Artificial Intelligence

nil by Pei Wang, and comments by Joscha Bach

• Lack of Theory & Theoretical Nihilism: Fields of science or engineering are commonly guided by corresponding theories but AI seems to be an exception. A major obstacle is "theoretical nihilism" — facing the well-known difficulty in establishing a theory of AI, the research community as a whole has not made enough effort in this task.
• Look, ma, no hands syndrome: The "problem-oriented" attitude toward AI focuses on the problem-solving capability of a computer system, while does not care much for the underlying theory.
• If we have a theory, it is not considered AI anymore: The "theory of AI" is impossible by definition, since we cannot have a theory for "those processes we don’t yet understand" — as soon as we have a good theory for such a process, it is no longer considered as AI anymore.
• Theory of AI need to be both descriptive and normative: The theory of AI must cover certain mechanisms in the human mind (Human Intelligence; Descriptive), then generalize and abstract them (to be the General Intelligence; Descriptive and Normative both), and finally specify them in a computational form (to become the Computer Intelligence; Normative).
• Scientific Theory must have following properties:
  • Correctness: A theory should be supported by available evidence.
  • Concreteness: A theory should be instructive in problem solving.
  • Compactness: A theory should be simple.

• Library of AI tools: https://www.futurepedia.io/
• Generalist AI beyond Deep Learning
• AI Deception

Backlinks

• Eliezer Yudkowsky
• NAAMI
• Compression is AI
• Student outsmarts the teacher
• LessWrong
• Adversarial Attack
• AI And The Limits Of Language
• MIT 6.S191: Introduction to Deep Learning