1. Clinical hypotheses guide the question
Ad-Infer uses the current hypotheses, causal relationships and belief state to determine what information is needed and why.
Ad-Infer · Current architecture
Ad-Infer: an evolving
system of clinical inference
Ad-Infer integrates clinical knowledge, causal reasoning, fuzzy evaluation, learning and an LLM directed by the inference engine within one continuous reasoning loop.
Current architecture
Knowledge, inference and language within Ad-Infer
Ad-Infer provides the system-level architecture. EVAL-KBS combines domain knowledge with Type-2 non-monotonic causal reasoning, Fuzzy Evaluation Learning, neural and deep-learning mechanisms, and Bayesian belief updating. The LLM is an internal functional component of this integrated system.
Ad-InferOne integrated inferential system
Clinical representation
Domain knowledge
Concepts, questions, diagnostic criteria, exclusions, clinical context and temporal relationships.
Sleep-EVAL provides the sleep-disorders knowledge base.
→
Inferential controlEVAL-KBS
EVAL-KBS
Inference & learning
Type-2 causal reasoning, fuzzy evaluation, neural and deep-learning processes, and Bayesian belief updating.
Directs the interview, evaluates evidence and retains diagnostic authority.
↔
Internal language functions
Integrated LLM
Standardized questioning, reformulation, clarification, translation and explanations of why a question is asked.
Interprets new information in relation to direct and indirect clinical causality.
Continuous return to inference. Questions asked or skipped, understood or misunderstood, accepted or refused, and answers that support or challenge a hypothesis return to Ad-Infer. EVAL-KBS re-evaluates their clinical meaning, adjusts the belief state and directs the next step.
A bidirectional clinical loop
The inference engine directs the interaction
2. Language functions support understanding
The integrated LLM supports standardized wording, reformulation, translation, clarification and WHY explanations, and interprets how new information relates to the clinical context.
3. Returned information changes inference
The content of the answers and the course of the interaction return to EVAL-KBS for clinical re-evaluation. Ad-Infer then directs the next question or revises its conclusions.
Evidence is re-evaluated
Both concordant and challenging evidence matter
Strengthening a positive diagnosis
Concordant responses can support the current hypothesis when required criteria, context and exclusions are evaluated.
Reopening the differential
Contradictions, qualifiers and alternative explanations can change the belief state and prompt further exploration.
Learning from the interview itself
Understanding, misunderstanding, skipped questions and refusals inform how Ad-Infer interprets the available evidence and what needs clarification.
The inferential engine retains authority over diagnostic reasoning. Language-model output is returned to the structured causal, fuzzy and belief-updating processes before it contributes to an inferential conclusion.
Temporal & longitudinal reasoning
Direct and indirect causality over time
Clinical meaning depends on context, order and timing: which symptom appeared first, whether treatment preceded a change, whether a feature is persistent or episodic, and whether later information changes the interpretation of an earlier observation.
Ad-Infer can follow competing hypotheses, evaluate exclusions and comorbidities, and reopen a prior inference when new information changes the causal picture. Repeated assessments support the study of incidence, persistence, remission, recurrence and functional trajectories.
From foundations to current capabilities
An architecture in continuous development
Hopfield neural computation and Zadeh fuzzy logic are part of Ad-Infer’s computational foundations. The platform has continued to evolve in its knowledge representation, causal reasoning, learning processes and interview functions. Integrated LLM capabilities extend these functions within the inferential architecture.
- 1987
Development begins
Maurice M. Ohayon begins developing Ad-Infer for clinical interviewing and inferential reasoning.
- 1992
Computational foundations
Doctoral work formalizes the architecture, including Hopfield neural computation and Zadeh fuzzy logic.
- 1990s onward
Population & longitudinal research
Domain knowledge bases support structured assessment in population samples, clinical groups and repeated interviews.
- Since 2016
Web-based assessment
Research applications extend the structured interview to web-based assessments.
- Current architecture
Integrated language-model capabilities
The evolving Ad-Infer system incorporates an LLM as an internal functional component directed by causal inference, fuzzy evaluation and the current belief state.
Research evidence
Validation and the scientific record
Foundational studies assessed earlier Sleep-EVAL implementations against specialist clinical assessment and polysomnographic data. The publication record also documents population-based and longitudinal applications. These historical studies describe the configurations evaluated at the time; current language-model integration requires evaluation appropriate to its intended application.