Mind Matters The Limits and Possibilities of Artificial Intelligence
May 30, 2024
Join George Montañez, an Assistant Professor of Computer Science, and William Dembski, a mathematician and philosopher, as they tackle the multifaceted world of artificial intelligence. Montañez dives into generative AI models, discussing how they encode relationships between text and images while highlighting the risks of model collapse. Dembski brings philosophical insights about human consciousness versus AI's limitations. Together, they explore digital immortality and the paradoxes of AI creativity, raising crucial questions about the future of humanity and technology.
AI Snips
Chapters
Books
Transcript
Episode notes
Correlation-Based Learning
- Generative AI models learn by encoding relations between text and images and interpolating from huge datasets.
- Their apparent rationality depends heavily on human-designed architectures, biases, and training choices.
Human Fingerprints In AI
- Large models' reasoning is highly parasitic on human creativity and design choices in architecture and hyperparameters.
- Their outputs simulate rationality but lack independent human-like rationality or consciousness.
Non-Computable Limits
- There exist mathematically non-computable problems (e.g., the halting problem) that algorithms cannot solve.
- If aspects of human minds are non-computable, silicon machines cannot fully reproduce creativity, sentience, or understanding.
