The AI would function as translation tissue rather than authority. Foundation models of physiology and disease would remain coupled to explicit causal hypotheses, provenance and calibrated uncertainty. Generative interfaces would learn a temporary sensory vocabulary with one patient and care team instead of imposing a universal representation. Active inference would identify observations that separate competing worlds; multimodal models would map metaphor and gesture into testable possibilities without declaring them literal; and adversarial agents would reveal when the representation erases experience, imports population bias or creates false coherence. The model would be required to expose how it sees, where it has no adequate analogue and which evidence could make it change. Its final act would be controlled decomposition, not indefinite retention.
The Third Body
“What if doctor and patient could temporarily share a body that belongs to neither of them?”
Grow a body between you.
Three perspectives remain distinct. The organism lives only while they learn to perceive together.
One perspective produces coherence. All three produce understanding.
The beginning
Medicine has spent centuries learning to look inside the patient, yet the fundamental gap remains: the patient inhabits illness without clinical interpretation, while the physician interprets illness without inhabiting it. The Third Body imagines a temporary synthetic sensory organism grown between them. It would translate physiology, lived experience, medical knowledge and uncertainty into a shared perceptual environment that neither person could create alone. It is not a digital twin, an AI doctor or a simulation of empathy. It is an intersubjective medical organ: a new sense through which doctor and patient can encounter illness together without surrendering their distinct perspectives.
Clinical language compresses pain, fatigue, cognitive fog and bodily wrongness into scales, codes and categories. Tests reveal fragments of biology but not what it is like to live inside their consequences. AI can integrate more signals, yet it still observes the distance between physician and patient from outside. The asymmetry is dangerous: the patient becomes increasingly transparent while the model and the clinician's uncertainty remain opaque. Medicine lacks an object capable of holding biological evidence, subjective experience, competing explanations and personal meaning simultaneously—without forcing them into premature agreement or permanent institutional memory.
What it could become
The Third Body would be born for one clinical relationship and die when its work is complete. A multimodal world model would translate live physiology, longitudinal records, movement, molecular data, narrative, gesture and patient-created metaphor into a negotiated sensory language: spatial, sonic, tactile and temporal. Doctor and patient could enter competing diagnostic worlds, feel where a model fits or fractures, construct impossible anatomies for distributed phenomena and rehearse uncertain treatment futures. The observer would also become observable: patients could perceive what the AI attends to, ignores and assumes, while clinicians could expose the evidence, doubt and judgment shaping a recommendation. Before a consequential decision, both would complete an epistemic handshake proving reciprocal understanding rather than one-sided consent.
For whom
- Patients whose experience exceeds existing medical language
- Physicians working with uncertainty, chronic illness and complex decisions
- Clinical teams caring for people with impaired communication
- Researchers in interoception, human-computer interaction and computational medicine
- Bioethicists and patient-rights organisations governing intimate representations
Core capabilities
- Multimodal physiological world models
- Generative spatial, sonic and haptic clinical interfaces
- Patient-authored phenomenological representations
- Competing causal worlds and counterfactual treatment rehearsal
- Visible model attention, assumptions and uncertainty
- Ephemeral computation with selective patient-owned memory
- Zero-knowledge verification of clinical process
- Reciprocal epistemic-consent protocols
The project becomes meaningful only when a new technical possibility is translated into a clear human advantage, an experience people can understand, and a system capable of earning trust over time.
Intelligence and mathematics
The Third Body can be formalised as a shared latent dynamical system constructed from three non-equivalent observation spaces: embodied patient experience, clinical interpretation and machine inference. Each has its own measurement operators, uncertainty and inaccessible variables. Rather than collapsing them, a category-theoretic translation layer preserves which relationships survive between representations and which do not. Bayesian world models maintain rival causal structures; active inference chooses reversible observations that maximise expected information gain; information geometry measures distortion as experience is translated into clinical form; and topological methods detect holes where no representation can explain the whole phenomenon. The goal is not maximum agreement, but a minimally lossy joint model whose disagreements remain legible.
How it might live
The first prototype should not attempt whole-body simulation. It could focus on one domain where subjective experience and measurable physiology repeatedly diverge, such as chronic pain, treatment-related cognitive change or medically unexplained symptoms. A research hospital could pair a patient-authored spatial interface with live signals and competing causal models, measuring whether it reveals decision-changing information that conventional consultation misses. The long-term form could be regulated clinical infrastructure licensed to health systems, with an open perceptual-sovereignty protocol and independent governance. Its defensibility would lie not in owning intimate data, but in verified translation methods, prospective evidence and a constitutional architecture that makes extraction technically difficult.
For me, a venture is more than an interesting technology. It needs a narrow first user, a repeated problem, a distribution path, a credible advantage and a reason to improve as more people use it. I would test those conditions before deciding whether this idea should become a company, a product, an open technology or an ongoing research programme.
Rules for making it real
- 01
Never claim to reproduce another person's consciousness or pain.
- 02
The patient retains perceptual sovereignty over every interpretation.
- 03
Make the observer, its assumptions and its uncertainty observable.
- 04
Preserve disagreement instead of manufacturing coherence.
- 05
No insurer, employer or advertiser may access the Third Body.
- 06
No credibility score for pain, emotion or subjective experience.
- 07
Information may be necessary for understanding without being necessary for storage.
- 08
The physician remains personally responsible for consequential decisions.
- 09
Participation must be optional and an unrecorded encounter must remain possible.
- 10
Mortality is a feature: the Third Body must know how to die.
From question to company
- 01Frame
Choose one clinical domain where lived experience and measurement routinely diverge.
- 02Prototype
Co-design a nonverbal sensory vocabulary with patients and clinicians.
- 03Prove
Build a bounded multimodal model that preserves three distinct perspectives.
- 04Build
Create competing diagnostic worlds with visible uncertainty and provenance.
- 05Advance
Test whether shared perception reveals clinically consequential information.
- 06Advance
Prototype reciprocal epistemic consent and patient-controlled decomposition.
- 07Advance
Run adversarial studies for suggestion, bias, surveillance and false embodiment.
- 08Advance
Publish an open perceptual-sovereignty charter before clinical expansion.
What could go wrong
Serious imagination includes the possibility that an idea should change radically—or should not exist. These are the tensions the project would need to resolve:
- Mistaking a generated representation for the patient's actual experience.
- Creating the most intimate surveillance system ever deployed in medicine.
- Allowing institutions to define legitimate pain or invisible illness.
- Model suggestion changing how patients perceive or narrate symptoms.
- False sensory precision making weak hypotheses feel physically true.
- Clinicians deferring responsibility to an apparently shared experience.
- Excluding people unable or unwilling to use immersive interfaces.
- Reconstructing a patient after they required their model to disappear.
- Commercialising intimate perceptual languages or impossible anatomies.
- Confusing technological intensity with better care.