Atlas record: https://www.aitimeline.dev/en/events/evt-0147/
Research · April 1985
Bayesian networks
Pearl proposed representing dependencies between events as a graph, and showed how to propagate new evidence through it locally, without recomputing everything.
Why it matters
Uncertainty stopped being a heuristic add-on: there was now a way of computing probabilities that scales.
A full joint distribution grows exponentially with the number of variables, which is why early systems made do with unjustified coefficients. The graph encodes conditional independence and makes the computation feasible. This returned probability to AI after two decades of symbolic dominance and led to modern probabilistic models. Pearl received the Turing Award in 2011.
Event record
- Event date
- April 1985
- Timeline date
- Event date
- Verification
- Sources gathered automatically · September 17, 2026
- Lines
- ID
- evt-0147
Presented at the seventh Cognitive Science Society conference in Irvine, 15-17 April 1985; UCLA report CSD-850017.
Related events
Earlier
- Related MYCIN 1974
MYCIN's certainty factors are replaced by a calculation with a probabilistic justification.
- Extends Semantic networks 1968
A graph of concepts acquires probabilistic semantics and a rule for propagating evidence.
- Related Fuzzy Sets 1965
A different answer to uncertainty: not a degree of truth but probability with explicit dependencies.
Later