Proceedings of the AAAI Symposium Series, Jan 21, 2024
Large language models (LLMs) provide capabilities far beyond sentence completion, including quest... more Large language models (LLMs) provide capabilities far beyond sentence completion, including question answering, summarization, and natural-language inference. While many of these capabilities have potential application to cognitive systems, our research is exploiting language models as a source of task knowledge for cognitive agents, that is, agents realized via a cognitive architecture. We identify challenges and opportunities for using language models as an external knowledge source for cognitive systems and possible ways to improve the effectiveness of knowledge extraction by integrating extraction with cognitive architecture capabilities, highlighting with examples from our recent work in this area.
Our long-term goal is to develop autonomous robotic systems that have the cognitive abilities of ... more Our long-term goal is to develop autonomous robotic systems that have the cognitive abilities of humans, including communication, coordination, adapting to novel situations, and learning through experience. Our approach rests on the recent integration of the Soar cognitive architecture with both virtual and physical robotic systems. Soar has been used to develop a wide variety of knowledge-rich agents for complex virtual environments, including distributed training environments and interactive computer games. For development and testing in robotic virtual environments, Soar interfaces to a variety of robotic simulators and a simple mobile robot. We have recently made significant extensions to Soar that add new memories and new non-symbolic reasoning to Soar's original symbolic processing, which should significantly improve Soar abilities for control of robots. These extensions include episodic memory, semantic memory, reinforcement learning, and mental imagery. Episodic memory and semantic memory support the learning and recalling of prior events and situations as well as facts about the world. Reinforcement learning provides the ability of the system to tune its procedural knowledgeknowledge about how to do things. Mental imagery supports the use of diagrammatic and visual representations that are critical to support spatial reasoning. We speculate on the future of unmanned systems and the need for cognitive robotics to support dynamic instruction and taskability.
Soar, an architecture for problem solving and learning based on heuristic search and chunking, ha... more Soar, an architecture for problem solving and learning based on heuristic search and chunking, has been applied to a variety of tasks during the development of the Soar project, the goal of which is to build a system capable of general intelligent behavior. The hypothesis being tested in this aspect of Soar research is that chunking, a simple experience-basmd learning mechanism, can form the basis for a general learning mechanism. Previous work has demonstrated how the combination of chunking and Soar could acquire search-control knowledge (strategy acquisition) and operator implementation rules in both search-based puzzle tasks and knowledge-based expert systems tasks. This paper provides a new demonstration of the capabilities of chunking in the context of the macro-operator technique and shows how: (1) this technique can be used in a general, learning problem solver without the addition of new mechanisms; (2) the learning can be incremental during problem solving rather than requiring a preprocessing phase; (3) the macros can be used for any goal stated in the problem; and (4) additional generality can be obtained via transfer of learning between macro-operators if an appropriate representation of the task is available. References and a distribution list are provided. (KM)
Incorrect knowledge can be a problem for any in-telligent system. Soar is a proposal for the unde... more Incorrect knowledge can be a problem for any in-telligent system. Soar is a proposal for the under-lying architecture that supports intelligence. It has a single representation of long-term memory and a single learning mechanism called chunking. This paper investigates the ...
Chunks have long been proposed as a basic organizational unit for human memory. More recently chu... more Chunks have long been proposed as a basic organizational unit for human memory. More recently chunks have been used to model human learning on simple perceptual-motor skills. In this paper we describe recent progress in extending chunking to be a general learning mechanism by implementing it within a general problem solver. Using the Soar problem-solving architecture, we take significant steps toward a general problem solver that can learn about all aspects of its behavior. We demonstrate chunking in Soar on three tasks: the Eight Puzzle, Tic-Tat-Toe, and a part of the RI computer-configuration task. Not only is there improvement with practice, but chunking also produces significant transfer of learned behavior, and strategy acquisition.
