Some Amateur Analysis of the Consequences of the Current AI-math Fever

Disclaimer:  I'm Not a mathematician, merely some guy who's interested in Mathematical linguistics. 

Here the term "AI-math fever" refers specifically to the narratives of the form "AI is gonna replace human mathematician", "AI is gonna revolutionize mathematics", "It shall be allowed to use AI (unrestrictedly) to do mathematical researches." etc..  

There are at leat three problems along with these narratives: 

1. If a mathematical probelm is not solved, we shall first of all ask why it is not solved. It's absolutely absurd to hype over some results without first knowing why the probelm is not solved. 

2. That the problems are not solved does not indicate that nobody has been working on the probelms. It's also well-known that these models are trained on large amounts of data produced by human. Human mathematicians who are working on the unsolved problems are human. Thus, the models are trained on (at least some) materials produced by them without crediting them. 

3. If this unethical use of AI is accepted by the mathematical community (actually any community),  those who have access to better models will gradually "monopolize the game", then it seems the whole mathematical enterprise, if having any universal value at all, would lose its universal value and be turned into some sort of monetary and power game, controlled by a small circle of clever theives.  

While my analysis might just be some layman's guesses. The Tech Entrepreneur Chomba Bupe, who's well aware of the field and its tricks, pointed out something more important: 

Anyone who understands reinforcement learning (RL) process knows exactly what's going on here. OpenAI is abusing the maths community to keep improving their "internal" model. They are trying to absorb the maths community as a mere effective & cheap verifier for their RL loop. [Source]

Thus, it seems to me, the best the math community (in fact any community) could do at the moment is simply, as many said, to ignore this absurd fever. 

Update: 

Some issues discovered: [Source: Gro-Tsen]  Thread-link 


Comments

Post a Comment

Popular posts from this blog

Some Sidenotes upon My Tentative Proof of Chomsky(1956)'s Claim concerning FSG.

The Claim : if S has an m -termed dependency set, then at least 2^m   states are necessary for the FSG that generates L. Here is The proof . It's proven based on the Myhill-Nerode Theorem .  I'm working on this simply because:  It's a crucial point in understanding the whole motivation of the development of transformational grammar.  Simply put, for a FSL contains m dependencies, the FSG that could generate it would need at least 2^m such states.  In other words, a FSG with 2^m states could at most generate a FSL with m dependencies.  More generally,  for any finite state grammar, the language it generates shall contain finite number of dependencies. Natural language Grammar, however, has no such upper bound.  It is for this reason that natural language could not be adequately described by the type of FSG defined in Chomsky(1956).  My interactions with people in the field (of linguistics) informed me that there is systematic ignorance of the ...
Read more

Some thoughts on classic optimality theory(Prince & Smolensky 1993/2004)

[This is a term paper of mine, I'm not that familar with OT, thus I cannot guarantee that similar proposals had not been made before , comments are welcome .] Basic structure of classic optimality theory   Generation Function : G(input) = {C1, C2,..., Cn}  Evaluation Function :  E({C1, C2, C3,..., Cn}, Con) = Ck (1<=k<=n);                                        Ck = <In_k , G(in_k)>, In_k is the kth input, 1<=k<=n         To be more precise, G-function could be considered as a function of the form P(g(x), x), in which x is input, g is a generation function that generates an infinite number of outputs,  and P is a pairing function which generates an order pair called candidate, in which an input is paired with its corresponding output. The G-function thus generates an infinite number of candidates, which are inputs to the E-function...
Read more