Human Function Intelligence

Beyond answering a question: reasoning about a person

If someone asks, “Why does my knee hurt when I climb stairs?”, generic question answering may list many causes. Human Function Intelligence first asks whose task this is, under what load and circumstances, what observations and history exist, and what information the authorized professional can use. It distinguishes task demand, available capacity, actual engagement, and performance or cost. Only when adequate evidence exists should explanations become more specific about anatomy, clinical conditions, intervention, or reassessment.

Task decomposition and bidirectional reasoning

One direction starts with the task and asks about required capacities: What functional conditions are needed and where might limitations in capacity or engagement arise? The reverse direction starts with an observed consequence and asks about competing explanations: What hypotheses could explain pain, fatigue, altered movement, or decline, and what assessment can differentiate them? These directions must be cross-checked against context, time, and evidence. An observed movement pattern cannot, by itself, establish a specific muscle, nerve, or disease diagnosis.

One shared function brain does not imply unlimited shared permissions

Movetips' architectural direction uses a common Human Function Brain Core for Free Chat, Learning Chat, Member Service Chat, and the human-function assessment entry point. They may share semantic and reasoning foundations, but they do not share the same authority to use personal evidence or create official records. Free Chat supports general knowledge and does not directly write real member-service facts. Learning Chat supports lessons, simulations, practice, and assessment. Member Service Chat requires an authorized member, service role, and real service context. Structured function assessment requires the relevant assessment content and admissible evidence.

The shared element is the intended reasoning discipline, not unrestricted access across contexts. End-to-end implementation and acceptance remain work in progress; the website must not claim that every planned inference path is already fully operational.

Evidence, memory, learning, and feedback

A reasoning result should distinguish recorded evidence, external knowledge, current inference, and candidate advice, retaining source information and versions. Long-term memory is not a process of converting every statement in a conversation into fact. It should preserve professionally authorized, confirmed, and auditable records along the relevant timeline. Execution, monitoring, feedback, and reassessment generate evidence for later decisions. Changes to the formal ontology require separate governance, not an automatic edit triggered by one conversation.

What a useful judgment should explain

For an individual function question, the system should aim to state what can presently be supported, which explanations remain hypothetical, what risks or professional boundaries apply, what information is missing, which next assessment is most informative, and how a conclusion should change if new evidence arrives. In nonclinical training it must not overstep into medical diagnosis; where a medical red flag is plausible, appropriate professional assessment and referral boundaries matter.

Research that can be evaluated

Beyond answer accuracy, Human Function Intelligence should be examined for missing-evidence recognition, traceability, consistency across cases, separation of capacity and engagement, time-dependent updating, and safe refusal to overclaim. Broader automatic learning, causal inference, longitudinal prediction, and coordination among domain models remain research directions; they must be evaluated rather than asserted.