Using Predicate Logic and LLM to prove Theorems, how far can we do it? How much is correct?

Lets start with the statement, “The kernel of the group homomorphism is a normal subgroup.” Lets try to use chatGPT to prove it?

Here is the first prompt where I tried Prolog first before the theoretical proof of Predicate Logic on which Prolog is based.

“Using prolog prove subgroup of an abelian group is abelian”

Here is a snippet of proof:

Similarly, we generated the prolog code for “Using prolog prove kernel of a group homomorphism is a normal subgroup”

But we need theoretical and mathematical proof.

Here is ChatGPT with the prompt “To prove that every subgroup of an abelian group is abelian, we will formalize the proof using predicate logic.”

Now, we prove, The kernel of the group homomorphism is a normal subgroup.

Can LLM and chatGPT prove more theorems using predicate logic?

Summary:

Using existential and universal quantifiers helps in theorem proving, but LLMs somehow can’t write full proofs using these quantifiers.

Till what we tested more data regarding predicate logic must be provided to LLMs in a way that machines can comprehend in a semi-supervised manner.

More to come…

Subscribe for updates..

Stay tuned..

Warm regards

Published by Nidhika

In the Futuristic with AI and Tech blog, Nidhika Yadav covers topics of and related to Future of World with Artificial Intelligence. She primarily talks about AI applications for good. She also talks about how AI can become harmful. She manages two independent blogs here, and one is hobby blog you can subscribe one or all of them. 1. Blog on Artificial Intelligence and future. https://nidhikayadav.org 2. In Blog on Global Issues and future, she covers important international issues and their future implications. https://nidhikayadav.com/ 3. Blog on cooking. This is a hobby blog. Here she describes some delicious innovations and nutritious food. https://nidhikasrecipes.com/ Do subscribe to one or all of them.

Leave a comment