Why AI is Harder Than We Think

Explore a thought-provoking keynote address from GECCO 2021 that delves into the complexities and challenges of artificial intelligence. Gain insights from Melanie Mitchell, Davis Professor of Complexity at the Santa Fe Institute, as she examines common fallacies in AI research and discusses why the development of advanced AI technologies has proven more difficult than anticipated. Learn about the cyclical nature of AI progress, the limitations of current approaches, and the need for more robust, general, and adaptable AI systems. Discover Mitchell's perspective on major open challenges in the field, including transparency, robustness, and the barriers to creating truly intelligent machines. Engage with topics such as self-driving cars, AI winters, wishful mnemonics, and the nature of intelligence in the brain. Conclude with a discussion on the potential dangers of AI and the role of evolutionary computation in addressing these challenges.

Introduction
Selfdriving cars
AI winter
Are we still optimistic
First step fallacy
Second step fallacy
Third step fallacy
wishful mnemonics
over attributions
other examples
intelligence in the brain
open questions
Major open challenges
Transparency and robustness
Barrier to AI
Questions
AI without ontologies
Openended evolution
Evolutionary computation
Dangers of AI