Artificial Intelligence Expert Designs New Theory for Decision Making
How should humans make decisions when the outcomes of their choices are uncertainty and unpredictably determined by the theory of probability?
Such is the question posed by Prakash Shenoy, the Ronald G. Harper Distinguished Professor of Artificial Intelligence at the University of Kansas School of Business.
His resolution can be discovered in the brief article “An Interest-Valued Utility Theory for Decision making with Dempster-Shafer belief functions,” which appearing in the September issue of the International Journal of Approximate Reasoning.
Shenoy said that subjects tend to assume someone always can link likelihoods to unknown events.
Shenoy continues by telling that in realized life, one can never comprehend what probabilities really are. An actual individual does not know if they are 50 percent or 60 percent. This is the heart of the theory of belief roles that Arthur Dempster and Glenn Shafer established in the 1970s.
Their article (co-written with Thierry Denoeux) generalizes the decision-making concept from chance to belief function.
The probability theory of decision making is used for making any choice with a high degree of involvement. Like should someone accept a new job offer or a marriage proposal? Something high-stakes. Somebody wouldn’t demand it for wherever to go for lunch, Shenoy said.
However, in general, we never comprehend what is going to happen. A Person may take an assignment, but may end up with a bad driver. There is a lot of uncertainty. Someone may have two job offers, having to decide two choices of what to take. Then you do the pros and cons and plug the probability into them. The likabilities are great when you have a lot of repetition. But if it’s a one-time thing, you can’t “average your winnings.”
Shenoy maintained that one of the first answers to this issue was given by John von Neumann and Oskar Morgenstern in their 1947 book Theory of Games and Economic Behavior. In 1961, Daniel Ellsberg showed, using experimentation, that von Neumann and Morgenstern’s theory of decision was not descriptive of human actions, especially when ambiguity was represented by unpredictability by the likelihood theory.
During the late 1960s and mid-1970s, Arthur Dempster and Glenn Shafer (a previous KU professor in both mathematics and business) evolved a calculus of uncertainty called belief features that was a generalization of probability theory that was much better able to represent uncertainty. However, there was no decision theory to choose from when this theory described uncertainty.
Shenoy’s short article provides the first formalization of a theory for decision decision making when uncertainty is defined by Dempster-Shafer belief functions comparable to von Neumann-Morgenstern theory. Furthermore, Shenoy said that this concept is much more able to clarify Ellsberg’s speculative conclusions for choosing under ambiguity.
The professor first addressed Denoeux on that topic three years before, while both were addressing graduate school students.
(“Denoeux”) went through all of the theories of decision making with belief functions. After that, I went and told him, “All this stuff you said is unsatisfactory.” And he concurred with me! I said I would love to come and work with him on this. So he sent me an offer.”
Shenoy sought a sabbatical, then went to France in spring 2019, where he invested five months working with Denoeux at the Université de Technologie de Compiègne..
He said it was culturally very enriching, as well as suitably rewarding.
Now in his 43rd year at KU, Shenoy continues to be an expert in uncertain thoughts and their applications to an expert system. He is the inventor of Valuation Based Systems (VBS), a mathematical architecture for knowledge representation and inference that consisting of many uncertainty computations. His VBS architecture is currently used for multisensor combinations in ballistic missiles for the United States Department of Defense.
He hopes that his most current study can benefit those who rely on belief functions.
This includes many individuals in the military, for example, Shenoy stated. Shenoy added, they like belief function because of its flexibility, and they want to know exactly how decisions are made. If you’re going to minimize every little thing to possibilities in the end, why not make use of probabilities to begin with?”
Raed the original article on The University of Kansas.
