Cognitive architecture

Last updated

A cognitive architecture refers to both a theory about the structure of the human mind and to a computational instantiation of such a theory used in the fields of artificial intelligence (AI) and computational cognitive science. [1] These formalized models can be used to further refine comprehensive theories of cognition and serve as the frameworks for useful artificial intelligence programs. Successful cognitive architectures include ACT-R (Adaptive Control of Thought – Rational) and SOAR. The research on cognitive architectures as software instantiation of cognitive theories was initiated by Allen Newell in 1990. [2]

Contents

The Institute for Creative Technologies defines a cognitive architecture as a "hypothesis about the fixed structures that provide a mind, whether in natural or artificial systems, and how they work together — in conjunction with knowledge and skills embodied within the architecture — to yield intelligent behavior in a diversity of complex environments." [3]

History

Herbert A. Simon, one of the founders of the field of artificial intelligence, stated that the 1960 thesis by his student Ed Feigenbaum, EPAM provided a possible "architecture for cognition" because it included some commitments for how more than one fundamental aspect of the human mind worked (in EPAM's case, [4] human memory and human learning).

John R. Anderson started research on human memory in the early 1970s and his 1973 thesis with Gordon H. Bower provided a theory of human associative memory. [5] He included more aspects of his research on long-term memory and thinking processes into this research and eventually designed a cognitive architecture he eventually called ACT. He and his students were influenced by Allen Newell's use of the term "cognitive architecture". Anderson's lab used the term to refer to the ACT theory as embodied in a collection of papers and designs. (There was not a complete implementation of ACT at the time.)

In 1983 John R. Anderson published the seminal work in this area, entitled The Architecture of Cognition. [6] One can distinguish between the theory of cognition and the implementation of the theory. The theory of cognition outlined the structure of the various parts of the mind and made commitments to the use of rules, associative networks, and other aspects. The cognitive architecture implements the theory on computers. The software used to implement the cognitive architectures was also called "cognitive architectures". Thus, a cognitive architecture can also refer to a blueprint for intelligent agents. It proposes (artificial) computational processes that act like certain cognitive systems. Most often, these processes are based on human cognition, but other intelligent systems may also be suitable. Cognitive architectures form a subset of general agent architectures. The term 'architecture' implies an approach that attempts to model not only behavior, but also structural properties of the modelled system.

Distinctions

Cognitive architectures can be symbolic, connectionist, or hybrid. [7] Some cognitive architectures or models are based on a set of generic rules, as, e.g., the Information Processing Language (e.g., Soar based on the unified theory of cognition, or similarly ACT-R). Many of these architectures are based on principle that cognition is computational (see computationalism). In contrast, subsymbolic processing specifies no such a priori assumptions, relying only on emergent properties of processing units (e.g., nodes [ clarification needed ]). Hybrid architectures such as CLARION combine both types of processing. A further distinction is whether the architecture is centralized, with a neural correlate of a processor at its core, or decentralized (distributed). Decentralization has become popular under the name of parallel distributed processing in mid-1980s and connectionism, a prime example being the neural network. A further design issue is additionally a decision between holistic and atomistic, or (more concretely) modular structure.

In traditional AI, intelligence is programmed in a top-down fashion. Although such a system may be designed to learn, the programmer ultimately must imbue it with their own intelligence. Biologically-inspired computing, on the other hand, takes a more bottom-up, decentralized approach; bio-inspired techniques often involve the method of specifying a set of simple generic rules or a set of simple nodes, from the interaction of which emerges the overall behavior. It is hoped to build up complexity until the end result is something markedly complex (see complex systems). However, it is also arguable that systems designed top-down on the basis of observations of what humans and other animals can do, rather than on observations of brain mechanisms, are also biologically inspired, though in a different way[ citation needed ].

