Active Inference

• Identifying indicators of consciousness in AI systems

  mentions
• There is no self-evidence: A physics of emptiness realisation

  cites
• Active inference on discrete state-spaces: a synthesis

  mentions
• Koan Battery: Measuring Reflective Mode Accessibility in AI

  cites
• World models, artificial general intelligence and the hard problems of life–mind continuity: toward a unified understanding of natural and artificial intelligence

  associated_with
• Johnson Vasocomputation 2023

  implements

• Karl Friston

  associated_withintroduces
  Author of the free energy principle framework; central thinker in the paper.
• Giovanni Pezzulo

  associated_withstudies
• Michael Levin

  associated_with
• Michael Edward Johnson

  studies
  Author of the vasocomputation paper; researcher at Symmetry Institute (QRI) studying consciousness, active inference, and Buddhist phenomenology.
• Dalton Sakthivadivel

  studies
• Karl J. Friston

  studies
• Ruben Laukkonen

  studies
• Maxwell J. D. Ramstead

  introduces
• Adeel Razi

  studies
• Michael D. Kirchhoff

  introduces
• Sandved Smith et al.

  authored
  Authors of Computational phenomenology framework (2021) used to ground meta-awareness scoring dimension.
• Zahra Sheikhbahaee

  studies
• Beren Millidge

  studies
  Proposed deep active inference as variational policy gradients; extends active inference to high-dimensional problems

• Koan Battery

  associated_with
  Assessment framework for measuring introspection and self-observation in LLMs; grounded in Janus's architectural theory.
• Gradient Descent on Variational Free Energy

  implements
  Process by which neuronal dynamics minimize free energy; produces empirically observable neural phenomena.
• Belief Propagation

  implements
  Inference mechanism underlying active inference; updates posterior beliefs via gradient descent on free energy.
• Mean-Field Approximation

  uses
  Variational technique used in active inference to tractably compute posterior beliefs.
• Expected Free Energy Minimization

  implements
  Minimizing expected free energy for planning, decision-making, and action selection.
• Abstract Rule Learning Paradigm

  implements
  Experimental simulation paradigm where agents learn a rule mapping central cue color to correct response location
• Bayesian Inference

  associated_with
• Softmax policy selection

  implements
  Selecting policies using a softmax (normalized exponential) function of negative expected free energy.
• Bayesian Smoothing

  implements
  State estimation that combines prior expectations with likelihood; updates informed by past and future states.
• Computational fMRI

  implements
  Application of active inference to fMRI data; cited as prior use of the framework

• Generative Model

  associated_withimplements
  Agent's internal probabilistic model of environment; enables belief inference about hidden states given outcomes.
• Expected Free Energy

  associated_withimplements
  Free energy expected under future outcomes; guides policy selection via epistemic and extrinsic value.
• World Models

  associated_with
  Theme issue context: relates to internal models of environment, central to consciousness and cognition across substrates.
• Homeostasis

  associated_withgates
  Biological principle whereby agents maintain sensations within hospitable range; basis for active inference motivation.
• Self-Evidencing

  associated_withimplements
  Concise framing of action-perception cycle whereby agents minimize surprise through perception and action.
• Markov Blanket

  associated_withimplements
  A statistical partition of states that separates internal states from external hidden states; fundamental to self-organization in the paper.
• Vasocomputation

  extendsimplements
  Unifying framework proposing that Buddhist tanha operates through vascular smooth muscle cells as the brain's compression/prediction infrastructure.
• Prior Preferences over Outcomes

  associated_withimplements
  Replaces explicit reward signal in active inference; encodes agent's preferred observations independent of environment.
• Collective Intelligence

  associated_with
  Recognition that selves are composite systems of competent parts; all intelligences are higher-level selves made of cells or components.
• AI: a Bridge toward Diverse Intelligence and Humanity's Future

  mentions
  The source paper under extraction — a philosophical essay by Michael Levin arguing that AI debates neglect deeper questions about diverse intelligence, developmental biology, and humanity's future
• Active Inference, Curiosity and Insight (Friston et al., 2017)

