Machines Like Us

A new paper by Rodrick Wallace from the New York State Psychiatric Institute suggests that multiple-core chip technology may lend itself to implementations of recent mathematical models of the massively parallel, high level mental processes in animals. The author acknowledges that multi-core computers -- with the potential to have machines with thousands or tens of thousands of cores -- present a programming nightmare as the number of cores increases. He proposes a solution: adopt the parallel computation model that was optimized by millions of years of evolution. One consequence of doing so is that it may bring machine intelligence closer to reality.

From the paper's Abstract:

The irresistable march toward multiple-core chip technology presents currently intractable programming challenges. High level mental processes in many animals, and their analogs for social structures, appear similarly massively parallel, and recent mathematical models addressing them may be adaptable to the multi-core programming problem. Key words bandpass, cognition, consciousness, directed homotopy, global workspace, groupoid, institution, information theory, multitasking, random network, rate distortion, topology.

Recent developments in multiple-core chip technology create a powerful, hardware-driven, impetus toward highly parallel computing (e.g. Asanovic et al., 2006). Programming single core machines is something of a nightmare, and the challenges presented by hundreds, thousands, or tens of thousands, of cores suggest the necessity of draconian solutions. Asanovic et al. (2006) suggest that, "Since real world applications are naturally parallel and hardware is naturally parallel, what we need is a programming model, system software, and a supporting architecture that are naturally parallel."

There is a successful massively parallel -- computation -- model in nature, which has recently been formalized (Wallace et al., 2007), and which may be adaptable to the coming generation of highly parallel machines. It is the model of distributed cognition which applies in particular to humans, their cultural artifacts, and their institutions. For nearly a half-million years, hominids in small, well trained, well equipped and disciplined groups, have been the most efficient and fearsome predators on Earth. Humans, in large-scale organization, have recast the surface features and ecological dynamics of the entire planet. Human organizations, at all scales, are cognitive, taking the perspective of Atlan and Cohen (1998), in that they perceive patterns of threat or opportunity, compare those patterns with some internal, learned or inherited, picture of the world, and then choose one or a small number of responses from a vastly larger repertory of what is possible to them.

Read entire paper (pdf format).