Articles A Standard Model of the Mind: Toward a Common Computational Framework Across Artificial Intelligence, Cognitive Science, Neuroscience, and Robotics John E. Laird, Christian Lebiere, Paul S. Rosenbloom n A standard model captures a com - mind is a functional entity that can think, and thus munity consensus over a coherent region support intelligent behavior. Humans possess minds, of science, serving as a cumulative ref - as do many other animals. In natural systems such as erence point for the field that can pro - A these, minds are implemented through brains, one particular vide guidance for both research and class of physical device. However, a key foundational hypoth - applications, while also focusing efforts to extend or revise it. Here we propose esis in artificial intelligence is that minds are computational developing such a model for humanlike entities of a special sort — that is, cognitive systems — that minds, computational entities whose can be implemented through a diversity of physical devices structures and processes are substan - (a concept lately reframed as substrate independence tially similar to those found in human [Bostrom 2003]), whether natural brains, traditional general- cognition. Our hypothesis is that cogni - purpose computers, or other sufficiently functional forms of tive architectures provide the appropri - hardware or wetware. ate computational abstraction for defin - ing a standard model, although the standard model is not itself such an architecture. The proposed standard model began as an initial consensus at the 2013 AAAI Fall Symposium on Integrated Cognition, but is extended here through a synthesis across three existing cognitive architectures: ACT-R, Sigma, and Soar. The resulting standard model spans key aspects of structure and processing, memory and content, learning, and perception and motor, and highlights loci of architectural agreement as well as disagreement with the consensus while identifying poten - tial areas of remaining incompleteness. The hope is that this work will provide an important step toward engaging the broader community in further develop - ment of the standard model of the mind. Copyright © 2017, Association for the Advancement of Artificial Intelligence. All rights reserved. ISSN 0738-4602 WINTER 2017 13 Articles Artificial intelligence, cognitive science, neuro - facilitates shared cumulative progress. For integrative science, and robotics all contribute to our under - researchers concerned with modeling entire minds, a standing of minds, although each draws from a dif - standard model can help focus work on differences ferent perspective in directing its research. Artificial between particular approaches and the standard intelligence concerns building artificial minds, and model, and on how to both extend and break the thus cares most for how systems can be built that model. Also, instead of each such researcher needing exhibit intelligent behavior. Cognitive science con - to describe all the assumptions and constraints of cerns modeling natural minds, and thus cares most their particular approach from scratch, given the for understanding cognitive processes that generate standard model they can simply state how their own human thought. Neuroscience concerns the struc - approach differs from it. Tables 1 and 2 in the sum - ture and function of brains, and thus cares most for mary of this article, for example, specify the standard how minds arise from brains. Robotics concerns model developed in this article and the standing of building and controlling artificial bodies, and thus three distinct approaches with respect to it. In this cares most for how minds control such bodies. process, the standard model itself could serve as Will research across these disciplines ultimately something of an interlingua or shared ontology, pro - converge on a single understanding of mind, or will viding a vehicle for mapping the common aspects, the result be a large but structured space of possibil - and possibly uncommon terminology, of disparate ities, or even a cacophony of approaches? This is a architectures onto a common base. deep scientific question to which there is as yet no For theoretical and systems researchers who mod - answer. However, there must at least be a single el/build specific components of mind — whether answer for cognitive science and neuroscience, as learning, memory, reasoning, language — a standard they are both investigating the same mind, or nar - model can provide guidance when they seek to row class of minds, albeit at different levels of expand to include aspects of other components. For abstraction. Biologically, or cognitively, or psycho - experimental researchers who tease out the details of logically inspired research in artificial intelligence how natural minds and brains work, a standard mod - and robotics also may fit within this particular class el can provide top-down guidance in interpreting the of minds, particularly if the class is slightly abstract - results, as well as suggesting new experiments that ed; but so may other work that has no aspiration to may be worth trying. For all researchers, a standard such inspiration yet still finds itself in the same model can serve as a framework around which data neighborhood for functional reasons. This broader that is used in evaluating single components or com - class comprises what can be called humanlike minds, binations of components may be organized and with an overall focus more on the bounded ration - made available for use by the community; potential - ality hypothesized to be central to human cognition ly growing to yield standard tests and testbeds. A (Simon 1957; Anderson 1990) than on the optimal - standard model can also provide a sound basis for ity that is the focus in much of artificial intelligence guiding practitioners in constructing a broad range and robotics. The class is broader than the more of intelligent applications. familiar one of naturally inspired minds, as it also The intent, at least for the foreseeable future, is not includes both natural minds and some artificial to develop a single implementation or model of minds that are not necessarily naturally inspired yet mind by which everyone concerned with humanlike functionally related. However, it is narrower in scope minds would abide, or even a theory in which all of than human-level intelligence, as it excludes minds the details are agreed to as correct. What is sought that are sufficiently inhuman in how they achieve though is a statement of the best consensus given the this level of intelligence. community’s current understanding of the mind, The purpose of this article is to begin the process of plus a sound basis for further refinement as more is engaging the international research community in learned. Much of the existing work on integrative developing what can be called a standard model of the models of mind focuses on implementations rather mind, where the mind we have in mind here is than theory, with too little interchange or synthesis humanlike. The notion of a standard model has its possible across these implementations. The develop - roots in physics, where for over more than a half-cen - ment of a standard model provides an opportunity tury the international community has developed and for the community to work together at a more tested a standard model that combines much of what abstract level, where such interchange and synthesis is known about particles. This model is assumed to should be more practicable. be internally consistent, yet still have major gaps. Its For this to transpire though will depend on function is to serve as a cumulative reference point researchers within the community being interested for the field while also driving efforts to both extend in relating their own approaches to the standard and break it. model and participating in its further evolution. In As with physics, developing a standard model of the process, it is fully expected that they will disagree the mind could accelerate work across the relevant with some aspects of the standard model presented disciplines by providing a coherent baseline that here, leading ideally to efforts to either disprove or 14 AI MAGAZINE Articles improve parts of it. It is also expected that the stan - been accomplished, including a précis of the pro - dard model will be incomplete in significant ways, posed standard model, an analysis of where the same not because those parts that are left out are unim - three cognitive architectures sit with respect to it, portant, but because an adequate consensus on them and a discussion of where we hope it will lead. has not yet been achieved. Omission from the stan - dard model is thus often a statement of where a con - Background sensus is needed, rather than a consensus on a lack of This attempt at a standard model of the mind, either existence or importance. although originating at the 2013 symposium, did not Although the boundary around the class of spring there from nothingness; and Allen Newell was humanlike minds is ill defined, at least at present, we at the root of much of what came before. One do anticipate an evolving dialogue around this, driv - notable precursor from three decades earlier is the en by a sequence of challenges from ideas and data model human processor (Card, Moran, and Newell that conflict in substantive ways with the standard 1983), which defines an abstract model of structural model. For each such challenge, it will be critical to and timing regularities in human perceptual, mental, determine whether the consensus is ultimately that and motor processes. It supports predicting approxi - the standard model should be altered — either mate timings of human behavior, but does not changed to eliminate the conflict or abstracted to include any details of the underlying computational cover both old and new approaches — or that the processes. new ideas or data should be deemed insufficiently A second, albeit rather different, precursor is humanlike, and thus outside of the class of interest.