Proceedings of the ... AAAI Conference on Artificial Intelligence, Aug 4, 2011
Effective access to knowledge within large declarative memory stores is one challenge in the deve... more Effective access to knowledge within large declarative memory stores is one challenge in the development and understanding of long-living, generally intelligent agents. We focus on a sub-component of this problem: given a large store of knowledge, how should an agent's task-independent memory mechanism respond to an ambiguous cue, one that pertains to multiple previously encoded memories. A large body of cognitive modeling work suggests that human memory retrievals are biased in part by the recency and frequency of past memory access. In this paper, we evaluate the functional benefit of a set of memory retrieval heuristics that incorporate these biases, in the context of the word sense disambiguation task, in which an agent must identify the most appropriate word meaning in response to an ambiguous linguistic cue. In addition, we develop methods to integrate these retrieval biases within a task-independent declarative memory system implemented in the Soar cognitive architecture and evaluate their effectiveness and efficiency in three commonly used semantic concordances.
This paper brings together work in modeling episodic memory and reinforcement learning. We demons... more This paper brings together work in modeling episodic memory and reinforcement learning. We demonstrate that is possible to learn to use episodic memory retrievals while simultaneously learning to act in an external environment. In a series of three experiments we investigate learning what to retrieve from episodic memory and when to retrieve it, learning how to use temporal episodic memory retrievals, and learning how to build cues that are the conjunctions of multiple features. Our empirical results demonstrate that it is computationally feasible to learn to use episodic memory in all three experiments, and furthermore, that learning to use internal episodic memory accomplishes tasks that reinforcement learning alone does not. These experiments also expose some important interactions that arise between reinforcement learning and episodic memory.
Episodic memory endows autonomous agents with useful cognitive capabilities. However, for long-li... more Episodic memory endows autonomous agents with useful cognitive capabilities. However, for long-lived agents, there are numerous unexplored computational challenges in supporting useful episodic-memory functions while maintaining real-time reactivity. This paper presents and summarizes the evaluation of an algorithmic variant to the task-independent episodic memory of Soar that expands the class of tasks and cues the mechanism can support while remaining reactive over long agent lifetimes.
Over time, the range and complexity of behaviors in computer-generated forces has expanded as the... more Over time, the range and complexity of behaviors in computer-generated forces has expanded as they move to include more autonomy and intelligence. To support the representation and execution of these behaviors, there is a tendency to add more and more features to the underlying software architecture. However, with the addition of these features, two potential costs may be incurred: increased execution time and additional memory requirements. As architectures progress, it is important to continually evaluate the cost and value of each new architectural feature. Seemingly very similar architectures may require significantly different resources; small changes to the features in a single architecture may have a large impact on its performance. Thus, it is necessary to understand and to quantify the resources consumed by different architectures and by the components of a single architecture. Unfortunately, there is no standard method for evaluating features of an architecture or for comparing sets of architectures. In this paper, we discuss a methodology for evaluating a specific architecture. We dissect the Soar architecture into a core set of functionality and examine how incrementally adding each of the features found in the original implementation affects the overall performance and resource requirements. Finally, we show how the same methodology can be used to compare two different architectures. We discuss initial results of a comparison that indicates both qualitative and quantitative differences between the Soar and CLIPS architectures.
Learning novel tasks is a complex cognitive activity requiring the learner to acquire diverse dec... more Learning novel tasks is a complex cognitive activity requiring the learner to acquire diverse declarative and procedural knowledge. Prior ACT-R models of acquiring task knowledge from instruction focused on learning procedural knowledge from declarative instructions encoded in semantic memory. In this paper, we identify the requirements for designing computational models that learn task knowledge from situated taskoriented interactions with an expert and then describe and evaluate a model of learning from situated interactive instruction that is implemented in the Soar cognitive architecture.
A rarely studied issue with using persistent computational models is whether the underlying compu... more A rarely studied issue with using persistent computational models is whether the underlying computational mechanisms scale as knowledge is accumulated through learning. In this paper we evaluate the declarative memories of Soar: working memory, semantic memory, and episodic memory, using a detailed simulation of a mobile robot running for one hour of real-time. Our results indicate that our implementation is sufficient for tasks of this length. Moreover our system executes orders of magnitudes faster than real-time, with relatively modest storage requirements. We also project the computational resources required for extended operations.
We are creating an environment in which to investigate the role of advanced AI in computer games.... more We are creating an environment in which to investigate the role of advanced AI in computer games. This environment is based on the Unreal Tournament (UT) game engine and the Soar AI engine. Unreal provides a 3D virtual environment, while Soar provides a flexible architecture for developing complex AI characters. This paper describes our progress to date, starting with our game, Haunt 2, which is designed so that complex AI characters will be critical to the success (or failure) of the game. We also describe the extensions we have made to UT to support AI characters with complex physiology so that the AI characters' behavior is driven by their interaction with their environment, their internal long-term goals, and any story-based goals. Finally, we describe the overall system design and interfaces between Soar and UT to support flexible development as well as efficient implementation.
