Presentation is loading. Please wait.

Presentation is loading. Please wait.

Pat Langley Computational Learning Laboratory Center for the Study of Language and Information Stanford University, Stanford, California

Similar presentations


Presentation on theme: "Pat Langley Computational Learning Laboratory Center for the Study of Language and Information Stanford University, Stanford, California"— Presentation transcript:

1 Pat Langley Computational Learning Laboratory Center for the Study of Language and Information Stanford University, Stanford, California http://cll.stanford.edu/ Architectures for Intelligent Systems Thanks to Dongkyu Choi, Kirstin Cummings, Wayne Iba, Negin Nejati, David Nicholas, Seth Rogers, Stephanie Sage, and Daniel Shapiro for contributions to this research.

2 In the Beginning... AI monobloc

3 The Big Bang Theory of AI action perception reasoning learning planning language

4 Twenty Questions At the 1971 Carnegie Symposium, Allen Newell argued that: psychology had focused too much on isolated issues; psychology had focused too much on isolated issues; this strategy would never lead us to understand the mind; this strategy would never lead us to understand the mind; we could only achieve this by developing integrated theories. we could only achieve this by developing integrated theories. He claimed You cant play twenty questions with nature and win. The same critique holds for what John Laird has called niche AI.

5 An Alternative to Twenty Questions Newells alternative to twenty questions was that we should: move beyond isolated phenomena and capabilities to develop complete models of intelligent behavior; move beyond isolated phenomena and capabilities to develop complete models of intelligent behavior; demonstrate their intelligence on the same range of domains and tasks as humans can handle; demonstrate their intelligence on the same range of domains and tasks as humans can handle; view artificial intelligence and cognitive psychology as close allies with distinct but related goals; view artificial intelligence and cognitive psychology as close allies with distinct but related goals; evaluate systems in terms of generality and flexibility rather than success on a single application domain. evaluate systems in terms of generality and flexibility rather than success on a single application domain. He associated these aims with the idea of a cognitive architecture.

6 Intelligent Systems as Communicating Modules action perception reasoning learning planning language software engineering / multi-agent systems

7 action perception reasoning learning planning language short-term beliefs and goals Modules that Share Short-Term Memory blackboard architectures

8 Modules that Share Long-Term Memory action perception reasoning learning planning language short-term beliefs and goals long-term memory structures cognitive architectures

9 As traditionally defined and utilized, a cognitive architecture: specifies the infrastructure that holds constant over domains, as opposed to knowledge, which varies. specifies the infrastructure that holds constant over domains, as opposed to knowledge, which varies. models behavior at the level of functional structures and processes, not the knowledge or implementation levels. models behavior at the level of functional structures and processes, not the knowledge or implementation levels. commits to representations and organizations of knowledge and processes that operate on them. commits to representations and organizations of knowledge and processes that operate on them. comes with a programming language for encoding knowledge and constructing intelligent systems. comes with a programming language for encoding knowledge and constructing intelligent systems. Aspects of Cognitive Architectures Early candidates were cast as production system architectures, but alternatives have gradually expanded the known space.

10 Examples of Cognitive Architectures ACTSoarP RODIGY EPICRCS APEX C LARION Dynamic Memory Society of Mind CAPS GIPS 3T

11 Newells vision for research on theories of intelligence was that: cognitive architectures should incorporate strong theoretical assumptions about the nature of the mind. cognitive architectures should incorporate strong theoretical assumptions about the nature of the mind. architectural designs should change only gradually, as new structures or processes are determined necessary. architectural designs should change only gradually, as new structures or processes are determined necessary. later design choices should be constrained heavily by earlier ones, not made independently. later design choices should be constrained heavily by earlier ones, not made independently. Listen to the Architecture The research program was all about mutual constraints, and it was important to listen to the architecture. A successful architecture should provide a unified theory of intelligent behavior.

12 Mutually Constraining Components action perception reasoning learning planning language short-term beliefs and goals long-term memory structures

13 The I CARUS Architecture supports not only cognition but also perception and action; supports not only cognition but also perception and action; always operates in the context of an external environment. always operates in the context of an external environment. We can use a candidate architecture, I CARUS, to illustrate the idea of mutually constraining components. Traditional architectures were influenced by theories of human problem solving and other mental phenomena. In contrast, I CARUS is an architecture for physical agents that: I CARUS emphasis on embodied agents distinguishes it from production system architectures like ACT-R and Soar.

