Dr. Jeff Beck, a computational neuroscientist, discusses Bayesian modeling in empirical inquiry, active inference, and the integration of internalist and externalist cognition. They also explore the challenges of using a Bayesian approach in multi-agent systems, the relationship between data cardinality and representation, and the idea of creating specialized agents. Additionally, they touch on representation learning in neural networks and the development of fully Bayesian and generative transformers.
Bayesian analysis is a normative and logical approach to empirical investigation.
Active inference involves building dynamic models in real-time for optimal experimental design and dynamic updating.
Steering emergent systems through macroscopic optimization requires defining goals for the system as a whole and avoiding rigid behavioral profiles.
Deep dives
Modeling Humans and Animal Behaviors as Bayesian Inference
The podcast episode explores the work of computational neuroscientist Jeff Beck, who studies how human and animal behaviors can be modeled as Bayesian inference. They are interested in how the brain encodes probability distributions and performs probabilistic reasoning.
The Importance and Power of Bayesian Analysis
The podcast highlights the significance of Bayesian analysis in empirical inquiry. It is considered a normative approach to empirical investigation and is valued for its principled and logical reasoning under uncertainty.
Finding the Right Model and Structure
The discussion delves into the challenge of finding the right model or structure when applying Bayesian analysis. It emphasizes the importance of having models that are intelligible to humans for effective communication and understanding.
Active Inference and Building Dynamic Models
The podcast introduces the concept of active inference, which involves building dynamic models in real-time. It explores the benefits of this approach, including optimal experimental design and the ability to dynamically update models based on information obtained from the environment.
Agents achieving homeostatic equilibrium
The podcast discusses the concept of agents achieving homeostatic equilibrium as a fundamental component of active inference. It highlights how active inference defines an agent as consisting of an inference engine, a prediction model, and a reward function. Instead of using a traditional reward function to motivate behavior, active inference replaces it with an expression of homeostatic equilibrium. This approach aims to create a more stable system by focusing on achieving balance and equilibrium rather than optimizing a specific reward function. By prioritizing homeostatic equilibrium, active inference avoids brittleness and provides a robust framework for agent behavior.
Steering emergent systems through macroscopic optimization
Another key point raised in the podcast is the idea of steering emergent systems through macroscopic optimization. The discussion highlights the importance of defining the target macroscopic behavior or goal for the system. It emphasizes the notion that goals and behavior emerge as properties of the system as a whole, rather than being explicitly programmed into individual agents. The podcast explores the challenge of steering autonomous agents with a fixed objective without introducing brittleness. It suggests different approaches, such as including desired behaviors as part of the agents' definition of self or deploying specialized agents to adapt behavior. By maintaining flexibility and avoiding rigid behavioral profiles, the podcast suggests that steering emergent systems through macroscopic optimization can lead to more adaptable and resilient systems.
Dr. Jeff Beck is a computational neuroscientist studying probabilistic reasoning (decision making under uncertainty) in humans and animals with emphasis on neural representations of uncertainty and cortical implementations of probabilistic inference and learning. His line of research incorporates information theoretic and hierarchical statistical analysis of neural and behavioural data as well as reinforcement learning and active inference.
https://www.linkedin.com/in/jeff-beck...
https://scholar.google.com/citations?...
Interviewer: Dr. Tim Scarfe
TOC
00:00:00 Intro
00:00:51 Bayesian / Knowledge
00:14:57 Active inference
00:18:58 Mediation
00:23:44 Philosophy of mind / science
00:29:25 Optimisation
00:42:54 Emergence
00:56:38 Steering emergent systems
01:04:31 Work plan
01:06:06 Representations/Core knowledge
#activeinference
Get the Snipd podcast app
Unlock the knowledge in podcasts with the podcast player of the future.
AI-powered podcast player
Listen to all your favourite podcasts with AI-powered features
Discover highlights
Listen to the best highlights from the podcasts you love and dive into the full episode
Save any moment
Hear something you like? Tap your headphones to save it with AI-generated key takeaways
Share & Export
Send highlights to Twitter, WhatsApp or export them to Notion, Readwise & more
AI-powered podcast player
Listen to all your favourite podcasts with AI-powered features
Discover highlights
Listen to the best highlights from the podcasts you love and dive into the full episode