Maybe it’s a fantasy but if they could replace humans with machines that never bargained and went on holiday, they would in a heartbeat. And that’s what they want to do. Maybe they will fail, maybe the technology will develop. It can do many things today that it couldn’t so a few years ago. It’s already solved more difficult mathematical proofs than any human alive today for example.
Yeah, that’s not true. The math proofs it has solved are not “more difficult than any solved by a living person”. Many of the proofs AI has solved are corner case conditions that are not being actively worked on by mathematicians. In a trivial example. Image a problem like:
x = n + 25
AI says, hey, let’s assume n=3.
AI solved the proof that 3 + 25 = 28.
Then we get an article posted about how AI solved the impossible problem of n + 25
Now, it can be helpful to have these less general proofs solved. It can help mathematicians in creating a more general proof (what they are usually focused on).
But, as far as I’m aware, these are most of the contributions of AI so far in the field of mathematics. You’re falling for the hype around it. It’s mostly filling in gaps for corner cases that are not actively being worked on.
Which makes sense. Because it’s just a text generation algorithm that predicts the most likely characters. It’s using training data for the n = 3 case that it learned from another already solved problem and then injecting it into the unsolved problem. Sometimes is spits out something that makes sense, a human being has to verify it, and then a corner case for mathematical proof is “solved”.
It’s not capable of thinking or rationalizing to anything new. It is filling in gaps in a field of research that is not primarily focused on “gap filling”.
It’s why it’s “having success” in Math but not Physics. Mathematicians focus on general proofs that apply under all conditions. So, the field inevitably has a lot of problems that could be solved by a lot of brute force for specific conditions.
In Physics, scientist are constantly trying to “fill gaps”. They work with real world measurements and observations. Any new “gap” is constantly being tested. AI has no gaps to fill with its brute force way of “thinking”. And it’s completely incapable of coming up with theories or hypothesis on a level a human does.
This is why it’s not solving ground breaking mathematics proofs and why it is mostly useless in replacing humans in the scientific process. Because it isn’t actually thinking.
Sorry, rant. But this is something that bothers me and you hit a nerve.
Sorry the wording of that phrase was too ambiguous. I meant it had solved more proofs than any human had in their lifetime, and those were at least difficult enough to stump humans for years or decades. I didn’t mean that they were more difficult than proofs that humans had solved, simply that the AI had solved a higher quantity of them, and that, on average, they would be considered difficult. At least difficult enough to stump humans for a pretty good stretch of time.
Sure. That’s fine. But I think you still misunderstand exactly what is being solved in most of those cases. It’s not about quantity. If it was then all mathematicians would be spending time on filling “gaps”. Mathematicians aren’t focused on filling in gaps as a lot of the time that type of brute force work is not meaningful to the field of mathematics. They are focused on proving (or disproving) axioms that are foundational to the field. AI has not proven or disproven any mathematical axiom.
The biggest proof it’s ever provided was for “Erdos Problem”. Which is essentially just a fun “dot placement optimization problem” that Erdos came up with, was never taken too seriously, and was difficult enough that spending time on it in academics would likely get you laughed at.
It’s formal proof yields almost nothing to the field.
On top of that, it was “solved” by an AI math startup with closed source software. There is literally no confirmation that it was AI solving it organically or if the mathematicians themselves solved it during developing the AI and saw a means of getting headlines and more capital investment.
Please, be more skeptical. I double majored in Math and CS and have worked in the field for over a decade. There are applications for AI that are helpful. But, right now, people are falling for a lot of the hype and lies that are being pushed around to attract capital investment. You’re not an investor, you can be smarter and more skeptical than that.
In a decade we’re going to have a lot of reflection on this time and the lies being told about the usefulness of AI.
We live in a world where one of the most valued companies in the world pretends it’s AI robots are real and not just being remote operated by a human. You think AI startups aren’t giving credit to AI for the work of humans? They all are. Because it gets them investment.
A mostly fun and non helpful math problem gets solved by a human. The math community enjoys it. Laughs a little at the guy who spent too much of his life solving it. And no one outside the community hears about it.
But if you say “AI did it”. You get millions of dollars in investment.
I don’t have evidence a human solved it. But they also don’t have evidence AI solved it either. And all financial incentives right now are in saying “AI did it”. That is why you should be extremely skeptical of all these things you hear about the capability of AI.
