Could ML be trained to help mathematicians arrive at proofs to theorems by guiding them at the right direction? Can it find new worthy theorems?
I’m not a maths person, someone with a background in that can confirm or refute this, but afaics:
ML mainly works via looking for patterns. So that’s useful, ish, but to get to a point where it’s practically useful in this or a similar case, you need to have an in-depth understanding of the problem (eg you’re already generating results). You would need strong AI to allow a computer be useful before this pattern matching part. And as we’re not even vaguely close to strong AI atm, ML is currently of very limited use in most cases.
It seems like possibly a useful research subject, lots of people seem interested in it & there are loads and loads of papers on it. For example, cursory googling, all of these link to multiple papers:
In short, yes. Automated proof systems are not new. Neither are genetic approaches to proving theorems. Regarding ML in particular, it helps to remember that ML is effectively a statistical search process so it too can help with theorem generation and proof.
The literature is not small. Google Scholar gives 42,600 results for ‘automated theorem proving machine learning’ and ~272,000 for ‘automated theorem proving’.
Hope this helps.
This topic was automatically closed 182 days after the last reply. New replies are no longer allowed.