“Believing on the Basis of Evidence” outlines a program of research Kyburg has pursued over most of his career. The program addresses the difficult problem of how a rational agent can acquire beliefs about the world given the intrinsic uncertainty of empirical facts. This problem is also central to AI. Early A1 systems depended simply on deduction for generating beliefs, which can perhaps be viewed as “believing on the basis of proof.” Of course, deduction still requires a set of premises, and the question of how these are acquired returns one to the same problem. Furthermore, even given the premises, that is, a knowledge base, deduction is not sufficient to generate all of the conclusions that need to be drawn. This initiated research in nonmonotonic inference. Kyburg addresses the issue of how we can use our information to reason to a set of conclusions that are plausible but not certain. This set of conclusions goes beyond the set of deductive conclusions. His program is certainly ambitious and the approach he suggests meets with varying degrees of success and failure when it comes to particular issues. In the rest of the paper, we examine a few parts of Kyburg’s approach.
[1]
Henry E. Kyburg,,et al.
The Reference Class
,
1983,
Philosophy of Science.
[2]
Matthew L. Ginsberg,et al.
Readings in Nonmonotonic Reasoning
,
1987,
AAAI 1987.
[3]
Isaac Levi,et al.
The Enterprise Of Knowledge
,
1980
.
[4]
L. M. M.-T..
Theory of Probability
,
1929,
Nature.
[5]
Hector Geffner,et al.
A Framework for Reasoning with Defaults
,
1990
.
[6]
Fahiem Bacchus,et al.
Representing and reasoning with probabilistic knowledge
,
1988
.
[7]
Joseph Y. Halpern,et al.
Statistical Foundations for Default Reasoning
,
1993,
IJCAI.
[8]
Hector J. Levesque,et al.
Reasoning about Noisy Sensors in the Situation Calculus
,
1995,
IJCAI.
[9]
Joseph Y. Halpern.
An Analysis of First-Order Logics of Probability
,
1989,
IJCAI.