Maybe it’s a fantasy but if they could replace humans with machines that never bargained and went on holiday, they would in a heartbeat. And that’s what they want to do. Maybe they will fail, maybe the technology will develop. It can do many things today that it couldn’t so a few years ago. It’s already solved more difficult mathematical proofs than any human alive today for example.
Yeah, that’s not true. The math proofs it has solved are not “more difficult than any solved by a living person”. Many of the proofs AI has solved are corner case conditions that are not being actively worked on by mathematicians. In a trivial example. Image a problem like:
x = n + 25AI says, hey, let’s assume n=3.
AI solved the proof that
3 + 25 = 28.Then we get an article posted about how AI solved the impossible problem of
n + 25Now, it can be helpful to have these less general proofs solved. It can help mathematicians in creating a more general proof (what they are usually focused on).
But, as far as I’m aware, these are most of the contributions of AI so far in the field of mathematics. You’re falling for the hype around it. It’s mostly filling in gaps for corner cases that are not actively being worked on.
Which makes sense. Because it’s just a text generation algorithm that predicts the most likely characters. It’s using training data for the
n = 3case that it learned from another already solved problem and then injecting it into the unsolved problem. Sometimes is spits out something that makes sense, a human being has to verify it, and then a corner case for mathematical proof is “solved”.It’s not capable of thinking or rationalizing to anything new. It is filling in gaps in a field of research that is not primarily focused on “gap filling”.
It’s why it’s “having success” in Math but not Physics. Mathematicians focus on general proofs that apply under all conditions. So, the field inevitably has a lot of problems that could be solved by a lot of brute force for specific conditions.
In Physics, scientist are constantly trying to “fill gaps”. They work with real world measurements and observations. Any new “gap” is constantly being tested. AI has no gaps to fill with its brute force way of “thinking”. And it’s completely incapable of coming up with theories or hypothesis on a level a human does.
This is why it’s not solving ground breaking mathematics proofs and why it is mostly useless in replacing humans in the scientific process. Because it isn’t actually thinking.
Sorry, rant. But this is something that bothers me and you hit a nerve.
Sorry the wording of that phrase was too ambiguous. I meant it had solved more proofs than any human had in their lifetime, and those were at least difficult enough to stump humans for years or decades. I didn’t mean that they were more difficult than proofs that humans had solved, simply that the AI had solved a higher quantity of them, and that, on average, they would be considered difficult. At least difficult enough to stump humans for a pretty good stretch of time.
Sure. That’s fine. But I think you still misunderstand exactly what is being solved in most of those cases. It’s not about quantity. If it was then all mathematicians would be spending time on filling “gaps”. Mathematicians aren’t focused on filling in gaps as a lot of the time that type of brute force work is not meaningful to the field of mathematics. They are focused on proving (or disproving) axioms that are foundational to the field. AI has not proven or disproven any mathematical axiom.
The biggest proof it’s ever provided was for “Erdos Problem”. Which is essentially just a fun “dot placement optimization problem” that Erdos came up with, was never taken too seriously, and was difficult enough that spending time on it in academics would likely get you laughed at.
It’s formal proof yields almost nothing to the field.
On top of that, it was “solved” by an AI math startup with closed source software. There is literally no confirmation that it was AI solving it organically or if the mathematicians themselves solved it during developing the AI and saw a means of getting headlines and more capital investment.
Please, be more skeptical. I double majored in Math and CS and have worked in the field for over a decade. There are applications for AI that are helpful. But, right now, people are falling for a lot of the hype and lies that are being pushed around to attract capital investment. You’re not an investor, you can be smarter and more skeptical than that.
In a decade we’re going to have a lot of reflection on this time and the lies being told about the usefulness of AI.
We live in a world where one of the most valued companies in the world pretends it’s AI robots are real and not just being remote operated by a human. You think AI startups aren’t giving credit to AI for the work of humans? They all are. Because it gets them investment.
A mostly fun and non helpful math problem gets solved by a human. The math community enjoys it. Laughs a little at the guy who spent too much of his life solving it. And no one outside the community hears about it.
But if you say “AI did it”. You get millions of dollars in investment.
I don’t have evidence a human solved it. But they also don’t have evidence AI solved it either. And all financial incentives right now are in saying “AI did it”. That is why you should be extremely skeptical of all these things you hear about the capability of AI.