Somatic AI & Research

Somatic AI vs. Embodied AI, Affective Computing and Bioadaptive Interfaces

Five overlapping fields describe machines that deal with bodies. They are not interchangeable — and the differences reveal what a system senses, whose state it models, and whose interests its adaptation serves.

By Hannes BendAugust 1, 20269 min read

Concept visualization: Somatic AI uses consented physiological, behavioral and experiential signals as context for adaptive interfaces. Bioadaptive systems can adjust presentation or environments in response, while affective computing and embodied AI address related but distinct questions.

Artificial intelligence is increasingly described through the language of bodies: embodied AI, affective computing, physiological computing, bioadaptive interfaces and, more recently, Somatic AI.

These terms overlap, but they are not interchangeable.

Embodied AI usually asks how an intelligent agent can perceive and act within a physical or simulated environment. Affective computing focuses on recognizing, interpreting or expressing emotion. Physiological computing uses biological signals as inputs to computer systems. A bioadaptive interface closes the loop by changing an interface or environment in response to those signals.

I use Somatic AI to describe a related but distinct direction: artificial intelligence that senses aspects of a human being's embodied state and uses that information to adapt an experience, interface or environment in ways intended to support the person's agency, attention, performance or wellbeing.

This article explains where these fields differ, where they overlap and how they connect to my work across art, biofeedback, physiological sensing and adaptive interfaces.

A concise comparison

Field	Primary subject	Typical input	Typical output or action	Central question
Somatic AI	The embodied state of a person	Heart rate, breathing, posture, movement, expression, context or self-report	Personalized content, feedback, interface or environment	How can AI respond to how a person is, not only to what the person clicks or says?
Bioadaptive interface	The closed interaction loop between a user and a system	Physiological or behavioral signals	Real-time interface adaptation	How should a system change in response to measured human state?
Physiological computing	Biological signals used within computing	ECG, PPG, EEG, EDA, respiration, eye activity and related signals	Measurement, classification, control or adaptation	How can physiology become an input to a computer system?
Affective computing	Emotion and affect	Face, voice, text, physiology and behavior	Emotion recognition, response or expression	How can computers recognize, understand or communicate affect?
Embodied AI	An intelligent agent acting through a body or within an environment	Vision, touch, proprioception, language and environmental sensors	Navigation, manipulation or physical action	How can intelligence emerge through perception and action in a world?

Somatic AI

Primary subject:: The embodied state of a person
Typical input:: Heart rate, breathing, posture, movement, expression, context or self-report
Output or action:: Personalized content, feedback, interface or environment
Central question:: How can AI respond to how a person is, not only to what the person clicks or says?

Bioadaptive interface

Primary subject:: The closed interaction loop between a user and a system
Typical input:: Physiological or behavioral signals
Output or action:: Real-time interface adaptation
Central question:: How should a system change in response to measured human state?

Physiological computing

Primary subject:: Biological signals used within computing
Typical input:: ECG, PPG, EEG, EDA, respiration, eye activity and related signals
Output or action:: Measurement, classification, control or adaptation
Central question:: How can physiology become an input to a computer system?

Affective computing

Primary subject:: Emotion and affect
Typical input:: Face, voice, text, physiology and behavior
Output or action:: Emotion recognition, response or expression
Central question:: How can computers recognize, understand or communicate affect?

Embodied AI

Primary subject:: An intelligent agent acting through a body or within an environment
Typical input:: Vision, touch, proprioception, language and environmental sensors
Output or action:: Navigation, manipulation or physical action
Central question:: How can intelligence emerge through perception and action in a world?

The categories can intersect. A robot can be embodied AI and use affective computing. A meditation application can use physiological computing without becoming a fully bioadaptive system. A bioadaptive environment can use machine learning without being a robot.

The distinctions matter because they reveal what a system is sensing, whose state it is modeling, how it acts and whose interests its adaptation serves.

What is Somatic AI?

"Somatic" relates to the body as lived and experienced, rather than the body understood only as an external object.

The term Somatic AI does not yet have one universally accepted technical definition. It is currently used in several ways.

Some writers use it for medical AI that builds computational representations of anatomy and physiology [9]. Other companies use it for collecting physical-movement data to train robots [10]. In somatic psychology and interaction design, the emphasis may instead be on bodily awareness, interoception and lived experience.

My use of the term is centered on the relationship between human physiological state and adaptive technology.

In this framework, Somatic AI has four defining characteristics:

It treats the human body and embodied state as meaningful context.
It senses that context through consented physiological, behavioral or experiential data.
It adapts a digital or physical experience in response.
It is designed to increase human agency rather than merely optimize engagement.

The last point is essential.

An advertising platform could theoretically use physiological cues to hold attention more aggressively. That would be biologically responsive, but it would not represent the human-centered direction I am advocating.

