Vincent Aleven
Carnegie Mellon University
Abstract
Since
the days
of Bacon and Galileo, formulating
hypotheses
about natural phenomena and testing them against empirical data have
been
cornerstones of the natural sciences. As a cognitive framework,
hypothesis
formation and testing are also important in legal reasoning. The legal
domain,
however, is different from natural science and mathematics in a
significant
respect. Determining whether a hypothesized rule and proposed outcome
are
consistent with past legal decisions is much more a matter of
interpretation. The
aims of this project are to (1) design and evaluate an Artificial
Intelligence (AI)
cognitive model of framing and testing hypotheses in an interpretive
domain,
legal reasoning, and (2) incorporate the model in an intelligent
tutoring
system (ITS) to teach law students the process.
The project builds upon two recent
developments: (1) a
newly invented means to frame and evaluate hypotheses predicting the
outcomes
of new cases based on an AI database of existing precedents; (2) a
convenient,
on-line corpus of U.S. Supreme Court oral arguments in aural and
written form, including
many concrete examples of legal hypothesis framing and testing. In
response to an
advocate’s proposed hypothesis of how the case should be decided, the
Justices often
challenge it by posing hypotheticals,
sometimes
forcing the advocate to modify or abandon the hypothesis.
By studying these examples, the researchers,
participating
law students and law faculty will schematize and model the process of
framing
and testing legal hypotheses, implement it computationally, evaluate it
empirically,
and use it to design the ITS.
The tutor will implement the model in various
legal domains,
each with a body of legal rules, issues, precedents, and principles, operationalized in a way that supports
hypothesis
formulation, prediction, testing, and explanation. Using the model, it
will guide
and challenge students’ arguments. It will predict outcomes of cases,
help
students construct tests and rationales justifying the prediction, and
help
them evaluate the hypothesis by posing or responding to hypothetical
challenges.
The researchers will evaluate the project’s
success in
terms of: (1) the accuracy of the model’s predictions for new cases and
the
extent it improves case retrieval; (2) how well model-generated
arguments compare
to those in the Supreme Court oral arguments or generated by law
students; (3)
how well ITS-trained students compare to a control group taught the
same process
using conventional law school methods; (4) whether ITS-trained students
generate more accurate self-explanations of the Supreme Court oral
arguments.
This work extends AI techniques to a
much less well-structured domain than natural science and mathematics,
one more
like the common sense domains AI has yet to address.
By using AI to investigate empirically a
cognitive phenomenon, framing and testing hypotheses in an interpretive
domain,
it will contribute to research in AI & Law, Case-based Reasoning,
AI &
Education, and Cognitive Science.
*
NSF Grant IIS-0412830, effective September 15, 2004 and expires
February 28, 2006, in the amount of $650,000.