The artificial intelligence (AI) components of computer games often appear to be very complex, po... more The artificial intelligence (AI) components of computer games often appear to be very complex, possibly having abilities beyond the state of the art in computer generated forces. In this paper we study the similarities and differences between AIs for computer games and computer generated forces (CGFs). We contrast the goals of AIs and CGFs, their behavioral requirements, and the underlying resources available for developing and fielding them, with an eye to how they impact the complexity of their behaviors. Our conclusion is that CGFs are currently far ahead of game AIs, but that this may change soon. We argue that computer games have advantages for doing certain types of research on complex, human-level behavior. We support this argument with a demonstration of research we have done on AI and computer games. We have developed the Soar Quakebot, which is a Soar program that plays the death match version of Quake II. The design of the Soar Quakebot is based on TacAir-Soar, a real-time expert system that flies U.S. military air missions in simulation, and that is used for training in the U.S. Air Force. The Soar Quakebot incorporates complex tactics and the ability of the bot to anticipate the actions of its enemy.
International Joint Conference on Artificial Intelligence, Aug 8, 1983
The weak methods occur pervasively in Al systems and may form the basic methods for all intellige... more The weak methods occur pervasively in Al systems and may form the basic methods for all intelligent systems. The purpose of this paper is to characterize the weak methods and to explain how and why they arise in intelligent systems. We propose an organization, called a universal weak method, that provides functionality of all the weak methods. A universal weak method is an organizational scheme for knowledge that produces the appropriate search behavior given the available task-domain knowledge. We present a problem solving architecture in which we realize a universal weak method. We also demonstrate the universal weak method with a variety of weak methods on a set of tasks. 1
Innovative Applications of Artificial Intelligence, Jul 25, 2004
We are creating an environment for investigating the role of advanced AI in interactive, story-ba... more We are creating an environment for investigating the role of advanced AI in interactive, story-based computer games. This environment is based on the Unreal Tournament (UT) game engine and the Soar AI engine. Unreal provides a 3D virtual environment, while Soar provides a flexible architecture for developing complex AI characters. This paper describes our progress to date, starting with our game, Haunt 2, which is designed so that complex AI characters will be critical to the success (or failure) of the game. It addresses design issues with constructing a plot for an interactive storytelling environment, creating synthetic characters for that environment, and using a story director agent to tell the story with those characters.
This is a detailed analysis and comparison of the ACT-R and Soar cognitive architectures, includi... more This is a detailed analysis and comparison of the ACT-R and Soar cognitive architectures, including their overall structure, their representations of agent data and metadata, and their associated processing. It focuses on working memory, procedural memory, and long-term declarative memory. I emphasize the commonalities, which are many, but also highlight the differences. I identify the processes and distinct classes of information used by these architectures, including agent data, metadata, and meta-process data, and explore the roles that metadata play in decision making, memory retrievals, and learning.
We propose a computational model of situated language comprehension based on the Indexical Hypoth... more We propose a computational model of situated language comprehension based on the Indexical Hypothesis that generates meaning representations by translating amodal linguistic symbols to modal representations of beliefs, knowledge, and experience external to the linguistic system. This Indexical Model incorporates multiple information sources, including perceptions, domain knowledge, and short-term and long-term experiences during comprehension. We show that exploiting diverse information sources can alleviate ambiguities that arise from contextual use of underspecific referring expressions and unexpressed argument alternations of verbs. The model is being used to support linguistic interactions in Rosie, an agent implemented in Soar that learns from instruction.
The ultimate goal of work in cognitive architecture is to provide the foundation for a system cap... more The ultimate goal of work in cognitive architecture is to provide the foundation for a system capable of general intelligent behavior. That is, the goal is to provide the underlying structure that would enable a system to perform the full range of cognitive tasks, employ the full range of problem-solving methods and representations appropriate for the tasks, and learn about all aspects of the tasks and its performance on them. In this article we present Soar, an implemented proposal for such an architecture. We describe its organizational principles, the system as currently implemented, and demonstrations of its capabilities.