Somatic AI should help people understand, regulate or express their own state. It should not turn the body into another invisible source of extractive behavioral data.

Read the core definition: What Is Somatic AI?

What is a bioadaptive interface?

A bioadaptive interface is an interface that changes in response to biological or physiological information from its user.

A conventional adaptive system may learn from clicks, purchasing behavior, past choices or navigation patterns. A bioadaptive system can additionally respond to signals associated with the user's present state.

For example, an interface might:

reduce visual intensity when signs of strain increase;
alter pacing when breathing becomes irregular;
offer a pause after sustained cognitive load;
adjust audiovisual elements in relation to heart rate;
change guidance based on posture or movement;
personalize feedback against an individual baseline.

The defining feature is not simply measurement. It is the closed loop:

sense → interpret → adapt → sense again

This is why bioadaptive interfaces are more specific than biometric analytics. A dashboard that displays a heart rate is providing information. A system that changes its behavior in response to the heart rate is participating in a bioadaptive loop.

Read the technical overview: Bioadaptive Interfaces: Definition, History and Applications

What is physiological computing?

Physiological computing is the broader research field in which real-time physiological measurements are used as inputs to computer systems [3].

Stephen Fairclough described physiological computing systems as systems that communicate information about a user's psychological state to an adaptive system through real-time psychophysiological measurement [2].

The input can include:

heart activity;
respiration;
brain activity;
electrodermal activity;
muscle activity;
eye behavior;
temperature;
movement;
combinations of several signals.

Not every physiological-computing system is AI, and not every one is adaptive.

A research instrument can acquire and visualize physiological signals without making autonomous decisions. A biofeedback artwork can translate breath into light without using a trained machine-learning model. A Somatic AI system can build on both approaches while adding inference, personalization and adaptive decision-making.

Physiological computing therefore provides much of the technical and conceptual foundation on which bioadaptive interfaces and Somatic AI can operate.

What is affective computing?

Affective computing is the field associated especially with Rosalind Picard's foundational work at MIT [1].

It investigates how computers can recognize, interpret, process, model or express affect and emotion.

Affective systems may use:

facial expression;
voice;
language;
physiological signals;
movement;
interaction patterns;
contextual information.

Affective computing and Somatic AI overlap when bodily signals are used to understand emotion or when an AI system responds to an affective state.

But the two are not identical.

Somatic AI need not infer a named emotion. A system could respond to breathing rhythm, physical stillness, fatigue, arousal or postural change without classifying the user as "happy," "sad" or "angry."

This distinction can be ethically important. Human states are contextual and ambiguous. Designing around observable changes and user-controlled feedback can sometimes be more respectful than making confident hidden claims about a person's emotions.

What is embodied AI?

Embodied AI generally concerns intelligent agents that perceive, learn and act within physical or simulated environments [4].

Robots, autonomous vehicles, drones and agents operating in interactive 3D environments are common examples.

The "body" in embodied AI usually belongs to the artificial agent.

The "body" in the form of Somatic AI described here belongs primarily to the human participant.

An embodied robot asks:

How can this agent perceive the world and act within it?

A human-centered Somatic AI system asks:

How can this system understand relevant aspects of a person's embodied state and adapt responsibly around that person?

The fields can converge. A socially assistive robot could be both embodied and somatically responsive. It could navigate a room through its own sensors while also adapting its behavior to a consenting user's breathing, posture or stress-related signals.

But the design priorities remain different. Embodied AI often emphasizes an agent's successful action. Somatic AI should emphasize the human participant's experience, agency and wellbeing.

Where biofeedback fits

Biofeedback predates contemporary machine learning.

A biofeedback system measures a physiological process and returns information about that process to the participant. The feedback may be visual, auditory, tactile, spatial or immersive.

When I began connecting breath, heart rate, immersive environments and artistic experiences, the central question was not how to automate a person. It was how technology could make internal processes perceptible.

That work progressed through art, neuroscience collaborations, VR, conference presentations, patents and later camera-based physiological sensing.

The history matters because today's AI terminology can make older work appear to have begun only when the current label became popular.

The underlying research questions existed earlier:

Can an environment respond to breathing?
Can physiological change alter an image or virtual world?
Can an interface support attention rather than fragment it?
Can personalization be based on present state rather than historical behavior?
Can biological data remain under the user's control?

See the documented chronology: Progression: From Art and Biofeedback to Bioadaptive AI

Read the longer historical article: From Immersive Art to Remote Photoplethysmography: The Evolution of Bioadaptive AI

How the fields converge in my work

My work belongs most directly to the traditions of biofeedback, physiological computing and bioadaptive interfaces.

The artistic practice explored perception, attention, material transformation and the body as an active participant in an artwork. Later immersive works used breath and heart-rate feedback to make internal state part of the experience.

The 2016 AAAI Spring Symposium paper "Mindful Technologies: Research and Developments in Science and Art" documented interdisciplinary projects combining biofeedback, virtual reality and human-machine interaction [5, 6].

