Commonsense Reasoning In BDI Agents
Speaker: Jeff Blee, Griffith University
When: 2006-06-20 10:30:00
Venue: N13 0.23. Nathan campus Environment 2 Building, Griffith University
Host: Prof Abdul Sattar
Abstract:Agent Technology is slowly becoming more accepted within the
software development industry and the business community, but it is
essentially still in its infancy. While some agent applications are
appearing what is needed is some âkiller applicationsâ to grab
peopleâs attention. These applications may well come in domains
where computer programs have not even be attempted to be written
before. One domain which is ripe for, and has received a small
amount of, attention is that of Assistant Agents, and particularly
Smart Personal Assistants, agents that are tailored (or can tailor
themselves) to assist humans in their everyday IT and Internet
related tasks. This, and other domains, requires agents to exhibit
human-like reasoning. Modern agents, with goals and plans are some
small way towards this goal. BDI agents, with their human like
mentalistic notions of belief, desire and intention are further in
the right direction. Conventional agents tend to use logics that
are monotonic in nature, that is, they are not able to reason with
the uncertainty inherent in peoplesâ everyday life. There are
nonmonotonic logics, and other logics that reason with uncertainty,
that have been integrated or melded into agents, but none so far
that have been totally successful or that have produced successful
implementations or platforms.
Philosophically and intuitively, it can be stated that humanâs
beliefs are divided into levels of certainty or plausibility, such
as: âI am absolutely sure ofâ; âthis usually happensâ; it may or may
not happenâ; this might be the caseâ; and so on. A similar division
can be made within desires and intentions. What is being suggested
for this research is to investigate the division of the mentalistic
notions of BDI into these âlevelsâ. The proposal is that this would
be a more effective way of reasoning with uncertainty than current
nonmonotonic logics, or agent integrations. The prime choices for
targeting would be the multi-modal logic of Rao & Georgeff, the
language AgentSpeak(L) and its implemented platform JASON. If the
investigations follow the expected course, a suitable example
application will be built from the resulting platform to demonstrate
the viability of the concept.
Biography:(biography unavailable)
Type: Ph.D confirmation
Contact:Prof Abdul Sattar, seminar host (n.dunstan@griffith.edu.au)
or Guido Governatori (ITEE seminar co-ordinator)
(guido@itee.uq.edu.au)