14 I CARUS Theoretical Constraints 1.Cognitive reality of physical objects 2.Cognitive separation of categories and skills 3.Hierarchical organization of long-term memory 4.Correspondence of long-term/short-term structures 5.Modulation of symbolic structures with utility functions 6.Primacy of categorization and execution over problem solving I CARUS also incorporates some other strong design principles: These constraints influence the nature of I CARUS agents and the manner in which one constructs them.

15 Overview of the I CARUS Architecture* Long-TermConceptualMemory Long-Term Skill Memory Short-TermConceptualMemory Short-Term Categorization and Inference SkillExecution Perception Environment PerceptualBuffer * without learning MotorBuffer SkillRetrieval

16 Organization of Long-Term Memory concepts skills Each concept is defined in terms of other concepts and/or percepts. Each skill is defined in terms of other skills, concepts, and percepts. I CARUS organizes both concepts and skills in a hierarchical manner.

17 Organization of Long-Term Memory conceptsskills For example, the skill highlighted here refers directly to the highlighted concepts. I CARUS interleaves its long-term memories for concepts and skills.

18 An In-City Driving Environment

19 The I CARUS Language: Driving Concepts (in-segment (?self ?sg) :percepts((self ?self segment ?sg) (segment ?sg))) (aligned-with-lane (?self ?lane) :percepts((self ?self) (lane-line ?lane angle ?angle)) :positives((in-lane ?self ?lane)) :tests((> ?angle -0.05) ( ?angle -0.05) (< ?angle 0.05)) ) (on-street (?self ?packet) :percepts((self ?self) (packet ?packet street ?street) (segment ?sg street ?street)) :positives((not-delivered ?packet) (current-segment ?self ?sg)) :value80 ) (increasing-direction (?self) :percepts((self ?self)) :positives((increasing ?b1 ?b2)) :negatives((decreasing ?b3 ?b4)) )

20 (turn-around-on-street (?self ?packet) :percepts((self ?self segment ?segment direction ?dir) (building ?landmark)) :start ((on-street-wrong-direction ?packet)) :effects((on-street-right-direction ?packet)) (building ?landmark)) :start ((on-street-wrong-direction ?packet)) :effects((on-street-right-direction ?packet)) :ordered((get-in-U-turn-lane ?self) (prepare-for-U-turn ?self) :ordered((get-in-U-turn-lane ?self) (prepare-for-U-turn ?self) (steer-for-U-turn ?self ?landmark)) ) (steer-for-U-turn ?self ?landmark)) ) (get-aligned-in-segment (?self ?sg) :percepts((lane-line ?lane angle ?angle)) :requires((in-lane ?self ?lane)) :effects((aligned-with-lane ?self ?lane)) :actions(( steer ( times ?angle 2))) ) (steer-for-right-turn (?self ?int ?endsg) :percepts((self ?self speed ?speed) (intersection ?int cross ?cross) (segment ?endsg street ?cross angle ?angle)) :start((ready-for-right-turn ?self ?int)) :effects((in-segment ?self ?endsg)) :actions(( times steer 2)) ) :actions(( times steer 2)) ) The I CARUS Language: Driving Skills

21 The I CARUS Control Cycle 1.places descriptions of sensed objects in the perceptual buffer; 2.infers instances of concepts implied by the current situation; 3.finds paths through the skill hierarchy from top-level intentions; 4.selects one or more applicable skill paths for execution; 5.invokes the actions associated with each selected path. On each successive execution cycle, the I CARUS architecture: Thus, I CARUS is an example of what Nilsson (2003) refers to as a three-tower architecture.

22 Basic I CARUS Processes concepts skills Concepts are matched bottom up, starting from percepts. Skill paths are matched top down, starting from intentions. I CARUS matches patterns to recognize concepts and select skills.