Proceedings of the AAAI Symposium Series, Jan 21, 2024
Large language models (LLMs) provide capabilities far beyond sentence completion, including quest... more Large language models (LLMs) provide capabilities far beyond sentence completion, including question answering, summarization, and natural-language inference. While many of these capabilities have potential application to cognitive systems, our research is exploiting language models as a source of task knowledge for cognitive agents, that is, agents realized via a cognitive architecture. We identify challenges and opportunities for using language models as an external knowledge source for cognitive systems and possible ways to improve the effectiveness of knowledge extraction by integrating extraction with cognitive architecture capabilities, highlighting with examples from our recent work in this area.
Our long-term goal is to develop autonomous robotic systems that have the cognitive abilities of ... more Our long-term goal is to develop autonomous robotic systems that have the cognitive abilities of humans, including communication, coordination, adapting to novel situations, and learning through experience. Our approach rests on the recent integration of the Soar cognitive architecture with both virtual and physical robotic systems. Soar has been used to develop a wide variety of knowledge-rich agents for complex virtual environments, including distributed training environments and interactive computer games. For development and testing in robotic virtual environments, Soar interfaces to a variety of robotic simulators and a simple mobile robot. We have recently made significant extensions to Soar that add new memories and new non-symbolic reasoning to Soar's original symbolic processing, which should significantly improve Soar abilities for control of robots. These extensions include episodic memory, semantic memory, reinforcement learning, and mental imagery. Episodic memory and semantic memory support the learning and recalling of prior events and situations as well as facts about the world. Reinforcement learning provides the ability of the system to tune its procedural knowledgeknowledge about how to do things. Mental imagery supports the use of diagrammatic and visual representations that are critical to support spatial reasoning. We speculate on the future of unmanned systems and the need for cognitive robotics to support dynamic instruction and taskability.
Soar, an architecture for problem solving and learning based on heuristic search and chunking, ha... more Soar, an architecture for problem solving and learning based on heuristic search and chunking, has been applied to a variety of tasks during the development of the Soar project, the goal of which is to build a system capable of general intelligent behavior. The hypothesis being tested in this aspect of Soar research is that chunking, a simple experience-basmd learning mechanism, can form the basis for a general learning mechanism. Previous work has demonstrated how the combination of chunking and Soar could acquire search-control knowledge (strategy acquisition) and operator implementation rules in both search-based puzzle tasks and knowledge-based expert systems tasks. This paper provides a new demonstration of the capabilities of chunking in the context of the macro-operator technique and shows how: (1) this technique can be used in a general, learning problem solver without the addition of new mechanisms; (2) the learning can be incremental during problem solving rather than requiring a preprocessing phase; (3) the macros can be used for any goal stated in the problem; and (4) additional generality can be obtained via transfer of learning between macro-operators if an appropriate representation of the task is available. References and a distribution list are provided. (KM)
Incorrect knowledge can be a problem for any in-telligent system. Soar is a proposal for the unde... more Incorrect knowledge can be a problem for any in-telligent system. Soar is a proposal for the under-lying architecture that supports intelligence. It has a single representation of long-term memory and a single learning mechanism called chunking. This paper investigates the ...
Chunks have long been proposed as a basic organizational unit for human memory. More recently chu... more Chunks have long been proposed as a basic organizational unit for human memory. More recently chunks have been used to model human learning on simple perceptual-motor skills. In this paper we describe recent progress in extending chunking to be a general learning mechanism by implementing it within a general problem solver. Using the Soar problem-solving architecture, we take significant steps toward a general problem solver that can learn about all aspects of its behavior. We demonstrate chunking in Soar on three tasks: the Eight Puzzle, Tic-Tat-Toe, and a part of the RI computer-configuration task. Not only is there improvement with practice, but chunking also produces significant transfer of learned behavior, and strategy acquisition.
Proceedings of the ... AAAI Conference on Artificial Intelligence, Aug 4, 2011
Effective access to knowledge within large declarative memory stores is one challenge in the deve... more Effective access to knowledge within large declarative memory stores is one challenge in the development and understanding of long-living, generally intelligent agents. We focus on a sub-component of this problem: given a large store of knowledge, how should an agent's task-independent memory mechanism respond to an ambiguous cue, one that pertains to multiple previously encoded memories. A large body of cognitive modeling work suggests that human memory retrievals are biased in part by the recency and frequency of past memory access. In this paper, we evaluate the functional benefit of a set of memory retrieval heuristics that incorporate these biases, in the context of the word sense disambiguation task, in which an agent must identify the most appropriate word meaning in response to an ambiguous linguistic cue. In addition, we develop methods to integrate these retrieval biases within a task-independent declarative memory system implemented in the Soar cognitive architecture and evaluate their effectiveness and efficiency in three commonly used semantic concordances.
