community
active
leiden_hybrid_concepts
label: sonnet
community:leiden_hybrid_concepts-run4-c8Active inference & agent ecology
Free energy minimization, Markov blankets, trust gradients, and multi-agent rhythm/deferral frameworks
70 members. Each node is clickable.
Loading graph…
Sub-communities (12)
Finer clusters this community splits into. Each is its own community page.
Agent ecology design principles7Active inference and free energy minimization7Empirical operationalization of agency attribution7Substrate-neutral sentience criteria7Functional criteria for moral patienthood7Substrate-agnostic systems theory6Speech acts in artificial agents5Hybrid agentic-latent reasoning architectures5Free energy minimization in active inference4Active inference as Bayes-optimal exploration4Agent design through restraint and simplicity3Confabulation as adaptive neural coherence3
Drawn from 28 sources
The papers/notes whose extracted claims & findings make up this cluster.
- agent-harness-design.md9 members
- Generalizing frameworks for sentience beyond natural species6 members
- Active inference: demystified and compared5 members
- guo-atlas-2026.md4 members
- Self-Improvising Memory: A Perspective on Memories as Agential, Dynamically Reinterpreting Cognitive Glue4 members
- A Free energy principle for the brain (lecture summary)4 members
- Multiple ways to implement and infer sentience4 members
- Elephant 2000: A Programming Language Based on Speech Acts4 members
- Technological Approach to Mind Everywhere: An Experimentally-Grounded Framework for Understanding Diverse Bodies and Minds3 members
- Active inference on discrete state-spaces: a synthesis3 members
- Active Inference: A Process Theory3 members
- Technological Approach to Mind Everywhere: An Experimentally-Grounded Framework for Understanding Diverse Bodies and Minds3 members
- Multiple ways to implement and infer Consciousness3 members
- AI: A Bridge Toward Diverse Intelligence.md2 members
- 2026-05-14_phil-trans-A-goodfire-aboutblank-impact.md2 members
- SFR-DeepResearch: Towards Effective Reinforcement Learning for Autonomously Reasoning Single Agents2 members
- Endless forms most beautiful 2.0: teleonomy and the bioengineering of chimaeric and synthetic organisms1 member
- Mechanistic Knobs in LLMs: Retrieving and Steering High-Order Semantic Features via Sparse Autoencoders1 member
- Self-Improvising Memory: A Perspective on Memories as Agential, Dynamically Reinterpreting Cognitive Glue1 member
- Project Oberon: the design of an operating system and compiler1 member
- 2026 02 02_2246_Search_Papers_The Found Papers Model Wealth Distribution Dynamic1 member
- Consciousness in Artificial Intelligence: Insights from the Science of Consciousness1 member
- Towards a computational phenomenology of mental action: modelling meta-awareness and attentional control with deep parametric active inference1 member
- Taking AI Welfare Seriously1 member
- Generalizing frameworks for sentience beyond natural species1 member
- Toward an ethics of autopoietic technology: Dukkha, care, and Intelligence1 member
- koan-battery-section.md1 member
- 15-properties-of-aliveness-in-AI.md1 member
Bridges (20)
Other communities that share members with this one — cross-cutting threads or papers that sit at the seam between two themes.
- Substrate-agnostic behavioral inference of cognition16 shared
- Active inference & free energy minimization11 shared
- Active inference and free energy minimization7 shared
- Empirical operationalization of agency attribution7 shared
- Functional criteria for moral patienthood7 shared
- Agent ecology aliveness & trust7 shared
- Substrate-neutral sentience criteria7 shared
- Agent ecology design principles7 shared
- Substrate-agnostic systems theory6 shared
- Speech acts in artificial agents5 shared
- Hybrid agentic-latent reasoning architectures5 shared
- Free energy minimization in active inference4 shared
- Alexander's 15 Properties for AI Interface Aliveness4 shared
- Active inference as Bayes-optimal exploration4 shared
- Functional token unified reasoning3 shared
- Confabulation as adaptive neural coherence3 shared
- Natural Language Auditing of Neural Models3 shared
- Agent design through restraint and simplicity3 shared
- Adaptive confabulation framework2 shared
- Ontology-ethics stress-care link1 shared
Claims (64)
- A system-agnostic approach offers the best roadmap for a deep, empirically fruitful unification of frameworks across cybernetics, behavioural science, bioengineering, robotics, materials science, and biomedicine.Prescriptive claim that ignoring substrate will unify diverse fields.
- Double standard in sentience attribution across substratesRouleau & Levin argue: behavioral patterns attributed to sentience in animals should be afforded same interpretation in plants and other systems without neural substrate bias.
- A system-agnostic approach evaluating observable response patterns without reference to substrate offers the best roadmap for empirically fruitful unification.Core argumentative position: sentience assessment should focus on behavior, not substrate composition; extends to AI and robotic systems.