23 Perceptual Buffer from Driving Domain (self me speed 24.0 wheel-angle 0.02 limit 25.0 road-angle 0.06) (segment g1059 street 2 dist -5.0 latdist 15.0) (segment g1050 street A dist -45.0 latdist nil) (segment g1049 street A dist oor latdist nil) (lane-line g1073 length 100.0 width 0.5 dist 35.0 angle 1.57 color white) (lane-line g1074 length 100.0 width 0.5 dist 15.0 angle 1.57 color white) (lane-line g1072 length 100.0 width 0.5 dist 25.0 angle 1.57 color yellow) (lane-line g1100 length 100.0 width 0.5 dist -15.0 angle 0.0 color white) (lane-line g1101 length 100.0 width 0.5 dist 5.0 angle 0.0 color white) (lane-line g1099 length 100.0 width 0.5 dist -5.0 angle 0.0 color yellow) (lane-line g1104 length 100.0 width 0.5 dist 5.0 angle 0.0 color white) (intersection g1021 street A cross 2 dist -5.0 latdist nil) (building g943 address 246 c1dist 43.69 c1angle -0.73 c2dist nil c2angle nil) (building g941 address 246 c1dist 30.10 c1angle -1.30 c2dist 43.70 c2angle -0.73) (building g939 address 197 c1dist 30.10 c1angle -1.30 c2dist 33.40 c2angle -2.10) (building g943 address 172 c1dist 33.40 c1angle -2.09 c2dist 50.39 c2angle -2.53) (sidewalk g975 dist 15.0 angle 0.0) (sidewalk g978 dist 5.0 angle 1.57)

24 Short-Term Memory from Driving Domain (buildings-on-right me g2231 g2230 g2480)(increasing me g2231 g2230 g2480) (buildings-on-right me g2231 g2222 g2480)(increasing me g2231 g2222 g2480) (buildings-on-right me g2231 g2211 g2480)(increasing me g2231 g2211 g2480) (buildings-on-right me g2230 g2222 g2480)(increasing me g2230 g2222 g2480) (buildings-on-right me g2230 g2211 g2480)(increasing me g2230 g2211 g2480) (buildings-on-right me g2222 g2211 g2480)(increasing me g2222 g2211 g2480) (buildings-on-left me g2366 g2480)(buildings-on-left me g2368 g2480) (buildings-on-left me g2370 g2480)(buildings-on-left me g2372 g2480) (not-on-street me g2980)(currrent-building me g2222) (not-approaching-cross-street me g2980)(not-on-cross-street me g2980) (current-street me A) (current-segment me g2480) (not-delivered g2980) (in-U-turn-lane me g2533) (in-leftmost-lane me g2533)(lane-to-right me g2533) (fast-for-right-turn me)(fast-for-U-turn me) (driving-in-segment me g2480 g2533)(at-speed-for-cruise me) (steering-wheel-straight me)(centered-in-lane me g2533) (aligned-with-lane me g2533)(in-lane me g2533) (on-right-side-of-road me)(in-segment me g2480)

25 Basic I CARUS Architecture action perception reasoning short-term beliefs and goals long-term memory structures

26 The Challenge of Hierarchy Learning are consistent with knowledge of human behavior; are consistent with knowledge of human behavior; operate in an incremental and cumulative manner; operate in an incremental and cumulative manner; satisfy constraints imposed by other I CARUS components. satisfy constraints imposed by other I CARUS components. I CARUS commitment to hierarchical organization raises a serious question about the origin of its structures. We want mechanisms which acquire these structures in ways that: This requires some source of experience from which to create hierarchical structures.

27 An Approach to Hierarchy Learning decomposes complex problems into subproblems; decomposes complex problems into subproblems; relies on heuristic search to find useful decompositions. relies on heuristic search to find useful decompositions. We have extended I CARUS to incorporate a module for means-ends problem solving that: When I CARUS cannot execute a skill because its start condition is unmet, this mechanism: chains backward off skills that would achieve the condition; or chains backward off skills that would achieve the condition; or chains backward off definitions of the unsatisfied concept. chains backward off definitions of the unsatisfied concept. Traces of successful problem solving serve as the basis for new hierarchical structures.

28 The Extended I CARUS Architecture* Long-TermConceptualMemory Long-Term Skill Memory Short-TermConceptualMemory Goal/SkillStack Categorization and Inference SkillExecution Perception Environment PerceptualBuffer * without learning Means-EndsAnalysis MotorBuffer SkillRetrieval

29 1 2 8 5 3 6 4 7 9 1011 An Abstract Means-Ends Trace

30 new skills based on backward chaining over skills; new skills based on backward chaining over skills; new skills based on backward chaining over concepts: new skills based on backward chaining over concepts: skills that achieve the desired concept; skills that achieve the desired concept; guard skills that ensure execution in desired order; guard skills that ensure execution in desired order; concepts for the start fields of newly built skills. concepts for the start fields of newly built skills. I CARUS learns four types of structures from means-ends traces: Each of these is based on a single solution to an impasse and may build on skills and concepts learned previously. Learning is fully embedded within problem solving and execution. Learning Skills from Means-Ends Traces

31 1 2 8 5 3 6 4 7 A 9 1011 Learning Skills from Means-Ends Traces concept chaining