Notable examples

Some well-known cognitive architectures, in alphabetical order:

NameDescription
4CAPS developed at Carnegie Mellon University by Marcel A. Just and Sashank Varma.
4D-RCS Reference Model Architecture developed by James Albus at NIST is a reference model architecture that provides a theoretical foundation for designing, engineering, integrating intelligent systems software for unmanned ground vehicles. [8]
ACT-R developed at Carnegie Mellon University under John R. Anderson.
Extended Artificial Memory developed at TU Kaiserslautern under Lars Ludwig. [9]
ASMO [10] developed by Rony Novianto, Mary-Anne Williams and Benjamin Johnston at the University of Technology Sydney. This cognitive architecture is based on the idea that actions/behaviours compete for an agents resources.
CHREST developed under Fernand Gobet at Brunel University and Peter C. Lane at the University of Hertfordshire.
CLARION the cognitive architecture, developed under Ron Sun at Rensselaer Polytechnic Institute and University of Missouri.
CMAC The Cerebellar Model Articulation Controller (CMAC) is a type of neural network based on a model of the mammalian cerebellum. It is a type of associative memory. [11] The CMAC was first proposed as a function modeler for robotic controllers by James Albus in 1975 and has been extensively used in reinforcement learning and also as for automated classification in the machine learning community.
Copycat by Douglas Hofstadter and Melanie Mitchell at the Indiana University.
DAYDREAMER developed by Erik Mueller at the University of California in Los Angeles under Michael G. Dyer
DUAL developed at the New Bulgarian University under Boicho Kokinov.
FORR developed by Susan L. Epstein at The City University of New York.
Framsticks a connectionist distributed neural architecture for simulated creatures or robots, where modules of neural networks composed of heterogenous neurons (including receptors and effectors) can be designed and evolved.
Google DeepMind The company has created a neural network that learns how to play video games in a similar fashion to humans [12] and a neural network that may be able to access an external memory like a conventional Turing machine, [13] resulting in a computer that appears to possibly mimic the short-term memory of the human brain. The underlying algorithm is based on a combination of Q-learning with multilayer recurrent neural network. [14] (Also see an overview by Jürgen Schmidhuber on earlier related work in deep learning. [15] [16] )
Holographic associative memory This architecture is part of the family of correlation-based associative memories, where information is mapped onto the phase orientation of complex numbers on a Riemann plane. It was inspired by holonomic brain model by Karl H. Pribram. Holographs have been shown to be effective for associative memory tasks, generalization, and pattern recognition with changeable attention.
Hierarchical temporal memory This architecture is an online machine learning model developed by Jeff Hawkins and Dileep George of Numenta, Inc. that models some of the structural and algorithmic properties of the neocortex. HTM is a biomimetic model based on the memory-prediction theory of brain function described by Jeff Hawkins in his book On Intelligence . HTM is a method for discovering and inferring the high-level causes of observed input patterns and sequences, thus building an increasingly complex model of the world.
CoJACK An ACT-R inspired extension to the JACK multi-agent system that adds a cognitive architecture to the agents for eliciting more realistic (human-like) behaviors in virtual environments.
IDA and LIDA implementing Global Workspace Theory, developed under Stan Franklin at the University of Memphis.
MANIC (Cognitive Architecture) Michael S. Gashler, University of Arkansas.
PRS 'Procedural Reasoning System', developed by Michael Georgeff and Amy Lansky at SRI International.
Psi-Theory developed under Dietrich Dörner at the Otto-Friedrich University in Bamberg, Germany.
Spaun (Semantic Pointer Architecture Unified Network) by Chris Eliasmith at the Centre for Theoretical Neuroscience at the University of Waterloo – Spaun is a network of 2,500,000 artificial spiking neurons, which uses groups of these neurons to complete cognitive tasks via flexibile coordination. Components of the model communicate using spiking neurons that implement neural representations called "semantic pointers" using various firing patterns. Semantic pointers can be understood as being elements of a compressed neural vector space. [17]
Soar developed under Allen Newell and John Laird at Carnegie Mellon University and the University of Michigan.
Society of Mind proposed by Marvin Minsky.
The Emotion Machine proposed by Marvin Minsky.
Sparse distributed memory was proposed by Pentti Kanerva at NASA Ames Research Center as a realizable architecture that could store large patterns and retrieve them based on partial matches with patterns representing current sensory inputs. [18]
Subsumption architectures developed e.g. by Rodney Brooks (though it could be argued whether they are cognitive).

See also

Related Research Articles

<span class="mw-page-title-main">Cognitive science</span> Interdisciplinary scientific study of cognitive processes

Cognitive science is the interdisciplinary, scientific study of the mind and its processes. It examines the nature, the tasks, and the functions of cognition. Mental faculties of concern to cognitive scientists include language, perception, memory, attention, reasoning, and emotion; to understand these faculties, cognitive scientists borrow from fields such as linguistics, psychology, artificial intelligence, philosophy, neuroscience, and anthropology. The typical analysis of cognitive science spans many levels of organization, from learning and decision to logic and planning; from neural circuitry to modular brain organization. One of the fundamental concepts of cognitive science is that "thinking can best be understood in terms of representational structures in the mind and computational procedures that operate on those structures."

