Neuro-symbolic inventive reasoning

ARIZ, the algorithm for inventive problem solving, is a long sequence of reformulations that experts take years to master. Our doctoral work has studied how to orchestrate language models through this sequence, step by step, and how to anchor the generated concepts in real patents.

The symbolic layer matters: where a model gives an opinion, the contradiction matrix gives a recommendation that is invariant from one run to the next. We study how to combine both without losing either.