Creating General Intelligent Systems

1
Creating General Intelligent Systems
Goal of AI Design and construct computational
artifacts that combine many cognitive abilities
in an integrated system ? possess similar
intellectual capacity to humans ? exhibit
intelligence in a general way across multiple
domains
Cognitive Architectures to support general
intelligent systems ? ICARUS (Langley) ? SOAR
(Newell, Laird, Rosenbloom) and ACT-R
(Anderson) ? multi-agent systems (Sycara)
2
Essential elements of cognitive architectures

• Design of any cognitive architecture must
  consider
• short-term and long-term memories that store the
  agents beliefs, goals and knowledge
• representation and organization of structures
  embedded in these memories
• functional processes that operate on these
  structures, including performance mechanisms and
  learning mechanisms
• programming language used to construct
  knowledge-based systems that embody the
  architectures assumptions

ICARUS is also concerned with physical agents
operating in an external environment (e.g.
agent driving in a city)
3
Design principles used in ICARUS

• cognition is grounded in perception and action
• concepts and skills are distinct cognitive
  structures
• long-term memory is organized in a hierarchical
  fashion
• skill and concept hierarchies are acquired in a
  cumulative way
• long-term and short-term structures have strong
  correspondence

These design principles are guided by empirical
observations from psychology
4
Memories in ICARUS
long-term memories store knowledge and
procedures, and change gradually short-term
memories store agents beliefs and goals, and
change rapidly in response to the environment and
the agents agenda
conceptual memories encode knowledge about
classes of objects and relations between
them skill memories encode knowledge about ways
to act and achieve goals
5
Some concepts for in-city driving(long-term
conceptual memory)
((in-rightmost-lane ?self ?clane) percepts
((self ?self) (segment ?seg) (lane-line
?clane segment ?seg)) relations
((driving-well-in-segment ?self ?seg ?clane)
(last-lane ?clane) (not (lane-to-right
?clane ?anylane)))) ((driving-well-in-segment
?self ?seg ?lane) percepts ((self ?self)
(segment ?seg) (lane-line ?lane segment
?seg)) relations ((in-segment ?self ?seg)
(in-lane ?self ?lane) (aligned-with-lane-in
-segment ?self ?seg ?lane)
(centered-in-lane ?self ?seg ?lane)
(steering-wheel-straight ?self))) ((in-lane
?self ?lane) percepts ((self ?self segment
?seg) (lane-line ?lane segment ?seg
dist ?dist)) tests ((gt ?dist -10)
(lt ?dist 0)))
6
Some skills for in-city driving(long-term skill
memory)
((in-rightmost-lane ?self ?line) percepts
((self ?self) (lane-line ?line)) start
((last-lane ?line)) subgoals
((driving-well-in-segment ?self ?seg ?line)))
((driving-well-in-segment ?self ?seg ?line)
percepts ((segment ?seg) (lane-line ?line)
(self ?self)) start ((steering-wheel-st
raight ?self)) subgoals ((in-segment ?self
?seg) (centered-in-lane ?self ?seg
?line) (aligned-with-lane-in-segment ?self
?seg ?line) (steering-wheel-straight
?self))) ((in-segment ?self ?endsg)
percepts ((self ?self speed ?speed)
(intersection ?int cross ?cross)
(segment ?endsg street ?cross angle
?angle)) start ((in-intersection-for-rig
ht-turn ?self ?int)) actions ((?steer
1)))
7
Representing short-term beliefs and
goals(short-term conceptual memory)
(current-street me A) (current-segment me
g550) (lane-to-right g599 g601) (first-lane
g599) (last-lane g599) (last-lane
g601) (at-speed-for-u-turn me) (slow-for-right-tu
rn me) (steering-wheel-not-straight me)
(centered-in-lane me g550 g599) (in-lane me
g599) (in-segment me g550) (on-right-side-in-seg
ment me) (intersection-behind g550
g522) (building-on-left g288) (building-on-left
g425) (building-on-left g427) (building-on-left
g429) (building-on-left g431) (building-on-left
g433) (building-on-right g287) (building-on-right
g279) (increasing-direction me)
(buildings-on-right g287 g279)
8
Perceptual buffer
9
Functional processes of ICARUS architecture
10
Structure and use of conceptual memory
high-level beliefs about current state

• On each cycle
• Match low-level concept definitions in
  long-term memory to perceptions and beliefs
• If theres a match, add instance of concept to
  short-term belief memory (to support other
  inferences)
• Work up to higher-level concepts that match
  against lower-level concepts
• Continue until system has deduced all beliefs
  implied by conceptual knowledge and immediate
  perceptions

match concepts to current percepts
11
Skill execution in ICARUS
high-level goal
Skill execution starts from the current goal and
proceeds in a top-down direction Finds applicable
paths through skill hierarchy that terminate in
executable actions For example
(in-rightmost-lane me ln3)
(driving-well-in-segment me s5 ln3)
executable actions
(in-segment me s5)
(steer 1)
12
ICARUS Learns Skills from Problem Solving
Reactive Execution
no
impasse?
Primitive Skills
Executed plan
yes
Problem Solving
Skill Learning
13
Rule-based cognitive architectures
? SOAR (State Operator And Result) (Newell,
Laird, Rosenbloom, ) ? ACT-R (Adaptive Control
of Thought Rational) (Anderson)
Production systems are based on rules if-then
rules, condition-action rules, production rules,
Working memory analogous to human short-term
memory Productions are part of long-term
memory On each cycle ? productions are matched
to facts in working memory ? if conditions
satisfied, facts added to or deleted from working
memory
14
Organization of information in ACT-R
Intentional Module (not identified)
Declarative Module (Temporal/Hippocampus)
Retrieval Buffer (VLPFC)
Goal Buffer (DLPFC)
Matching (Striatum)
Productions (Basal Ganglia)
Selection (Pallidum)
Execution (Thalamus)
Visual Buffer (Parietal)
Manual Buffer (Motor)
Manual Module (Motor/Cerebellum)
Visual Module (Occipital/etc)
Environment
15
ACT-R
16
Memory structures in SOAR
17
Soar Decision Cycle
Perception
Cognition
Motor
Elaboration Phase

• Fire rules
• Generate preferences
• Update working memory

Input Phase
Output Phase

• Sense world
• Perceptual pre-processing
• Assert to WM

Decision Phase

• Command effectors
• Adjust perception

• Evaluate operator preferences
• Select new operator OR
• Create new state

18
Other approaches to general intelligent systems

• ? Multi-agent systems (Sycara)
• distinct modules for different facets of an
  intelligent system, that communicate directly
• specify inputs/outputs of each module and
  communication protocol
• no constraints on operation of individual
  modules
• ? Blackboard systems (Engelmore Morgan)
• similar to multi-agent systems, but modules read
  and modify a shared memory of beliefs, goals,
  and other short-term structures