Sean Carroll's Mindscape: Science, Society, Philosophy, Culture, Arts, and Ideas 68 | Melanie Mitchell on Artificial Intelligence and the Challenge of Common Sense
Oct 14, 2019
In this engaging discussion, Melanie Mitchell, a computer scientist and complexity researcher, delves into the perplexities of artificial intelligence, highlighting its struggles with common sense. She explores why AI excels in games but falters in real-world tasks, like driving. The conversation touches on the challenges of teaching AI fundamental concepts such as causality and object permanence and debates the contrasting methods of rule-based versus deep learning systems. Additionally, ethical concerns surrounding AI, including biases and societal impacts, are thoughtfully examined.
AI Snips
Chapters
Books
Transcript
Episode notes
Exponential Curve Limitations
- Exponential curves are rare in nature; many initially appear exponential but then plateau.
- Don't rely solely on curve fitting for AI predictions.
Perceptron Learning
- Perceptrons, early neural networks, could learn from examples by adjusting input weights.
- However, they lacked hidden layers to form complex representations, limiting their learning capacity.
Neural Networks and Backpropagation
- Neural networks have hidden layers enabling complex representation learning.
- The backpropagation algorithm allows networks to learn by apportioning errors through layers.