<span class="mw-page-title-main">Cognitive neuroscience</span> Scientific field

Cognitive neuroscience is the scientific field that is concerned with the study of the biological processes and aspects that underlie cognition, with a specific focus on the neural connections in the brain which are involved in mental processes. It addresses the questions of how cognitive activities are affected or controlled by neural circuits in the brain. Cognitive neuroscience is a branch of both neuroscience and psychology, overlapping with disciplines such as behavioral neuroscience, cognitive psychology, physiological psychology and affective neuroscience. Cognitive neuroscience relies upon theories in cognitive science coupled with evidence from neurobiology, and computational modeling.

Artificial consciousness, also known as machine consciousness, synthetic consciousness, or digital consciousness, is the consciousness hypothesized to be possible in artificial intelligence. It is also the corresponding field of study, which draws insights from philosophy of mind, philosophy of artificial intelligence, cognitive science and neuroscience. The same terminology can be used with the term "sentience" instead of "consciousness" when specifically designating phenomenal consciousness.

<span class="mw-page-title-main">Connectionism</span> Cognitive science approach

Connectionism is the name of an approach to the study of human mental processes and cognition that utilizes mathematical models known as connectionist networks or artificial neural networks. Connectionism has had many 'waves' since its beginnings.

A cognitive model is a representation of one or more cognitive processes in humans or other animals for the purposes of comprehension and prediction. There are many types of cognitive models, and they can range from box-and-arrow diagrams to a set of equations to software programs that interact with the same tools that humans use to complete tasks. In terms of information processing, cognitive modeling is modeling of human perception, reasoning, memory and action.

Bio-inspired computing, short for biologically inspired computing, is a field of study which seeks to solve computer science problems using models of biology. It relates to connectionism, social behavior, and emergence. Within computer science, bio-inspired computing relates to artificial intelligence and machine learning. Bio-inspired computing is a major subset of natural computation.

<span class="mw-page-title-main">ACT-R</span>

ACT-R is a cognitive architecture mainly developed by John Robert Anderson and Christian Lebiere at Carnegie Mellon University. Like any cognitive architecture, ACT-R aims to define the basic and irreducible cognitive and perceptual operations that enable the human mind. In theory, each task that humans can perform should consist of a series of these discrete operations.

<span class="mw-page-title-main">Neural network (biology)</span> Structure in nervous systems

A neural network, also called a neuronal network, is an interconnected population of neurons. Biological neural networks are studied to understand the organization and functioning of nervous systems.

Unified Theories of Cognition is a 1990 book by Allen Newell. Newell argues for the need of a set of general assumptions for cognitive models that account for all of cognition: a unified theory of cognition, or cognitive architecture. The research started by Newell on unified theories of cognition represents a crucial element of divergence with respect to the vision of his long-term collaborator, and AI pioneer, Herbert Simon for what concerns the future of artificial intelligence research. Antonio Lieto recently drew attention to such a discrepancy, by pointing out that Herbert Simon decided to focus on the construction of single simulative programs that were considered a sufficient mean to enable the generalisation of “unifying” theories of cognition. Newell, on the other hand, didn’t consider the construction of single simulative microtheories a sufficient mean to enable the generalisation of “unifying” theories of cognition and, in fact, started the enterprise of studying and developing integrated and multi-tasking intelligence via cognitive architectures that would have led to the development of the Soar cognitive architecture.

Computational cognition is the study of the computational basis of learning and inference by mathematical modeling, computer simulation, and behavioral experiments. In psychology, it is an approach which develops computational models based on experimental results. It seeks to understand the basis behind the human method of processing of information. Early on computational cognitive scientists sought to bring back and create a scientific form of Brentano's psychology.

General game playing (GGP) is the design of artificial intelligence programs to be able to play more than one game successfully. For many games like chess, computers are programmed to play these games using a specially designed algorithm, which cannot be transferred to another context. For instance, a chess-playing computer program cannot play checkers. General game playing is considered as a necessary milestone on the way to artificial general intelligence.

