In this edition of QuEST, Dr. Lenat, CEO of Cycorp, talks about his AI called Cyc and trustworthiness of AI systems.
Key Moments in the video include:
Brief history of Cyc project and origins
Cyc demo - Can an upside down coffee cup hold hot coffee?
Cyc represents things in predicate calculus
Power of statistical machine learning
Brittleness of AI systems
Foundation of commonsense knowledge for AI systems
Harnessing LLMs to get as close to automatic natural language understanding as possible
First order logic is not expressive enough
CycL as the language of Cyc
Trustworthiness of AI
Details of building Cyc
Audience Questions:
Could you talk a little bit more about how measures of trust would lead to better trustworthiness of AI?
Are you willing to consider flipping what you just said on its head since these LLMs can create code, are you willing to let in a small experiment then generate new pieces of Cyc?
When you asked about the knowledge acquisition for Romeo and Juliet, you said “well, it’s mostly done manually, but we want to do something more interesting” you wanted to harness the power of NLP for what I understand would be knowledge acquisition, could you elaborate on that?
More