  introduces
  The primary source paper being extracted
• free energy

  implements
  Thermodynamic potential ΔF = ΔE − TΔS; domain walls form if ΔF < 0
• Biological Computationalism

  contradicts
  Core theoretical framework: consciousness requires hybrid (discrete + continuous), scale-inseparable, metabolically embedded computation distinct from von Neumann architecture.
• in-context learning (ICL)

  supports
  Test-time adaptation from prompt or retrieved context with no parameter updates.
• Selflets

  associated_with
  Levin's model of continuous cognition as series of frames, each ~100-300ms thick; each Selflet is a temporal agent separated from others by time.
• Confabulation

  associated_with
  A form of cognitive plasticity where minds actively modify and reinterpret memory data to preserve psychological coherence; reframed as adaptive rather than pathological.
• Reinforcement learning (RL)

  associated_with
  Machine learning paradigm where agents learn to maximize cumulative reward through interaction.
• Epistemic Exploration

  cites
  Bayes-optimal exploration driven by uncertainty minimization; natural behavior in active inference without handcrafted mechanisms.
• Empowerment

  associated_with
  Information-theoretic quantification of options available to an agent; functional measure of affordance change.
• Prediction Error

  implements
  Role in optimizing sensory states; unified treatment shows value-learning and perception share error-minimization principle.
• Surprise Minimization

  implements
  Core principle: acting to maximize value is equivalent to minimizing surprise by sampling environment to conform to expectations.
• Metacognitive Particle

  implements
  System that encodes beliefs about a subset of its own internal states; prerequisite for the emergence of the separation prior
• Self-Evidencing Brain

  associated_with
  Hohwy's (2016) characterization: brain acts to maximize its own model evidence; consistent with active inference summary
• Three-Level Generative Model of Mindfulness

  implements
  Formal active inference model with perceptual, attentional, and meta-awareness states implementing mindfulness

+10 more

• Active Inference

  members_of

• Active inference agents can learn their own reward function (prior preferences) by interacting with the environment, bypassing the need for an explicit reward signal.

  supports
  Abstract and §3, preference learning section.
• Natural exploration-exploitation trade-offs emerge automatically from expected free energy minimization without hyperparameter tuning.

  supports
  Active inference achieves Bayes-optimal arbitration between exploration and exploitation without handcrafted mechanisms like ε-greedy.
• Agents perceive by minimizing variational free energy to ensure model consistency with past observations and act by minimizing expected free energy to make future sensations consistent with preferences.

  supports
  Formalization of perception-action cycle integrating inference and decision-making.
• Behavior prescribed by active inference dynamics is approximately Bayes-optimal.

  supports
  Process theory outcomes produce normatively sound decision-making.
• Curiosity, insight, decision-making, and diverse phenomena can all be accommodated by a single imperative: minimization of expected free energy (resolution of uncertainty).

  supports
  Central thesis of the paper unifying cognitive phenomena under one objective function

• Free Energy Principle

  associated_withextends
  A foundational variational principle from statistical physics that formalizes how self-organizing systems maintain structural integrity and adapt to their environment by minimizing free energy—a mathematical bound on surprise or prediction error. Originally developed by Karl Friston, the framework unifies action, perception, and learning as processes of active inference, where systems both update internal models of the world and act upon it to reduce the divergence between predictions and observations.
• Tame Technological Approach To Mind Everywhere

  associated_withcites
  A conceptual framework for understanding cognition and intelligence across diverse substrates—including evolved biological systems, artificial systems, and bioengineered systems—using empirically-grounded, gradualist approaches. TTAME enables comparative analysis of mind-like phenomena regardless of the physical or biological substrate in which it emerges, facilitating cross-disciplinary study of unconventional intelligences.
• Enactivism

  associated_withextends
  Theoretical approach treating cognition and self as emergent from embodied interaction; foundational to the paper's 'selfless self' model.
• Empathic Active Inference

  extendsrelated_to
  Matsumura et al.'s extension of active inference to include others' welfare in the generative model
• Stress Care Intelligence Loop

  associated_with
• Variational Bayesian Inference

  implementsuses
  A method for approximate Bayesian inference that optimizes a variational lower bound (ELBO) on log evidence.
• Autopoiesis

