Explanatory hypotheses Aetion's computer-assisted reasoning capability is founded upon the principles of abductive inference, which uses commonsense reasoning to assemble the best explanation for a set of data. In contrast with narrower probabilistic approaches, the merit of an explanatory hypothesis is measured on several criteria such as:

Internal consistency

Plausibility (including consistency with background knowledge)

Likelihood

Simplicity (Occam's razor may be applied)

Explanatory power

Specificity

Productive promise

One's willingness to accept a hypothesis H depends upon:

H's relative merit – how decisively it surpasses the alternatives

H's intrinsic merit – how good it is by itself

How thorough the search was for hypotheses that may compete with H

The need to reach a conclusion at this time (instead of, say, gathering further evidence)

The costs of being wrong, and the benefits of being correct

People understand these considerations and routinely justify their own arguments in similar terms. The basic inferential step is:

Exploit confidence A lot of the power and efficiency of abductive inference comes from the way it leverages corroboration and contradiction. The system accepts the essential (only possible explanation) and clear-best hypotheses, the effect of which will typically be to increase confidence in other hypotheses and to rule out incompatible hypotheses. A good indicator of a hypothesis' merit is the number of data that it explains; indeed, if there is no otherwise-unexplained datum that a hypothesis explains, it offers no additional explanatory value. Thus, driven by the currently-unexplained data, the abduction machine forms the most confident conclusion first, and works outward from this island of certainty.

Remain comprehensible The system performs commonsense evidential reasoning to manage uncertainty and finds the best consistent explanation for data. Not only can it automatically guess the real situation but, critically, it can provide an easy-to-understand argument that uses evidence to justify its interpretations and explanations. Therefore, the user can collaborate with the system, controlling its behavior and working with it in its inference process.

Conclusion Our technology provides means to maintain a working model of present reality: the set of mutually consistent hypotheses that is accepted as the best explanation for the observations. More information on the background to this approach can be found in Abductive Inference (1994), edited by John R. Josephson and Susan G. Josephson, ISBN 0-521-43461-0.

See how abuductive inference fits in with our other technologies here.