I’m relatively sure that NNs are learning an implicit type theory anyway. But: 1. we don’t know which one; and, 2. you can only use fuzzy reasoning, not synth

I’m relatively sure that NNs are learning an implicit type theory anyway. But: 1. we don’t know which one; and, 2. you can only use fuzzy reasoning, not synthesize and compute results. I think the next breakthrough will be using statistical models to build effective type theories for the data we show them, but extracting them so we can “reason” explicitly.