  associated_with
  Maturana-Varela principle of self-maintaining systems that organize themselves through internal feedback; extended here to biological, technological, and hybrid systems.
• Diverse Intelligence

  associated_with
  Research program studying intelligence at multiple scales and substrates; proposed as relevant to implications of mnemonic improvisation.
• Reinforcement Learning

  contradicts
  Alternative framework for agent behavior; based on reward maximization rather than free energy minimization.
• Self-Prior

  extends
  The key novel contribution: an internal model that learns the density of familiar multisensory experiences and drives mark-removal behavior through mismatch with the free energy principle
• Contemplative AI

  uses
  The paper's primary proposed framework embedding contemplative wisdom into AI alignment
• Partially Observable Markov Decision Process (POMDP)

  implements
  Modeling framework for discrete state-space decision-making under uncertainty, used as generative model in active inference.
• Hierarchical Precision Modulation Framework

  extends
  Novel framework introduced by this paper: three-level generative model (perception, attention, meta-awareness) for formalizing consciousness of one's own attentional states.
• Principles of Vasocomputation

  extends
  Mike Johnson's 2023 framework unifying Buddhist phenomenology, Active Inference, and physical reflex; introduces tanha as mental motion.
• Bayesian Brain Hypothesis

  extends
  Normative theory proposing biological systems perform approximate Bayesian inference through free energy minimization.
• Emptiness Realisation Model

  uses
  The paper's primary contribution: formalising Buddhist awakening as BMR of the separation prior sigma
• Free Energy Approach To Pattern Regulation

  extends
  Application of free-energy principle to understand pattern regulation in biological systems (Friston et al. 2015).
• Markov Decision Process (MDP)

  implements
  Generative model substrate for active inference; discrete states, actions, outcomes, and temporal policies.
• Morphogenesis As Bayesian Inference

  associated_with
  Variational approach to pattern formation and control in biology (Kuchling et al. 2020).
• Predictive Coding

  associated_with
  Related framework emphasizing prediction errors; active inference extends to Markov decision processes.
• Structural Representationalism

  contradicts
  Dominant interpretation of generative models as neural structures with representational content; main target of critique
• Bayesian Cognitive Science

  contradicts
  Umbrella framework for brain-centric Bayesian approaches; contrasted with active inference
• Bellman Optimality Principle

  contradicts
  Classical optimal control principle argued to be inapplicable to belief-based epistemic problems
• Good Regulator Theorem

  associated_with
  Theorem stating every good regulator of a system must be a model of that system.

• Transfer of Dopamine Responses

  supports
  Learning phenomenon reproduced by active inference: dopamine discharge shifts from unconditioned to conditioned stimuli.
• Mismatch Negativity

  associated_with
  ERP component reproduced by active inference: neural response to prediction violations.
• Theta-Gamma Coupling

  associated_with
  Hippocampal oscillatory phenomenon reproduced by active inference; phase-amplitude coupling.

• Modified OpenAI Gym FrozenLake environment

  about
  A 3×3 grid world with start, frozen, hole, and goal states used for comparing active inference and RL agents.
• SPM Academic Software (spm_MDP_VB_X.m)

  implements
  Matlab code implementing active inference belief updates; available at fil.ion.ucl.ac.uk/spm/
• spm_MDP_VB_X.m

  implements
  MATLAB routine in SPM software implementing discrete state-space active inference belief updating.

• FrozenLake OpenAI Gym Environment

  cites
  Modified discrete state-space environment used for experimental comparison of active inference and RL agents.
• Radial Maze Simulation

  about
  Benchmark task used to illustrate inference, learning, and foraging phenomena in active inference simulations.

• Tanha as Unskillful Active Inference (TUAI): tanha is a side-effect of active inference gone wrong through rate overload, uncontrollable domains, context desynchronization, and metabolic dysfunction.

  about
  Core hypothesis linking tanha to active inference failures.
• Vascular Clamp Hypothesis (VCH): Vascular contractions freeze local neural patterns and plasticity for the duration of contraction, with specific constrictions encoding specific predictions, functioning as medium-term memory.

  about
  Second core hypothesis, linking VSMC contraction to active inference predictions and memory.