文档介绍：TRAINING INTELLIGENT A GENTS USING
HUMAN D A T A
ELIZABETH SKLAR y
ALAN D. BLAIR z
P ABLO FUNES y
JORD AN POLLA CK y
y DEMO L ab, Dept. puter Scienc e,
Br andeis University,
Waltham, MA 02454-9110, USA
E-mail: sklar,p ablo,p ol ******@ du
z Dept. puter Scienc e,
University of Melb ourne,
Parkvil le, Victoria, 3052, A USTRALIA
E-mail: ******@
W e describ e a metho d for training in telligen t agen ts using h uman data collected at
ano v el In learning site where h umans and soft w are agen ts pla y games against
eac h other. T o facilitate h uman learning, it is desirable to select prop er opp onen ts
for h umans so that they will adv ance and not b ecome b ored or frustrated. In the
w ork presen ted here, w e use h uman data as the basis for constructing a p opulation
of graded agen ts, so that in future w e can c ho ose opp onen ts (from this p opulation)
that will c hallenge individual h uman learners appropriately .
Keywor ds: h uman-agen tin teraction, w orks, learning
1 In tro duction.
Hidden inside ev ery mouse clic k and ev ery k ey strok eisv aluable information
that can b e tapp ed, to rev eal something of the h uman who en tered eac h action.
Commercial pro ducts lik e MicrosoftW ord pro vide con text sensitiv e \wizards"
that observ e their users and p op up to assist with curren t tasks. In