Open Problems In AGI

  1. Limited Experience Problem - Many AI learning methods require huge quantities of data, but humans require few (some times as little as one) example(s) to learn a thing.
  2. Control Problem - How to prevent AGI from doing something?
  3. Consciousness Problem - What is the nature of consciousness, and what algorithm will allow us to reproduce it?
  4. Knowledge Transfer Problem - How do we transfer learning from one task to another.
  5. Ethics Problem - What ethics ought we give our AGI?
  6. Symbol Grounding Problem - How abstract symbols can be grounded in empirical experience.
  7. Catastrophic Forgetting Problem - Computer programs that learn to perform tasks also typically forget them very quickly.
  8. Black Box Problem - An understanding of what a neural network is doing can be difficult and often only guessed at.
  9. Bias Problem -
  10. Cost of Errors -
  11. Entanglement Problem -
  12. Brittle Dependencies -
  13. Freshness -
  14. Hidden Feedback Loops -

https://ai4life.github.io/problems/

A documentary based on Ray Kurzweil’s book, The Singularity Is Near.

References

Unsolved Problems in AI Rules of Machine Learning: Best Practices for ML Engineering