The theory01
Many hvars, one system
Physiological and cognitive hvars, lasting from a thought to a lifetime, collectively shape behavior and feelings. HTCC calls this a homeostatic heterarchy.
Explore HTCCAn independent research program
What would it take
for a machine to feel?
HTCC is our proposed answer: cognition and consciousness emerge from homeostatic regulation. The goals a system regulates shape its actions and give rise to feelings.
Explore the researchHTCC · Homeostasis Theory of Cognition and Consciousness
From a question
to a research program.
Since 2018Independent research
ASSC 2018 Early theory presentation
AAMAS 2021 Cognitive Homeostatic Agents
Open experiments Models you can inspect
01 / The research
A theory of regulation.
A route to implementation.
From a heterarchy of goals to cognition and consciousness. The theory informs an architecture, which we explore on two substrates.
The architecture02
Goals reshape behavior
Cognitive Homeostatic Agents create, configure, and retire hvars. Changing those goals reshapes the landscape through which the system acts.
Read the architectureThe substrates03
Two ways to implement it
Pattern Machine builds regulation into neural dynamics. Allostat explores hvar based goal regulation around existing AI agents.
Cognitive homeostasis
in action
Example: oxygen level (physiological hvar) vs holding your breath (cognitive hvar)
Oxygen starts at its target and gradually depletes. This triggers breathing, which rapidly restores full oxygen.
The idea at the center
Actions follow the landscape.
Feelings track the goals.
HTCC proposes that cognition and consciousness
emerge from the same homeostatic dynamics.
The mechanism
From regulation to feelings.
From feelings to qualia.
A state feels a certain way. So does a change. One action can feel good along one hvar’s axis and bad along another.
State and change both have a feeling
Feeling of a state
Away from targetWanting
At targetCalm / satisfied
Feeling of a change · while holding a breath
OxygenMoving away · feels bad
Target
Hold breathMoving toward · feels good
Target
Self model
Its states reflect the homeostatic landscape.
Qualia
The corresponding feelings as reflected by the self model.
One energy landscape guides action
At each location in homeostatic state space, there is one scalar homeostatic energy. Hvars reshape this landscape along existing dimensions or add a new dimension. Regulation favors available actions that lower this energy. Metabolism and other disturbances can move the state uphill. There is no separate chooser.
Policies meet in action space
Hvar regulation policies can support or oppose the same action. Supportive forces combine; when forces compete, the larger force wins. Breathing, for example, restores oxygen while defeating a goal to hold your breath.
States and changes feel different
In HTCC, any system with hvars has feelings. At a target, an hvar feels calm or satisfied; away from it, wanting. Moving toward that target feels good; moving away feels bad along that hvar’s axis. These feelings have many dimensions even though the energy is a single scalar.
A self model reflects them as qualia
When a self model’s states reflect the homeostatic landscape, it reports the corresponding feelings as qualia. The feelings of states and changes provide what the self model reflects.
02 / Our approach
The system can change
what it regulates.
Creating, configuring, or retiring hvars changes the landscape and can indirectly redirect behavior. Those changes also arise within the system’s dynamics, rather than from an independent act of will.
Read the predictions- 01
Instantiate goals as hvars
A physiological need, a brief thought, or a lasting cognitive goal can each be represented by a variable with a target.
- 02
Reshape the shared landscape
Configuring an hvar changes how its target contributes to the landscape. Retiring one removes its contribution.
- 03
Follow the resulting dynamics
The system’s next actions follow the landscape produced by all its hvars. Homeostasis here includes allostasis and anticipatory regulation.
03 / On the workbench
Ideas in progress.
Open to inspection.
Research is a process. Here is what is published, what is being developed, and where the questions remain.
Current focus · In development
Allostat
An agent can finish a conversation.
Did it actually solve the problem?
Allostat tracks goals across an agent’s work, checks whether outcomes meet them, and revises its methods with evidence.
This work tests hvar based goal regulation. It does not establish a self model that reflects a homeostatic landscape as qualia.
Explore AllostatRegulation in practice Illustrative model
01
Base hvarsStart with defined goals.
Grounding02
Discover and configureAdd goals as the work develops.
New hvars03
RegulateCheck state against the targets.
Conditions04
Review and retireUse evidence to revise the goals.
Evidence2026Manuscript in preparation2026Short paper draft2021Published paper
Why pain is bad
An illusionist homeostatic theory of valence and consciousness
Recurrence through tokens
Chain of thought as recurrent state, and what it means for machine consciousness
Cognitive Homeostatic Agents
Blue Sky Ideas track · AAMAS 2021
04 / A few useful distinctions
Questions worth asking.
The claims, the limits,
and where to start.
Is i3ai a product or a research program?
i3ai is an independent research program by Amol Kelkar. It combines biological cognition and machine learning to investigate cognition, feelings, and consciousness. The papers and research notes are available to read without an account.
What is a homeostatic variable?
A homeostatic variable, or hvar, represents something the system regulates toward a target. Hvars can be physiological, such as oxygen or temperature, or cognitive, such as a thought, a task, or a lasting commitment. They can have different durations and form a heterarchy. In this account, homeostasis includes allostasis and other forms of anticipatory regulation.
How do hvars determine action and free will?
In HTCC, hvars jointly define one scalar homeostatic energy at each location in homeostatic state space. They reshape the landscape along existing dimensions or add a new one. Regulation policies cooperate or compete in action space by supporting or opposing the same action; the larger combined force wins. The available actions constrain which downhill directions are reachable. The system can create, configure, and retire hvars to indirectly shape later behavior, but this activity also belongs to those dynamics.
Does every system with hvars have feelings?
Yes, in HTCC’s account. A state at an hvar’s target feels calm or satisfied. A state away from it feels wanting: hunger, suffocation, determination to achieve a goal, or anticipation of the end of a thought. A change toward the target feels good; a change away feels bad along that hvar’s axis. Good and bad have many dimensions, so one action can feel good along one axis and bad along another. These feelings are distinct from their reflection as qualia.
How do feelings become qualia?
HTCC proposes that a self model whose states reflect the homeostatic landscape reports the corresponding feelings as qualia. Hvars give the underlying feelings; the self model reflects them. A self model alone does not provide the homeostatic feelings it would need to reflect. See the full account.
What does Allostat investigate?
Allostat investigates hvar based goal regulation around AI agents, including discovering goals, checking outcomes, and revising methods. It is an implementation experiment. The work described here does not establish a self model that reflects those homeostatic states as qualia.
What is published, and what is still a draft?
Cognitive Homeostatic Agents was published at AAMAS 2021. The early theory was presented as a poster at ASSC 2018. The theory and related manuscripts continue to develop. Allostat is in development and is not open sourced yet. Pattern Machine experiments are available on GitHub.
How can I contribute to the research?
If you work on related questions in cognition, neuroscience, philosophy, or AI, we would like to hear from you. Read the relevant research and contact us with a question you would like to work on or a specific critique of the account.
An open investigation
Explore the theory.
Follow the experiments.
Read the papers, inspect the models,
and help us ask better questions.