Information Technology Reference
In-Depth Information
reasoning about how to modify the agent's plan, TMKL2 supports reasoning about
how to diagnose and redesign the agent itself.
17.5.2 Meta-Reasoning
Meta-reasoning refers to thinking about thinking [
3
,
16
,
17
]. In this article, we have
focused mostly on meta-reasoning for agent self-adaptation [
32
]. Much research on
meta-reasoning for self-adaptation has used self-models of agents to help localize
and identify modifications to the agent design, e.g., [
4
,
6
,
12
,
26
,
27
,
32
,
37
,
43
,
52
-
54
,
73
,
74
].
We can trace several themes in model-based self-adaptation in intelligent agents.
Firstly, self-adaptations can be retrospective [
4
,
6
,
12
,
26
,
27
,
37
,
43
,
52
,
73
,
74
], i.e.,
after the agent has executed an action in the world and received some feedback on the
result, or proactive [
53
,
54
], i.e., when the agent is given a newgoal similar and related
to but different from its original goal. The recent work on goal-driven autonomy [
51
]
appears to be related to proactive self-adaptation. As our four adaptation experiments
in Sect.
17.5
indicate, GAIA can handle both proactive and retrospective adaptations.
Secondly, the self-adaptation in an intelligent agent may pertain to the deliberative
element in an agent's architecture [
4
,
6
,
12
,
26
,
27
,
32
,
37
,
43
,
52
-
54
,
73
], or
the reactive element [
74
]. The experiments described in this chapter pertain to the
deliberative element but in principle GAIA should be able to manage both.
Thirdly, self-adaptation in an intelligent agent may pertain to the agent's reasoning
[
4
,
6
] Collins et al [
12
][
43
,
52
-
54
], or the agent's knowledge [
26
,
27
,
32
,
37
,
73
].
The experiments described in this chapter pertain to revising the reasoning, but again
in principle GAIA should be able to address both.
Fourthly, in general the meta-reasoner can call upon special-purpose reasoners.
For example, the original REM system invoked a situated learner when needed [
54
,
75
].
Autonomic computing [
40
] represents a different but related line of research.
Autonomic computing pertains to self-managing software systems, including systems
capable of self-configuration, self-optimization, self-healing, and self-protection.
The experiments described in Sect.
17.5
exhibit elements of self-configuration and
self-healing, and to this extent, they may qualify as autonomic computing systems.
17.5.3 Game Playing
Over the last generation, interactive games have emerged as an important domain
for AI research [
1
,
15
,
38
,
41
,
42
,
46
,
50
,
58
,
59
,
61
,
65
,
75
,
80
]. Laird and Van
Lent [
42
] call interactive games the “killer application” for human-level AI. Love,
Hinrichs and Genesereth [
46
] describe GDL, a general game description language.
Search WWH ::
Custom Search