Questions to Guide the Future of Artificial Intelligence Research

By Jordan Ott et al

Published on March 10, 2020

Read the original document by opening this link in a new tab.

Table of Contents

1. Key Points
2. Introduction
3. Can intelligence be discretized?
4. Why are we so dense?
5. Can you turn the noise down?
6. Where's my reward?

Summary

The field of machine learning has focused, primarily, on discretized sub-problems of intelligence. This article proposes leading questions to guide the future of artificial intelligence research, emphasizing the amalgamation of algorithmic and observational findings from neuroscience and machine learning. It discusses the limitations of gradient descent, computational principles from the brain, shortcomings of classical AI systems, understanding parts in the context of the whole, and the importance of approximating behavior. The document explores coding schemes for neural networks, robustness to noise and adversarial attacks, invariant representations in the brain, and robust properties from biology in achieving intelligence.