a lot of our inspiration for this approach is based on what’s been going on in math and ai research, especially with the recent news where openai apparently solved erdos’ planar unit distance problem. from what i understand from the ai + math field, ai is being used to search across a really diverse solution space, but for each promising candidate result, a team of mathematicians have to formally verify the solution
as i understand it, mathematicians have been using the functional programming language, Lean, for a while now, to formally verify mathematical proofs. we’re trying to take the same approach, with patent language (and i suspect, this approach could extend to other legal fields). and then i think formal verification could be applied in contexts of program repair or program synthesis, while using ai to explore the patent language space, formally verify, with each iteration
cool post! it's funny how many things in this world are naturally graphs. i think it's neat how, especially in biology, a lot of high-dimensional objects, like protien sequences, converge onto lower-dimensional representations, like protein structures.
i did neuroscience for grad school, and i was always amazed by how often complex neural activity could be well represented by lower dimensional representations--clean manifolds, attractor dynamics, etc. i think, in general, biology (evolution) doesn't penalize against redundancy too hard (hence things like genetic drift, neutral theory of evolution, etc.).
anyway, super cool stuff. agree with you that probs more useful to explore the search space via 'less natural' structures, given how forgiving evolution is to redundancy. probs where the most information can be found
a lot of our inspiration for this approach is based on what’s been going on in math and ai research, especially with the recent news where openai apparently solved erdos’ planar unit distance problem. from what i understand from the ai + math field, ai is being used to search across a really diverse solution space, but for each promising candidate result, a team of mathematicians have to formally verify the solution
as i understand it, mathematicians have been using the functional programming language, Lean, for a while now, to formally verify mathematical proofs. we’re trying to take the same approach, with patent language (and i suspect, this approach could extend to other legal fields). and then i think formal verification could be applied in contexts of program repair or program synthesis, while using ai to explore the patent language space, formally verify, with each iteration