CHREST is a symbolic cognitive architecture based on the concepts of limited attention, limited short-term memories, and chunking. The architecture takes into low-level aspects of cognition such as reference perception, long and short-term memory stores, and methodology of problem-solving and high-level aspects such as the use of strategies. Learning, which is essential in the architecture, is modelled as the development of a network of nodes (chunks) which are connected in various ways. This can be contrasted with Soar and ACT-R, two other cognitive architectures, which use productions for representing knowledge. CHREST has often been used to model learning using large corpora of stimuli representative of the domain, such as chess games for the simulation of chess expertise or child-directed speech for the simulation of children's development of language. In this respect, the simulations carried out with CHREST have a flavour closer to those carried out with connectionist models than with traditional symbolic models.

<span class="mw-page-title-main">CLARION (cognitive architecture)</span>

Connectionist Learning with Adaptive Rule Induction On-line (CLARION) is a computational cognitive architecture that has been used to simulate many domains and tasks in cognitive psychology and social psychology, as well as implementing intelligent systems in artificial intelligence applications. An important feature of CLARION is the distinction between implicit and explicit processes and focusing on capturing the interaction between these two types of processes. The system was created by the research group led by Ron Sun.

Ron Sun is a cognitive scientist who has made significant contributions to computational psychology and other areas of cognitive science and artificial intelligence. He is currently professor of cognitive sciences at Rensselaer Polytechnic Institute, and formerly the James C. Dowell Professor of Engineering and Professor of Computer Science at University of Missouri. He received his Ph.D. in 1992 from Brandeis University.

The LIDA cognitive architecture attempts to model a broad spectrum of cognition in biological systems, from low-level perception/action to high-level reasoning. Developed primarily by Stan Franklin and colleagues at the University of Memphis, the LIDA architecture is empirically grounded in cognitive science and cognitive neuroscience. It is an extension of IDA, which adds mechanisms for learning. In addition to providing hypotheses to guide further research, the architecture can support control structures for software agents and robots. Providing plausible explanations for many cognitive processes, the LIDA conceptual model is also intended as a tool with which to think about how minds work.

Google DeepMind Technologies Limited is a British-American artificial intelligence research laboratory which serves as a subsidiary of Google. Founded in the UK in 2010, it was acquired by Google in 2014 and merged with Google AI's Google Brain division to become Google DeepMind in April 2023. The company is based in London, with research centres in Canada, France, Germany, and the United States.

Newton Howard is a brain and cognitive scientist, the former founder and director of the MIT Mind Machine Project at the Massachusetts Institute of Technology (MIT). He is a professor of computational neurology and functional neurosurgery at Georgetown University. He was a professor of at the University of Oxford, where he directed the Oxford Computational Neuroscience Laboratory. He is also the director of MIT's Synthetic Intelligence Lab, the founder of the Center for Advanced Defense Studies and the chairman of the Brain Sciences Foundation. Professor Howard is also a senior fellow at the John Radcliffe Hospital at Oxford, a senior scientist at INSERM in Paris and a P.A.H. at the CHU Hospital in Martinique.

Deep reinforcement learning is a subfield of machine learning that combines reinforcement learning (RL) and deep learning. RL considers the problem of a computational agent learning to make decisions by trial and error. Deep RL incorporates deep learning into the solution, allowing agents to make decisions from unstructured input data without manual engineering of the state space. Deep RL algorithms are able to take in very large inputs and decide what actions to perform to optimize an objective. Deep reinforcement learning has been used for a diverse set of applications including but not limited to robotics, video games, natural language processing, computer vision, education, transportation, finance and healthcare.

Artificial intelligence and machine learning techniques are used in video games for a wide variety of applications such as non-player character (NPC) control and procedural content generation (PCG). Machine learning is a subset of artificial intelligence that uses historical data to build predictive and analytical models. This is in sharp contrast to traditional methods of artificial intelligence such as search trees and expert systems.

Timothy P. Lillicrap is a Canadian neuroscientist and AI researcher, adjunct professor at University College London, and staff research scientist at Google DeepMind, where he has been involved in the AlphaGo and AlphaZero projects mastering the games of Go, Chess and Shogi. His research focuses on machine learning and statistics for optimal control and decision making, as well as using these mathematical frameworks to understand how the brain learns. He has developed algorithms and approaches for exploiting deep neural networks in the context of reinforcement learning, and new recurrent memory architectures for one-shot learning.