This paper brings together work in modeling episodic memory and reinforcement learning. We demons... more This paper brings together work in modeling episodic memory and reinforcement learning. We demonstrate that is possible to learn to use episodic memory retrievals while simultaneously learning to act in an external environment. In a series of three experiments we investigate learning what to retrieve from episodic memory and when to retrieve it, learning how to use temporal episodic memory retrievals, and learning how to build cues that are the conjunctions of multiple features. Our empirical results demonstrate that it is computationally feasible to learn to use episodic memory in all three experiments, and furthermore, that learning to use internal episodic memory accomplishes tasks that reinforcement learning alone does not. These experiments also expose some important interactions that arise between reinforcement learning and episodic memory.
Episodic memory endows autonomous agents with useful cognitive capabilities. However, for long-li... more Episodic memory endows autonomous agents with useful cognitive capabilities. However, for long-lived agents, there are numerous unexplored computational challenges in supporting useful episodic-memory functions while maintaining real-time reactivity. This paper presents and summarizes the evaluation of an algorithmic variant to the task-independent episodic memory of Soar that expands the class of tasks and cues the mechanism can support while remaining reactive over long agent lifetimes.
Over time, the range and complexity of behaviors in computer-generated forces has expanded as the... more Over time, the range and complexity of behaviors in computer-generated forces has expanded as they move to include more autonomy and intelligence. To support the representation and execution of these behaviors, there is a tendency to add more and more features to the underlying software architecture. However, with the addition of these features, two potential costs may be incurred: increased execution time and additional memory requirements. As architectures progress, it is important to continually evaluate the cost and value of each new architectural feature. Seemingly very similar architectures may require significantly different resources; small changes to the features in a single architecture may have a large impact on its performance. Thus, it is necessary to understand and to quantify the resources consumed by different architectures and by the components of a single architecture. Unfortunately, there is no standard method for evaluating features of an architecture or for comparing sets of architectures. In this paper, we discuss a methodology for evaluating a specific architecture. We dissect the Soar architecture into a core set of functionality and examine how incrementally adding each of the features found in the original implementation affects the overall performance and resource requirements. Finally, we show how the same methodology can be used to compare two different architectures. We discuss initial results of a comparison that indicates both qualitative and quantitative differences between the Soar and CLIPS architectures.
Learning novel tasks is a complex cognitive activity requiring the learner to acquire diverse dec... more Learning novel tasks is a complex cognitive activity requiring the learner to acquire diverse declarative and procedural knowledge. Prior ACT-R models of acquiring task knowledge from instruction focused on learning procedural knowledge from declarative instructions encoded in semantic memory. In this paper, we identify the requirements for designing computational models that learn task knowledge from situated taskoriented interactions with an expert and then describe and evaluate a model of learning from situated interactive instruction that is implemented in the Soar cognitive architecture.
A rarely studied issue with using persistent computational models is whether the underlying compu... more A rarely studied issue with using persistent computational models is whether the underlying computational mechanisms scale as knowledge is accumulated through learning. In this paper we evaluate the declarative memories of Soar: working memory, semantic memory, and episodic memory, using a detailed simulation of a mobile robot running for one hour of real-time. Our results indicate that our implementation is sufficient for tasks of this length. Moreover our system executes orders of magnitudes faster than real-time, with relatively modest storage requirements. We also project the computational resources required for extended operations.
We are creating an environment in which to investigate the role of advanced AI in computer games.... more We are creating an environment in which to investigate the role of advanced AI in computer games. This environment is based on the Unreal Tournament (UT) game engine and the Soar AI engine. Unreal provides a 3D virtual environment, while Soar provides a flexible architecture for developing complex AI characters. This paper describes our progress to date, starting with our game, Haunt 2, which is designed so that complex AI characters will be critical to the success (or failure) of the game. We also describe the extensions we have made to UT to support AI characters with complex physiology so that the AI characters' behavior is driven by their interaction with their environment, their internal long-term goals, and any story-based goals. Finally, we describe the overall system design and interfaces between Soar and UT to support flexible development as well as efficient implementation.
