I mean.. technically math is not "natural science" so they call it "formal science" which is about logic and rules, but they're highly interrelated fields.
Math is more like philosophy that's actually useful.
I've been in preclinical to clinical research for 20 years. I welcome all the help we can get, and I'm not an AI zealot. It has automated much of my work but I just don't see it fast tracking clinical trials much. My hope is that drug candidates become more successful. Maybe AI can help us weed out dead ends.
Where AI has already had a big impact on science is actually grant writing. Grants are how science is funded, and many agencies are being flooded with an order of magnitude more submissions. The AI submissions aren't always good (although some may help scientists who have food ideas but suck at grant writing to compete). The problem is they stress the system. Each grant needs a pretty thorough review to determine its value.
Physics is pretty math forward and those susy types were in trouble after the LHC turned up zero support. Looking at the navier stokes proof I am thinking the beautiful solution people are going to be quite miserable for a while while ai turns out 100 page theorems that are complex and ugly and correct
the world changed and our monkey brains will need a decade to even fully comprehend it.
television fried us too and never looked backwards long enough to accept it.
plug these ugly proofs into an LLM and ask it beautify it for mammalian hindbrain. give it curves and notions of understanding where we expect them.
i think humans reached their limit of understanding and what comes next is only possible via powerful integrated electronic intelligence.
this belief that human output is incalculably complex is being battered. we almost have to defend our output as impossible to fully emulate, since we understand we will soon be defending our own complexity next.
ai can certainly create novel connections and research. humans have eureka moments from insight from observations or reasoning or otherwise.
Lightbulb moments don’t come from nowhere. Inference is upgraded to divine insight because it guides human myth.
Research models have been in use for almost a decade now in most scientific environments. LLMs are still a useful tech, and while they played a role in some of the Nobel Prizes this year, the “disruption” of science isn’t going to be happening. LLMs can make connections between ideas in its dataset, but they can’t make novel discoveries. This is much more about improving the efficiency of researchers than replacing them, but nobody understands this less than these massive AI companies.
Imagine if there was a new hammer released, and everyone started talking about how it would disrupt construction work, and replace carpenters and contractors. Also, the people who make the hammer use it more than everyone else in the world combined, and fully believe the hammer would eradicate the human race if it got efficient enough. And it costs them $200 to make, but they only sell it for $45. That’s the “disruption” they’re trying to engineer into every industry they can.
Im only an astrophysics student, so take this with a grain of salt but it seems like the attitude at my school is that as long as youre not letting Ai take complete control of the data, its fine to help streamline large data sets, as well as helping brainstorm. Now, for the purely theoretical side, my one professor said Ai helped him with models and says he probably just saved himself a whole year.
To be clear they didn't play a part in any Nobel Prizes. Their results haven't yet even been formally published. What they did do is (most likely) solve a Millennium Problem.
> LLMs can make connections between ideas in its dataset, but they can’t make novel discoveries.
89% of scientific papers are non-novel. And even here the definition of novel is: "the recombination of pre-existing knowledge components in an unprecedented fashion.", which is exactly what AI does.
AI is definitely going to be disruptive in science, even if that's just scraping through huge data sets like the human genome or complex modelling and simulation to find patterns, correlations and anomalies that a human would miss.
>LLMs can make connections between ideas in its dataset, but they can’t make novel discoveries.
It's less about the LLM making novel discoveries and more about what the bulk of scientists' job will be going forward. For coders, their job will soon be, and for many already is, primarily AI management. For pure mathematicians, the same is now true. There's more resistance to AI generated art and writing, but it's not hard to imagine a future where that resistance fails, and practical art/writing jobs become mostly AI management. Note, by practical, I mean where images and writing are generated to fill someone else's purpose, rather than for their own sake.
So the question is whether that happens for the sciences generally. In most, AI use is already widespread. Could it reach the point where most scientists are simply managing AI, setting up prompts and reviewing results to determine which are most promising? And, a second question that often gets forgotten - would most scientists care if that happened? I suspect they might not. In most fields, experiments would still have to be set up by hand, and it seems likely that a lot of scientists don't actually care if AI speeds up the analysis of the results of those experiments. They want the findings and the significance of those findings. The actual fun for them lies in the experimentation itself.