Patent records subsequently formalized methods for adapting screen-based output using information about a user [7].

Later work extended these questions into camera-based physiological sensing, digital-wellbeing systems and commercial interfaces [8].

I now use Somatic AI as an umbrella for this trajectory: AI and adaptive technology designed around embodied human state.

This does not mean that one person or project created all of physiological computing, affective computing or embodied interaction. Those fields have deep and diverse histories.

The contribution is the continuity between artistic investigation, physiological feedback, adaptive-interface invention and contemporary human-state-aware AI.

That dated continuity — not the repetition of a label — is the basis for describing the work as pioneering.

Explore the evidence: Research, Papers and Patents · Art and Biofeedback Works · Somatic AI and Bioadaptive Ventures

The ethical question: adaptation for whom?

A system that senses the body can support care, but it can also intensify surveillance.

The same classes of signals that help a user recognize overload could be used by an employer to evaluate workers, by an advertiser to optimize persuasion or by a platform to infer sensitive states without meaningful consent.

Responsible Somatic AI therefore requires more than technical accuracy.

It should include:

explicit and informed consent;
understandable explanations of what is being sensed;
user control over activation;
minimal collection;
limited retention;
local or on-device processing where practical;
uncertainty rather than overconfident emotional labels;
clear separation between wellness and medical claims;
meaningful opt-out mechanisms;
benefits directed toward the participant.

The body should not become an invisible interface for other people's objectives.

A working definition

I define human-centered Somatic AI as:

Artificial intelligence that senses relevant aspects of embodied human state and responsibly adapts an experience, interface or environment to support human agency, awareness, performance or wellbeing.

This is a working definition rather than an attempt to erase other uses of the term.

Its purpose is to identify a design direction: systems that understand people not only through language, clicks and historical behavior, but through consented, contextual and embodied information.

The most important question is not whether AI can detect more about the body.

It is whether that knowledge helps people become more aware, capable and free.

Frequently asked questions

Is Somatic AI the same as embodied AI?

No. Embodied AI generally refers to an artificial agent that perceives and acts through a physical or simulated body. Somatic AI, as defined here, focuses on AI responding to the embodied state of a human participant. A system can be both.

Is a smartwatch a Somatic AI system?

Not automatically. A smartwatch that records heart rate is a sensing device. It becomes closer to a Somatic AI system when intelligent interpretation and responsible adaptation are added—for example, when the system changes guidance or an experience based on the user's present state.

Is every bioadaptive interface based on AI?

No. A bioadaptive system can use direct mappings or deterministic rules. Machine learning becomes relevant when the system must estimate state, personalize a baseline, combine signals or make more complex adaptive decisions.

How is affective computing different?

Affective computing is centered on emotion and affect. Somatic AI may include emotion, but it can also respond to breathing, posture, physical activation, fatigue, attention or other embodied conditions without assigning a named emotional category.

Is Somatic AI medical technology?

Not necessarily. It may be used in art, wellness, accessibility, education, human-computer interaction or research. A system making diagnostic, therapeutic or clinical claims requires appropriate validation, professional oversight and regulatory consideration.

What data can a Somatic AI system use?

Depending on the application and consent model, it may use heart activity, breathing, posture, movement, facial behavior, voice, interaction patterns, environmental context or self-reported experience. Responsible systems should collect only what is necessary.

References and further reading

1.Rosalind W. Picard, "Affective Computing," MIT Press: https://mitpress.mit.edu/9780262161701/affective-computing/
2.Stephen H. Fairclough, "Fundamentals of Physiological Computing": https://dl.acm.org/doi/10.1016/j.intcom.2008.10.011
3.Giulio Jacucci et al., "Physiological Computing": https://dl.acm.org/doi/abs/10.1109/MC.2015.291
4.Jiafei Duan et al., "A Survey of Embodied AI: From Simulators to Research Tasks": https://arxiv.org/abs/2103.04918
5.Hannes Bend et al., "Mindful Technologies: Research and Developments in Science and Art," AAAI Spring Symposium 2016: https://aaai.org/proceeding/07-spring-2016/
6.Hannes Bend, DBLP publication record: https://dblp.org/rec/conf/aaaiss/BendSKMASJ16
7."Adaptive Interface for Screen-Based Interactions," patent record: https://patents.google.com/patent/US20190042981A1/en
8."Adaptive Interfaces for Personalized Digital Wellbeing": https://ceur-ws.org/Vol-2903/IUI21WS-HEALTHI-1.pdf
9.An alternative medical use of the term Somatic AI (DCVC): https://www.dcvc.com/news-insights/the-somatic-ai-revolution-how-techmed-is-reshaping-healthcare-as-we-know-it
10.An alternative physical-workforce and robotics-data use: https://somatic-ai.com/