The artificial intelligence (AI) components of computer games often appear to be very complex, po... more The artificial intelligence (AI) components of computer games often appear to be very complex, possibly having abilities beyond the state of the art in computer generated forces. In this paper we study the similarities and differences between AIs for computer games and computer generated forces (CGFs). We contrast the goals of AIs and CGFs, their behavioral requirements, and the underlying resources available for developing and fielding them, with an eye to how they impact the complexity of their behaviors. Our conclusion is that CGFs are currently far ahead of game AIs, but that this may change soon. We argue that computer games have advantages for doing certain types of research on complex, human-level behavior. We support this argument with a demonstration of research we have done on AI and computer games. We have developed the Soar Quakebot, which is a Soar program that plays the death match version of Quake II. The design of the Soar Quakebot is based on TacAir-Soar, a real-time expert system that flies U.S. military air missions in simulation, and that is used for training in the U.S. Air Force. The Soar Quakebot incorporates complex tactics and the ability of the bot to anticipate the actions of its enemy.
International Joint Conference on Artificial Intelligence, Aug 8, 1983
The weak methods occur pervasively in Al systems and may form the basic methods for all intellige... more The weak methods occur pervasively in Al systems and may form the basic methods for all intelligent systems. The purpose of this paper is to characterize the weak methods and to explain how and why they arise in intelligent systems. We propose an organization, called a universal weak method, that provides functionality of all the weak methods. A universal weak method is an organizational scheme for knowledge that produces the appropriate search behavior given the available task-domain knowledge. We present a problem solving architecture in which we realize a universal weak method. We also demonstrate the universal weak method with a variety of weak methods on a set of tasks. 1
Innovative Applications of Artificial Intelligence, Jul 25, 2004
We are creating an environment for investigating the role of advanced AI in interactive, story-ba... more We are creating an environment for investigating the role of advanced AI in interactive, story-based computer games. This environment is based on the Unreal Tournament (UT) game engine and the Soar AI engine. Unreal provides a 3D virtual environment, while Soar provides a flexible architecture for developing complex AI characters. This paper describes our progress to date, starting with our game, Haunt 2, which is designed so that complex AI characters will be critical to the success (or failure) of the game. It addresses design issues with constructing a plot for an interactive storytelling environment, creating synthetic characters for that environment, and using a story director agent to tell the story with those characters.
This is a detailed analysis and comparison of the ACT-R and Soar cognitive architectures, includi... more This is a detailed analysis and comparison of the ACT-R and Soar cognitive architectures, including their overall structure, their representations of agent data and metadata, and their associated processing. It focuses on working memory, procedural memory, and long-term declarative memory. I emphasize the commonalities, which are many, but also highlight the differences. I identify the processes and distinct classes of information used by these architectures, including agent data, metadata, and meta-process data, and explore the roles that metadata play in decision making, memory retrievals, and learning.
We propose a computational model of situated language comprehension based on the Indexical Hypoth... more We propose a computational model of situated language comprehension based on the Indexical Hypothesis that generates meaning representations by translating amodal linguistic symbols to modal representations of beliefs, knowledge, and experience external to the linguistic system. This Indexical Model incorporates multiple information sources, including perceptions, domain knowledge, and short-term and long-term experiences during comprehension. We show that exploiting diverse information sources can alleviate ambiguities that arise from contextual use of underspecific referring expressions and unexpressed argument alternations of verbs. The model is being used to support linguistic interactions in Rosie, an agent implemented in Soar that learns from instruction.
The ultimate goal of work in cognitive architecture is to provide the foundation for a system cap... more The ultimate goal of work in cognitive architecture is to provide the foundation for a system capable of general intelligent behavior. That is, the goal is to provide the underlying structure that would enable a system to perform the full range of cognitive tasks, employ the full range of problem-solving methods and representations appropriate for the tasks, and learn about all aspects of the tasks and its performance on them. In this article we present Soar, an implemented proposal for such an architecture. We describe its organizational principles, the system as currently implemented, and demonstrations of its capabilities.
Uploads
Papers by John Laird