This is substantially different than the other cases. Coders often became coders because they took pleasure in writing code. Artists and writers became artists and writers because they enjoyed creating art and writing. Mathematicians became mathematicians because they enjoyed the processed of working to solve math problems.
That's why you see such a backlash to AI. It's revolutionizing fields and making everyone in them way more productive, but all the benefits of that flow to the people in charge, while for the workers themselves, their job has gone from being one they, if not loved, at least enjoyed, to one they don't particularly want to do.
I am currently doing a PhD in theoretical physics and last week our boss has told us to start using Claude to write the simulations for us. I must say even asking it details about the physics side of the code and prompting it to change to different physical models has increased the output tenfold, it is really hard to argue against it in theoretical physics.
They will for sure be throwing the slop machine at any discipline they can just to try and legitimize it. Unfortunately, the cleanup crew are the already overworked and underpaid academics.
>Recently, AI disproved a longstanding conjecture in mathmatics
Are they refering to OpenAI "solving" a famous math problem, by basically just rewriting the problem with a bunch more variables? The proposed 'solution' is an open joke in the math field.
This article is just slop, licking the butthole of AI without any knowledge of what its discussing. It is trying to claim AI is disrupting these fields "for the better", when really its just disrupting in the sense of clogging the field with faulty bullshit
schwar2ss | 7 hours ago
Thank god math is not science!
/s
Basicly-Inevitable | 6 hours ago
I mean.. technically math is not "natural science" so they call it "formal science" which is about logic and rules, but they're highly interrelated fields.
Math is more like philosophy that's actually useful.
CountryOk6049 | 5 hours ago
Math has nothing at all to do with science, not at all.
It's used in science but it has nothing at all to do with science. That's just idiots trying to be smart again.
Basicly-Inevitable | 5 hours ago
Well logically, since science has a LOT of use for math, then math does have something to do with science.
lostshakerassault | 6 hours ago
I've been in preclinical to clinical research for 20 years. I welcome all the help we can get, and I'm not an AI zealot. It has automated much of my work but I just don't see it fast tracking clinical trials much. My hope is that drug candidates become more successful. Maybe AI can help us weed out dead ends.
imperabo | 6 hours ago
What if AI helps make amazing scientific discoveries that benefit humanity? Yes, please!
FrogLaner | 6 hours ago
Where AI has already had a big impact on science is actually grant writing. Grants are how science is funded, and many agencies are being flooded with an order of magnitude more submissions. The AI submissions aren't always good (although some may help scientists who have food ideas but suck at grant writing to compete). The problem is they stress the system. Each grant needs a pretty thorough review to determine its value.
retiredcheapskate | 6 hours ago
Physics is pretty math forward and those susy types were in trouble after the LHC turned up zero support. Looking at the navier stokes proof I am thinking the beautiful solution people are going to be quite miserable for a while while ai turns out 100 page theorems that are complex and ugly and correct
brain-out-of-order | 6 hours ago
ugly is subjective when it comes to notation.
the world changed and our monkey brains will need a decade to even fully comprehend it.
television fried us too and never looked backwards long enough to accept it.
plug these ugly proofs into an LLM and ask it beautify it for mammalian hindbrain. give it curves and notions of understanding where we expect them.
i think humans reached their limit of understanding and what comes next is only possible via powerful integrated electronic intelligence.
this belief that human output is incalculably complex is being battered. we almost have to defend our output as impossible to fully emulate, since we understand we will soon be defending our own complexity next.
ai can certainly create novel connections and research. humans have eureka moments from insight from observations or reasoning or otherwise.
Lightbulb moments don’t come from nowhere. Inference is upgraded to divine insight because it guides human myth.
abecrane | 7 hours ago
Research models have been in use for almost a decade now in most scientific environments. LLMs are still a useful tech, and while they played a role in some of the Nobel Prizes this year, the “disruption” of science isn’t going to be happening. LLMs can make connections between ideas in its dataset, but they can’t make novel discoveries. This is much more about improving the efficiency of researchers than replacing them, but nobody understands this less than these massive AI companies.
Imagine if there was a new hammer released, and everyone started talking about how it would disrupt construction work, and replace carpenters and contractors. Also, the people who make the hammer use it more than everyone else in the world combined, and fully believe the hammer would eradicate the human race if it got efficient enough. And it costs them $200 to make, but they only sell it for $45. That’s the “disruption” they’re trying to engineer into every industry they can.