32 1 2 8 5 3 6 4 7 A B 9 1011 Learning Skills from Means-Ends Traces skill chaining

33 1 2 8 5 3 6 4 7 A B 9 1011 C Learning Skills from Means-Ends Traces concept chaining

34 1 2 8 5 3 6 4 7 A B D 9 1011 C Learning Skills from Means-Ends Traces skill chaining

35 1 2 8 5 3 6 4 7 A B D E 9 1011 C Learning Skills from Means-Ends Traces concept chaining

36 Results from Chaining Driving Skills In the driving domain, skill chaining produces structures like: (s-skill-5 (?me ?packet) :percepts((self ?me) (packet ?packet) :start((concept-1 ?me)) :effects((delivered ?packet)) :ordered((s-skill-4 ?me ?packet) (drop-off ?me ?packet))) (s-skill-4 (?me ?packet) :percepts((self ?me) (packet ?packet)) :start((current-seg ?me ?sg) (street-4-ahead-on-A ?me)) :effects((at-address ?me ?packet ?sg ?building)) :ordered((s-skill-3 ?me ?packet) (continue-to-address ?me ?packet))) (continue-to-address ?me ?packet))) (s-skill-1 (?me) :percepts((self ?me)) :start((street-4-ahead-on-A ?me)) :objective((street-C-ahead-on-4 ?me)) :ordered((straight-until-4-on-A ?me) (turn-onto-4 ?me)))

37 More Results from Skill Chaining In the blocks world, skill chaining can produce structures like: (s-skill-21 (?A ?T1 ?B ?C) :percepts((block ?A) (table ?T1) (block ?B) (block ?C)) :percepts((block ?A) (table ?T1) (block ?B) (block ?C)) :start((concept-46 ?A ?T1 ?B ?C)) :start((concept-46 ?A ?T1 ?B ?C)) :effects((on ?B ?C)) :effects((on ?B ?C)) :ordered((s-skill-20 ?A ?T1 ?B) (stack ?B ?C)) :ordered((s-skill-20 ?A ?T1 ?B) (stack ?B ?C)) :duration4 ) :duration4 ) (s-skill-20 (?A ?T1 ?B) :percepts((block ?A) (table ?T1) (block ?B)) :percepts((block ?A) (table ?T1) (block ?B)) :start((concept-43 ?A ?T1 ?B)) :start((concept-43 ?A ?T1 ?B)) :effects((holding ?A)) :effects((holding ?A)) :ordered((c-skill-22 ?A ?T1 ?B) (pickup ?B ?T1)) :ordered((c-skill-22 ?A ?T1 ?B) (pickup ?B ?T1)) :duration3 ) :duration3 ) (concept-46 (?A ?T1 ?B ?C) :percepts((block ?A) (table ?T1) (block ?B) (block ?C)) :percepts((block ?A) (table ?T1) (block ?B) (block ?C)) :positives((concept-43 ?A ?T1 ?B) (clear ?C)) ) :positives((concept-43 ?A ?T1 ?B) (clear ?C)) )

38 Extended I CARUS Architecture action perception reasoning learning planning short-term beliefs and goals long-term memory structures

39 Modularity of Knowledge Elements can be added and deleted easily by a developer; can be added and deleted easily by a developer; enable automated composition through problem solving; enable automated composition through problem solving; support learning through composition and modification. support learning through composition and modification. The notion of modules is central to software engineering, and cognitive architectures benefit from similar ideas. Fine-grained modularity occurs with knowledge elements (e.g., rules or skills), which: This approach requires that knowledge elements share symbols, but otherwise is very general.

40 Modularity of Architectural Components seems closer to the coarse-grained view common in software engineering circles; seems closer to the coarse-grained view common in software engineering circles; occurs not through communication protocols but rather through shared knowledge structures; occurs not through communication protocols but rather through shared knowledge structures; is highly constrained due to cognitive architectures strong representational assumptions. is highly constrained due to cognitive architectures strong representational assumptions. Another form of modularity involves replacing one architectural component with another, which: Nevertheless, such component-level modularity is possible for architectures in the same representational class.

41 Modularity in PRISM (switchesdecay-strategytime-decay ordering-strategies(strength summed-activation randomize) selection-strategies(select-best select-best select-one) strengthening-function ( * x 1.1) weakening-function( x 0.5) discrimination-conditions (positive negative) ) PRISM (Langley & Neches, 1981) was an early framework that supported this notion of component-level modularity. PRISM offered a functional decomposition of production system architectures with alternative modules for each dimension.

