2017 Symposium

A mind is a functional entity that can think, and thus support intelligent behavior. Artificial intelligence, cognitive science, neuroscience, and robotics all contribute to our understanding of minds, although each draws from a different perspective. Artificial intelligence concerns building artificial minds, and thus cares most about how systems can be built that exhibit intelligent behavior. Cognitive science concerns modeling natural minds, and thus cares most about understanding cognitive processes that yield human thought. Neuroscience concerns the structure and function of brains, and thus cares most about how brains induce minds. Robotics concerns building and directing artificial bodies, and thus cares most about how minds control such bodies.

Will research across these disciplines ultimately converge on a single understanding of mind? This is a deep scientific question to which there is as yet no answer. However, there must at least be a single answer for cognitive science and neuroscience, as they both investigate the same mind, or narrow class of minds, albeit at different levels of abstraction. Research that is inspired by natural systems also may fit within this class of minds, particularly if it is slightly abstracted; but so may research that has no such aspiration yet still finds itself in the same neighborhood for functional reasons. This broader class comprises what can be called human-like minds.

The purpose of this symposium is to engage the international research community in developing a standard model of the mind, focused on human-like minds. The notion of a standard model has its roots in physics, where such a model has been developed for particles. This standard model is assumed to be internally consistent, yet still have major gaps. It serves as a cumulative reference point for the field while also driving efforts to both extend and break it.

As with physics, a standard model of the mind could provide a coherent baseline that facilitates shared cumulative progress, including a framework for organizing and sharing data for use in evaluating it and its alternatives. For integrative researchers, it could also help focus work on differences between particular approaches and the standard model, and on how to both extend and break the model. For theoretical and systems researchers, it could also provide guidance when they seek to expand to include aspects of other components. For experimental researchers, it could also provide top-down guidance in interpreting the results, as well as suggesting new experiments. For practitioners, it could also provide a sound basis for guiding development.

The intent, at least for the foreseeable future, is not to develop a single implementation or model of mind by which everyone concerned with human-like minds would abide, or even a theory in which all of the details are agreed to as correct. What is sought though is a statement of the best consensus given the community’s current understanding of the mind, plus a sound basis for further refinement as more is learned. Much of the existing work on integrative models of mind focuses on implementations rather than theory, with too little interchange or synthesis possible across these implementations. The development of a standard model provides an opportunity for the community to work together at a more abstract level, where such interchange and synthesis should be more practicable.

This symposium grew out of the 2013 AAAI Fall Symposium on Integrated Cognition, which was intended to bring together researchers across a set of disparate perspectives and communities concerned with an integrated view of human-level cognition. The organizing committee included representatives from cognitive science, cognitively and biologically inspired artificial intelligence, artificial general intelligence, and robotics. One of the summary presentations in the concluding session led to the startling finding that the wide range of researchers in the room at the time agreed that it was an appropriate consensus about the current state of the field. Given the field’s history of stark differences between competing approaches, this startled those in attendance. It implied that a consensus had implicitly begun to emerge – perhaps signaling the dawning maturity of the field – and that an attempt to make it explicit could provide significant value.

This was then followed up with an article that made an attempt to capture and extend the initial consensus articulated at the symposium (Laird, Lebiere, & Rosenbloom, 2017). Although this article hopefully provides a useful starting point, to truly create a standard model requires researchers from across the community being interested in relating their own approaches to it and participating in its evolution.  In the process, it is fully expected that they will disagree with some aspects of it, leading ideally to efforts to either disprove or improve parts of it. It is also expected that it will be incomplete in significant ways, not because those parts that are left out are unimportant, but because an adequate consensus on them has not yet been achieved.  Absence from the standard model is thus often a statement of where a consensus is needed, rather than a consensus on a lack of either existence or importance.

Laird, J. E., Lebiere, C. & Rosenbloom, P. S. (2017). A Standard Model of the Mind: Toward a Common Computational Framework across Artificial Intelligence, Cognitive Science, Neuroscience, and Robotics. AI Magazine38, 13-26. pdf