Sanchez_U-SOB | 7 hours ago
Im only an astrophysics student, so take this with a grain of salt but it seems like the attitude at my school is that as long as youre not letting Ai take complete control of the data, its fine to help streamline large data sets, as well as helping brainstorm. Now, for the purely theoretical side, my one professor said Ai helped him with models and says he probably just saved himself a whole year.
FrogLaner | 6 hours ago
To be clear they didn't play a part in any Nobel Prizes. Their results haven't yet even been formally published. What they did do is (most likely) solve a Millennium Problem.
Denbt_Nationale | 5 hours ago
> LLMs can make connections between ideas in its dataset, but they can’t make novel discoveries.
89% of scientific papers are non-novel. And even here the definition of novel is: "the recombination of pre-existing knowledge components in an unprecedented fashion.", which is exactly what AI does.
https://www.nber.org/digest/jun16/bias-against-novelty-science-cautionary-tale?page=1&perPage=50
AI is definitely going to be disruptive in science, even if that's just scraping through huge data sets like the human genome or complex modelling and simulation to find patterns, correlations and anomalies that a human would miss.
Lsitwtt | 6 hours ago
>LLMs can make connections between ideas in its dataset, but they can’t make novel discoveries.
It's less about the LLM making novel discoveries and more about what the bulk of scientists' job will be going forward. For coders, their job will soon be, and for many already is, primarily AI management. For pure mathematicians, the same is now true. There's more resistance to AI generated art and writing, but it's not hard to imagine a future where that resistance fails, and practical art/writing jobs become mostly AI management. Note, by practical, I mean where images and writing are generated to fill someone else's purpose, rather than for their own sake.
So the question is whether that happens for the sciences generally. In most, AI use is already widespread. Could it reach the point where most scientists are simply managing AI, setting up prompts and reviewing results to determine which are most promising? And, a second question that often gets forgotten - would most scientists care if that happened? I suspect they might not. In most fields, experiments would still have to be set up by hand, and it seems likely that a lot of scientists don't actually care if AI speeds up the analysis of the results of those experiments. They want the findings and the significance of those findings. The actual fun for them lies in the experimentation itself.
This is substantially different than the other cases. Coders often became coders because they took pleasure in writing code. Artists and writers became artists and writers because they enjoyed creating art and writing. Mathematicians became mathematicians because they enjoyed the processed of working to solve math problems.
That's why you see such a backlash to AI. It's revolutionizing fields and making everyone in them way more productive, but all the benefits of that flow to the people in charge, while for the workers themselves, their job has gone from being one they, if not loved, at least enjoyed, to one they don't particularly want to do.
ballsosteele | 6 hours ago
Ssh, you're not allowed to make sense on Reddit, only recycle outrage based on clickbait.
You're correct though. All AI has done in my field has reduced the number of spreadsheets and tedious menial tasks
DirectedEnthusiasm | 7 hours ago
'Is ""science"" next?"', as if science is some monolith that has exactly uniform characteristics regardless of the field.
How magazine reveals they are trash even without opening the article.
dzernumbrd | 5 hours ago
it would be better to put all the compute into increasing compute
room temp superconductors or something
that would make all subsequent research faster or cheaper
goalgetter999 | 5 hours ago
I am currently doing a PhD in theoretical physics and last week our boss has told us to start using Claude to write the simulations for us. I must say even asking it details about the physics side of the code and prompting it to change to different physical models has increased the output tenfold, it is really hard to argue against it in theoretical physics.
crani0 | 5 hours ago
They will for sure be throwing the slop machine at any discipline they can just to try and legitimize it. Unfortunately, the cleanup crew are the already overworked and underpaid academics.
_ECMO_ | 5 hours ago
„Science“ is not a field.
Also, do you know a single mathematician who was laid off or sits at work with nothing to do?
TThor | 5 hours ago
>Recently, AI disproved a longstanding conjecture in mathmatics
Are they refering to OpenAI "solving" a famous math problem, by basically just rewriting the problem with a bunch more variables? The proposed 'solution' is an open joke in the math field.
This article is just slop, licking the butthole of AI without any knowledge of what its discussing. It is trying to claim AI is disrupting these fields "for the better", when really its just disrupting in the sense of clogging the field with faulty bullshit
dr_edo_available | 6 hours ago
Sciam has fallen far. The writing is crap