42 Modularity of Task Knowledge knowledge is organized into sets of elements relevant to handling large-scale tasks; knowledge is organized into sets of elements relevant to handling large-scale tasks; these modules share some structures for communicating requests and results; these modules share some structures for communicating requests and results; the cognitive architecture is fixed and shared among these distinct modules. the cognitive architecture is fixed and shared among these distinct modules. Another form of coarse-grained modularity, which occurs at the knowledge level, assumes that: This form of modularity in cognitive architectures comes closest to its intended use in software engineering. Moreover, it supports hierarchical decomposition of knowledge available to an intelligent system.

43 Evaluation of Intelligent Systems Experimental studies of intelligent systems have lagged behind ones for component methods because: they focus on more complex, multi-step behavior; they focus on more complex, multi-step behavior; they require more engineering to develop them; they require more engineering to develop them; they rely on interaction among their components. they rely on interaction among their components. Together, these factors have slowed the widespread adoption of experimental evaluation.

44 Repositories for Intelligent Systems Public repositories are now common among the AI subfields, and they offer clear advantages for research by: providing fast and cheap materials for experiments; providing fast and cheap materials for experiments; supporting replication and standards for comparison. supporting replication and standards for comparison. However, they can also produce undesirable side effects by: focusing attention on a narrow class of problems; focusing attention on a narrow class of problems; encouraging a bake-off mentality among researchers. encouraging a bake-off mentality among researchers. To support research on intelligent systems, we need testbeds and environments designed with them in mind.

45 Desirable Characteristics of Testbeds Testbeds that are designed to support research on intelligent systems should: include a variety of domains to ensure generality; include a variety of domains to ensure generality; be well documented and simple for researchers to use; be well documented and simple for researchers to use; have standard formats to ease interface with systems. have standard formats to ease interface with systems. However, these features are already present in many existing repositories, and other characteristics are necessary.

46 Desirable Characteristics of Testbeds In addition, testbeds for integrated intelligent systems should: contain not data sets but task environments contain not data sets but task environments which support agents that exist over time which support agents that exist over time at least some of which involve physical domains at least some of which involve physical domains provide an infrastructure to ease experimentation with provide an infrastructure to ease experimentation with external databases (e.g., geographic information systems) external databases (e.g., geographic information systems) controlled capture, replay, and restart of scenarios controlled capture, replay, and restart of scenarios methods for recording performance measures methods for recording performance measures These features will reduce greatly the cost of experimentation.

47 Physical vs. Simulated Environments For domains that involve external settings, one can either a physical or a simulated environment for evaluation. Simulated environments have many advantages, including: ability to vary domain parameters and physical layout; ability to vary domain parameters and physical layout; ease of recording traces of behavior and cognitive state. ease of recording traces of behavior and cognitive state. One can make simulated environments more realistic by: using simulators that support kinematics and dynamics; using simulators that support kinematics and dynamics; including data from real sensors in analogous locations. including data from real sensors in analogous locations. This approach combines the relevance of physical testbeds with the affordability of synthetic ones.

48 Some Promising Domains A number of domains hold promise for the experimental study of intelligent systems: urban search and rescue (Balakirsky & Messina, 2002); urban search and rescue (Balakirsky & Messina, 2002); flying aircraft on military missions (Jones et al., 1999); flying aircraft on military missions (Jones et al., 1999); driving a vehicle in a city (Choi et al., 2004); driving a vehicle in a city (Choi et al., 2004); playing strategy games (Aha & Molineaux, 2004); playing strategy games (Aha & Molineaux, 2004); general game playing (Genesereth, 2004). general game playing (Genesereth, 2004). Each requires the integration of cognition, perception, and action in a complex, dynamical setting. However, successful testbeds also require a suite of tasks that can be used to drive research and evaluate systems.

49 Concluding Remarks are embedded within a unified cognitive architecture; are embedded within a unified cognitive architecture; incorporate modules that provide mutual constraints; incorporate modules that provide mutual constraints; demonstrate a wide range of intelligent behavior; demonstrate a wide range of intelligent behavior; are evaluated on multiple tasks in challenging testbeds. are evaluated on multiple tasks in challenging testbeds. We need more research on integrated intelligent systems that: We must also think about ways to overcome social obstacles to pursuing this research agenda.

50 End of Presentation


Download ppt "Pat Langley Computational Learning Laboratory Center for the Study of Language and Information Stanford University, Stanford, California"

Similar presentations


Ads by Google