- Acting to maximize value is the same as acting to minimize surprise; value is simply the probability of sensory input expected by an agent.Reinterprets classical reward/value concepts through free energy lens.
- Acting to optimize value and perception are two aspects of exactly the same principle: minimization of free energy.Foundational claim unifying action and perception within single optimization framework.
- Active inference agents can learn their own reward function (prior preferences) by interacting with the environment, bypassing the need for an explicit reward signal.Abstract and §3, preference learning section.
- Active inference describes the dynamics of systems that persist at non-equilibrium steady-state and that can be statistically segregated from their environment via a Markov blanket.Sets the theoretical grounding in Section 2.
- Agentic methods incur context-switching latency from external execution.Identifies a key limitation of agentic methods.
- 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.Formalization of perception-action cycle integrating inference and decision-making.
- Any system monitoring and controlling its own processing requires all three hierarchical levels: perception, attention, meta-awareness.Design principle with implications for AI and consciousness-UX; architectural requirement for self-directed cognition.
- ATLAS combines the strengths of agentic and latent reasoning by using a single discrete word (functional token) that serves both roles.Core claim of the paper: a unified token bridges the gaps.
- Attribution of agency is an empirical question, not a philosophical one.
- Behavior prescribed by active inference dynamics is approximately Bayes-optimal.Process theory outcomes produce normatively sound decision-making.
- Biology commits to optimizing salience and meaning, not fidelity, via high-level, agent-based interpretation.Central claim that biological memory prioritizes gestalt over details.
- Brain homology to humans as sentience criterionTraditional criterion inadequate across diverse embodiments; arbitrary constraint frozen in human phylogeny.
- Concentrate on essential functions and omit embellishments.
- Confabulation as adaptive cognitive plasticity
- Confabulation as adaptive sense-makingLevin reframes confabulation—often pejorative—as a fundamental biological strategy for extracting actionable wisdom from incomplete or contradictory information.
- Contingent properties like composition, origin, or similarity to human brains should not define ethical consideration for novel agents.Ethics must be based on empirically-determined cognitive properties (goals, preferences, concerns) rather than parochial markers.
- Current double standard applies behavioral criteria for sentience attribution to animals but not to plants or other non-neural systems.
- Developing principled sentience frameworks is an existential requirement for humanity as it encounters diverse intelligences.
- Dopamine discharge encodes changes in expected free energy under posterior vs. prior policy beliefs, representing precision updates.Links dopamine to precision modulation; reward prediction error reflects expected free energy changes.
- Elephant 2000 programs can perform some kinds of speech acts as genuinely as humans, including genuine promises.McCarthy's position against Searle: computers can genuinely promise without requiring Searle's conditions (e.g., belief that fulfillment benefits recipient).
- Epistemic behavior (exploration) emerges from maximizing mutual information between hidden states and observations.Formal mechanism for curiosity and information-seeking behavior derived from expected free energy.
- Ethics of novel beings should be based on teleonomic capacity (scale of goals they can pursue).Proposal for a new ethical framework.
- Exchanges with the environment are maintained within bounds that preserve the integrity of the agent through surprise minimization.Links free energy minimization to homeostatic preservation of biological systems.
- Expected free energy decomposes into risk (exploitation) and ambiguity (exploration) terms, providing optimal balance between goal-seeking and novelty-seeking.Key insight into structure of decision-making; explains intrinsic motivation and curiosity.
- Intelligence ratchet through substrate unreliability and competency
- Intrinsic links between ontology and ethics are evident in how all intelligent agents respond to perceived mismatch between what is and what should be.Bridges descriptive model (SCI loops) to normative framework (ethics).
- Many speech act properties are independent of whether the agent is human or machine.
- +34 more
Findings (6)
- Active inference agents engage in information-seeking behavior in reward-free FrozenLake environments, contrasting with Q-learning but similar to Bayesian RL.Empirical demonstration on FrozenLake; shows epistemic value drives exploration absent reward signal.
- All three agent types (active inference, Q-learning, Bayesian RL) perform adequately in stationary FrozenLake; only active inference achieves Bayes-optimal behavior in non-stationary settings.Key empirical result validating online planning capability of active inference.
- Commissurotomy (Split-Brain) ConfabulationSplit-brain patients whose left hand performs unexpected movements generate narrative explanations to preserve self-model; demonstrates active memory reinterpretation.
- No papers directly compare cultures with vs without proportional wealth-sharing norms—all focus on modeling wealth distribution within single economic systems rather than cross-cultural comparative analysis of sharing institutionsCore empirical finding of the search: identifies the absence of cross-cultural comparative work on wealth-sharing institutions and their economic/social outcomes.
- Our method achieves superior performance compared to Contrastive Activation Addition.Performance gains over CAA in steering tasks.
- Single-turn agentic workflow yields 10% absolute improvement on FRAMES for QwQ-32B over default multi-turn template.Result demonstrating inference-time architectural gains from reformulating multi-turn interactions as single-turn contextual QA.