Thoughts about knowledge representation

From: Gene Michael Stover

To: “usual guys”

When: 2021 December 11

This file is at https://36f10c18-a119-4e7c-9f7f-484bb5db26f2.s3.us-west-2.amazonaws.com/4724dfb1-8d88-4764-94f1-e95de7446334.html and https://cybertiggyr.com/4724dfb1-8d88-4764-94f1-e95de7446334.html.

Greetings, programs! (I think that should become my trademark opening.)

I'm sure that I sent you guys an early version of a thoughtful essay rant about how artificial intelligence practitioners should be sure to return to knowledge representation in the near future. I may also have talked about some implementation ideas for a symbolic, general purpose knowledge representation system.

Since then, I've been reading a lot about knowledge representation & had a realization.

If a person (me) wants a problem solver in a box, then traditional databases are fine. Yeah, we need more data, need to think about the schema, need to combine databases, so it's a tall order, but it's still traditional databases.

For example, if I want a problem solver in a box that finds the cure for cancer (or cures for many kinds of cancer), then I don't need a database schema that can represent “Jack used to wish that Jill thought he was hotter than Frank”, nor do I need an A.I. algorithm that can infer from that Jack's motivations for that wish & speculate why Jack no longer wishes it.

For problems like finding new drugs, I suspect we need...

• Big-assed, standard, shared, high-quality databanks of All the Data!
  This is a work in progress, I'm sure, though I fear that private profits impede the shared part.
• The machine learning techniques that are vogue these days.
  (They find patterns in data.)
  I'm confident this is being done.
• The automated search algorithms from symbolic A.I. to speculate on chemical combinations that are good candidates for drugs.
  This, or something better, is probably being done.
• The automated reasoning algorithms from symbolic A.I. to detect missing data that would probably be very useful to acquire.
  I don't know if this is being done.

Since I would like to make a contribution, even if it's tiny such as proof-reading documentation for machine-learning libraries for Python programmers, this realization leaves me wondering what I could do to help.

End.