

Alex Kiefer ~ Active Inference Insights 014 ~ Representations, Predictive Coding, Teleology
May 4, 2024
Alex Kiefer, a philosopher and machine learning expert, delves into the fascinating interplay of philosophy, cognition, and technology. He discusses the critical role of representations in understanding mental processes and how Bayesian perspectives reshape our views on reality and consciousness. Kiefer also explores predictive coding, the dynamics of active inference, and their implications for beliefs and desires. Finally, he highlights the journey of merging computational science with philosophical inquiry, advocating for a deeper understanding of consciousness.
AI Snips
Chapters
Transcript
Episode notes
From Art to Computational Neuroscience
- Alex Kiefer started with a fine arts background before moving into philosophy and then computational neuroscience.
- He built Helmholtz machines to concretely understand cognition, beyond theoretical philosophy.
Exploitable Structural Resemblance Defines Representation
- Representation arises from exploitable structural resemblance between internal models and hypothetical worlds.
- Representations do not require direct resemblance to the actual external world, enabling internal simulations and hallucinations.
Truth Conditions in Generative Models
- Generative models must partially track truth to be useful but perfect veridicality is unnecessary and inefficient.
- Structural resemblance allows truth conditions to be a matter of degree rather than binary true/false.