References

  1. Lieto, Antonio (2021). Cognitive Design for Artificial Minds. London, UK: Routledge, Taylor & Francis. ISBN   9781138207929.
  2. Newell, Allen. 1990. Unified Theories of Cognition. Harvard University Press, Cambridge, Massachusetts.
  3. "Cognitive Architecture". Institute for Creative Technologies. 2024. Retrieved 11 February 2024.
  4. "The Feigenbaum Papers". Stanford University . Retrieved 11 February 2024.
  5. "This Week's Citation Classic: Anderson J R & Bower G H. Human associative memory. Washington," in: CC. Nr. 52 Dec 24–31, 1979.
  6. John R. Anderson. The Architecture of Cognition, 1983/2013.
  7. Vernon, David; Metta, Giorgio; Sandini, Giulio (April 2007). "A Survey of Artificial Cognitive Systems: Implications for the Autonomous Development of Mental Capabilities in Computational Agents". IEEE Transactions on Evolutionary Computation. 11 (2): 151–180. doi:10.1109/TEVC.2006.890274. S2CID   9709702.
  8. Douglas Whitney Gage (2004). Mobile robots XVII: 26–28 October 2004, Philadelphia, Pennsylvania, USA. Society of Photo-optical Instrumentation Engineers. page 35.
  9. Dr. Lars Ludwig (2013). Extended Artificial Memory. Toward an integral cognitive theory of memory and technology (pdf) (Thesis). Technical University of Kaiserslautern. Retrieved 2017-02-07.
  10. Novianto, Rony (2014). Flexible Attention-based Cognitive Architecture for Robots (PDF) (Thesis).
  11. Albus, James S. (August 1979). "Mechanisms of planning and problem solving in the brain". Mathematical Biosciences. 45 (3–4): 247–293. doi:10.1016/0025-5564(79)90063-4.
  12. Mnih, Volodymyr; Kavukcuoglu, Koray; Silver, David; Graves, Alex; Antonoglou, Ioannis; Wierstra, Daan; Riedmiller, Martin (2013). "Playing Atari with Deep Reinforcement Learning". arXiv: 1312.5602 [cs.LG].
  13. Mnih, Volodymyr; Kavukcuoglu, Koray; Silver, David; Graves, Alex; Antonoglou, Ioannis; Wierstra, Daan; Riedmiller, Martin (2014). "Neural Turing Machines". arXiv: 1410.5401 [cs.NE].
  14. Mnih, Volodymyr; Kavukcuoglu, Koray; Silver, David; Rusu, Andrei A.; Veness, Joel; Bellemare, Marc G.; Graves, Alex; Riedmiller, Martin; Fidjeland, Andreas K.; Ostrovski, Georg; Petersen, Stig; Beattie, Charles; Sadik, Amir; Antonoglou, Ioannis; King, Helen; Kumaran, Dharshan; Wierstra, Daan; Legg, Shane; Hassabis, Demis (25 February 2015). "Human-level control through deep reinforcement learning". Nature. 518 (7540): 529–533. Bibcode:2015Natur.518..529M. doi:10.1038/nature14236. PMID   25719670. S2CID   205242740.
  15. "DeepMind's Nature Paper and Earlier Related Work".
  16. Schmidhuber, Jürgen; Kavukcuoglu, Koray; Silver, David; Graves, Alex; Antonoglou, Ioannis; Wierstra, Daan; Riedmiller, Martin (2015). "Deep learning in neural networks: An overview". Neural Networks. 61: 85–117. arXiv: 1404.7828 . doi:10.1016/j.neunet.2014.09.003. PMID   25462637. S2CID   11715509.
  17. Eliasmith, C.; Stewart, T. C.; Choo, X.; Bekolay, T.; DeWolf, T.; Tang, Y.; Rasmussen, D. (29 November 2012). "A Large-Scale Model of the Functioning Brain". Science. 338 (6111): 1202–1205. Bibcode:2012Sci...338.1202E. doi:10.1126/science.1225266. PMID   23197532. S2CID   1673514.
  18. Denning, Peter J. "Sparse distributed memory." (1989).Url: https://ntrs.nasa.gov/archive/nasa/casi.ntrs.nasa.gov/19920002425.pdf