Date

2007-06-04

About 1984, roughly 25 years later, at an annual meeting of the American Association for Artificial Intelligence (AAAI), a vote was taken and it was decided that probability was in no way relevant to Artiticial Intelligence. A protest group quickly formed and the next year there was a workshop at the AAAI meeting devoted to Probability and Uncertainty in A.I. This workshop has continued to the present day to be a yearly event.

Ray J. Solomonoff in THE UNIVERSAL DISTRIBUTION AND MACHINE LEARNING

Comments

What are the odds?!

Comment by q — June 4, 2007

well, that was before the era of Bayesian Networks when all probability driven methods were essentially ‘black box’ methods - so no explanations and hence not(AI).

Comment by Chintan — June 16, 2007

Chintan : what kind of “black box” method were them?

any pointer to a description/bibliographical paper? i would be interested in such hitorical facts

Comment by Pierre — June 24, 2007

The Uncertainty in AI workshop was spun out from AAAI by John Lemmer and Peter Cheeseman in 1985. In 1988 we incorporated into a non-profit professional society, Uncertainty in Artificial Intelligence, which continues to this day, and is usually held just before the AAAI conference, as it is this year in Vancouver.

Prior to the development of influence diagrams and its descendants, Bayesian networks and decision graphs, probability applications were “blackbox” in that they were systems of equations converted typically to a single piece of computer code.

Changes required going back to the original equations and re-deriving new systems of equations based on local changes to the structural or probabilistic parameters.

The development of influence diagrams and Bayesian Networks enabled local changes to be handled locally, visualized joint distributions, and enabled explanations of cause-effect events in terms of the topology of the networks.

Comment by Tod Levitt — June 29, 2007

Thank you Tod, that’s interesting.

Comment by Pierre — July 3, 2007