in fact it's guaranteed that they won't be. Even the squishy parts that a lot of people cling to, like communication, it's not even a close thing, AIs are massively better at it already, because they have virtually unlimited time to dedicate to everyone, without delay. It's as wide as the difference between a teacher with 5K students, and one with a single one, a dedicated tutor. This simply can't be compared, it's way out of proportion.
It's actually how odd how most of the limits people self-impose on what AI can do is just the same old self-centeredness that made it difficult to accept that Earth is not the center of the universe. People believe there must be something unique in human intelligence and socialization that cannot be replicated with computation. And yet it's already clearly the case that there is nothing unique about it, that it can all be replicated, even do better at every single part of it.
It's replacing software and math experts first because everything we do is verifiable, but there is very little in health care that is harder than what software developers do, with the same problems of there being too many possible solutions to brute force them, which requires reasoning, and how there are no true "right" answers as long as there is a solution to the problem works. Programming may involve a lot of math and is mostly verifiable, but it requires a ton of intuition, and debugging is very similar as diagnosis. It requires thinking at least as flexible, if not more because the space of possibilities when dealing strictly with information is massively larger than even the possible combinations of all the atoms in the universe.
It will do everything in most cases, better, faster and cheaper. The more I look at it, and the more it becomes obvious that one problem people see in what they perceive as madness from "tech bros" is that humanity has pretty much lost having ambitions, and this is the first time in a long while that there are people pursuing something truly ambitious, it looks completely out of norms. Because it is. Humanity has been overruled by a tyranny of low expectations, and thinking that AI, super smart intelligence with almost infinite capacity for work, will simply take of a few admin tasks and analyze lab findings is a good example of how low ambition dominates all thinking.
As one famous Quebecois said: think big, sti.
There’s far more than that in the podcast. So to answer the question no, not just those things
But they do cover that and one thing some patients and healthcare professionals would like is more human to human time. Let the computers do what the computers are good at and the humans the human things.
Last edited:
Posting a bit more as the cut and paste podcast description may not do the podcast justice. I’d like people to take away that this is a very listenable discussion with some good overviews of many different aspects and differing views (this whole series from The Francis Crick Institute is very good).
They cover failures/errors of doctors, problems with bias, how pain is often experienced as well as talked about and seen differently in women and men and how AI could reinforce or indeed solve these problems. On how AI can see patterns in data we cannot. On recent developments using AI to create far broader vaccines and potentially speed up medicines so we can respond to pandemics and save millions of lives.
They also cover future drug discovery and where in the process AI fits and that it is good as finding patterns in noise or ‘sorting out mess’ as it is put. And if we can keep patient data safe and private, partner with academia as well as companies to improve outcomes for us all and avoid monopolies.
It has the sort of nuance and public engagement that perhaps we need more of. These are really important topics with significant implications so we need both.
They cover failures/errors of doctors, problems with bias, how pain is often experienced as well as talked about and seen differently in women and men and how AI could reinforce or indeed solve these problems. On how AI can see patterns in data we cannot. On recent developments using AI to create far broader vaccines and potentially speed up medicines so we can respond to pandemics and save millions of lives.
They also cover future drug discovery and where in the process AI fits and that it is good as finding patterns in noise or ‘sorting out mess’ as it is put. And if we can keep patient data safe and private, partner with academia as well as companies to improve outcomes for us all and avoid monopolies.
It has the sort of nuance and public engagement that perhaps we need more of. These are really important topics with significant implications so we need both.
I read a news item yesterday (can't find a link now) about AI in the laboratory loop. It uses an AI to study (using simulations of cells) results from an experiment, and then suggest changes to that or proposes another experiment. It could even handle the lab equipment, saving time. That's the sort of progress I'm expecting from AI in medical research.
Further benefits: the knowledge bank for cell functions (and simulations in ever-increasing detail) accessible to AIs will keep growing, so if a drug is proposed, it can be checked for side-effects on all cells in the databank. Maybe the targeted mechanism in heart cells also exist in kidneys or retinas, so you can discard that pathway immediately, rather than wasting billions and years.
Further benefits: the knowledge bank for cell functions (and simulations in ever-increasing detail) accessible to AIs will keep growing, so if a drug is proposed, it can be checked for side-effects on all cells in the databank. Maybe the targeted mechanism in heart cells also exist in kidneys or retinas, so you can discard that pathway immediately, rather than wasting billions and years.
Everything I have heard is we are a some way off being able to simulate a cell, even a simple one let alone the huge array of more complex cells in the human body. It’s a goal and people are working on small pieces of the puzzle but being able to do so is not something I can see advancing things soon.
Maybe one of the things that could help drug development snd trials would be access to all the unpublished data all the different drug companies have. There’s a huge amount of hidden data which AI could help make sense of for fuure trials and usage. I can’t see them letting it out though, it will remain siloed in different companies for commercial reasons.
Maybe one of the things that could help drug development snd trials would be access to all the unpublished data all the different drug companies have. There’s a huge amount of hidden data which AI could help make sense of for fuure trials and usage. I can’t see them letting it out though, it will remain siloed in different companies for commercial reasons.
A complete simulation of a cell may still be a long way off, but simulating parts is still useful. Modelling a receptor, even if it's just accurate for 88% of situations, can still help design a molecule to bind to it, or to check whether a molecule meant for a different purpose might bind to it and cause problems. The accuracy would increase, as would the speed of the processing (knowing which checks should be done first).
mariovitali
Senior Member (Voting Rights)
It is good to see that AI is indeed helping to formulate and evaluate medical hypotheses, including now ME hypotheses, right in this forum.
Sure. My understanding is the simulations are pretty basic and narrow compared to the complexity of a cell or drug interactions. There’s some interesting tools in scGPT, Geneformer, etc which may have shifted things in specific areas of predicting some elements and I’m sure there will be tools which will help drug discovery. But they seem like expansions of bioinformatics tools and just another thing a professional would use in their pipeline to me.
More widely I suppose I take the approach that those of us who see the benefits have a responsibility to be really careful in what we claim is possible. There’s understandably a lot of scepticism and also just plain dislike of AI. some of hype we hear makes this worse IMHO not better and there are people, companies and news stories claiming all sorts of things which are not anywhere near possible. When challenged they just waive their hands around and say things are moving fast as if AI was magic and so normal rules and scepticism don’t apply.
If we stick to what is realistic and achievable now and see any surprise advances as a bonus that seems a better approach. And one which could be more successful than making claims which don’t come to fruition and so can be used to dismiss the whole field.
Last edited: