Designing ethical AI in the shadow of Hume’s guillotine

Abstract

Artificially intelligent systems can collect knowledge regarding epistemic information, but can they be used to derive new values? Epistemic information concerns facts, including how things are in the world, and ethical values concern how actions should be taken. The operation of artificial intelligence (AI) is based on facts, but it require values. A critical question here regards Hume’s Guillotine, which claims that one cannot derive values from facts. Hume’s Guillotine appears to divide AI systems into two ethical categories: weak and strong. Ethically weak AI systems can be applied only within given value rules, but ethically strong AI systems may be able to generate new values from facts. If Hume is correct, ethically strong AI systems are impossible, but there are, of course, no obstacles to designing ethically weak AI systems.