Vicente Raja & Jeff Beck ~ Active Inference Insights 007 ~ The Markov Blanket Trick Debate
Feb 15, 2024
auto_awesome
Vicente Raja and Jeff Beck explore the controversy surrounding 'The Markov Blanket Trick' paper. They discuss the concept of Markov Blankets and its implications, the debate on perception, Bayesian inference and the Free Energy Principle, criticism and clarity in the concept of the Markov Blanket, perspectives on inference in psychology and neuroscience, the Free Energy Principle and self organization, Jeremy England's work on complex formations, and the importance of instrumentalism and realism in science.
The choice of blankets in the FEP is based on the goal of labeling and simplifying complex systems, but the criteria for selecting significant blankets is not always explicitly addressed in the literature.
The selection of blankets for systems in the FEP literature raises philosophical questions about the presuppositions and ontological implications of labeling and boundary detection.
The FEP assumes that perception is a form of inference, although there are ongoing discussions on how perception can be understood without explicit Bayesian inference.
Vicente argues that perception should not rely on prior knowledge to make sense of impoverished stimulation, advocating for an ecological psychology framework based on the lawful relationship between patterns of life and the environment.
Deep dives
The Role of Blankets in the Free Energy Principle
The podcast discusses the concept of blankets in the context of the Free Energy Principle (FEP) and active inference. Blankets serve as a methodological tool for labeling and simplifying systems. The choice of a blanket depends on the needs of the model, and the FEP literature often selects blankets that align with intuitive boundaries of ordinary things. The FEP assumes that things are what they are by definition, even if the boundaries do not align with intuition. However, the choice of blankets is not always clear and may vary depending on the specific system of interest.
Blankets and Object Labels
The choice of blankets in the FEP is based on the goal of labeling and simplifying complex systems. By selecting specific blankets, the FEP allows for the application of Bayesian inference, which simplifies dynamics and provides predictions about the system's behavior. However, the issue of how blankets form and the criteria for selecting significant blankets is not always explicitly addressed in the literature. The FEP assumes that the chosen blanket statistics define the phenotype or object type, allowing for simplification and compression of the overall dynamics.
The Philosophical Implications of Blanket Selection
The selection of blankets for systems in the FEP literature raises philosophical questions about the presuppositions and ontological implications of labeling and boundary detection. The explicit criteria for blanket selection and the alignment with intuitive boundaries for ordinary things varies. The choosing the right blanket for any given system involves considerations of parsimony and simplification. The choice of blankets depends on the specific theoretical perspective and the ultimate aim of the model. However, sometimes the alignment of blankets with intuitive boundaries in the literature is not explicitly addressed.
Issues of Perception and Inference
The FEP assumes that perception is a form of inference, allowing for inference from impoverished sensory input to the real environment. Ecological psychologists propose that perception is more about the rich information exchange between the organism and the environment, rather than a process of internal states inferring external states. The FEP's as-if description suggests that systems perform Bayesian inference, although there are ongoing discussions on how perception can be understood without explicit Bayesian inference. The relationship between perception, blanket selection, and the underlying philosophical presuppositions requires further exploration and clarification within the FEP framework.
The importance of prior knowledge in perception
Vicente argues that perception should not rely on prior knowledge to make sense of impoverished stimulation. He asserts that ecological psychology offers an alternative framework where perception is based on the lawful relationship between patterns of life and the environment. His work focuses on operationalizing the concept of resonance in the brain to study the dynamics of perception.
Pushing the limits of the Bayesian brain hypothesis
Jeff explores the idea that the brain operates with multiple models rather than a single prior. He believes that the brain uses various models or statistical representations to understand and predict the world, reasoning by analogy with different models explaining different aspects of the data. His current research involves constructing a brain-like Bayesian graph neural network and exploring the interactions between these models in a computational intensive manner.
Convergence in scientific approach
Despite their different perspectives, Jeff and Vicente find common ground in their commitment to empirical science. They both emphasize the instrumental nature of their theories and recognize the need for simplifications and assumptions in scientific models. They agree that science is not about absolute truth, but about prediction, data compression, and pragmatic approaches to understanding the world.
In 2021, Raja et al. published ‘The Markov Blanket Trick’, which argued that “the FEP [Free Energy Principle] is not the general principle it is claimed to be, and that active inference is not the all-encompassing process theory it is purported to be either.” This led to a flurry of responses from within the active inference community, with commentaries published by Friston, Ramstead and Albarracin & Pitliya, amongst others. In this episode of Active Inference Insights, we unpick this controversial paper with the help of its primary author, Vicente Raja, and Jeff Beck, a computational neuroscientist studying probabilistic reasoning in humans and animals.
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