Artificial Brains and Natural Intelligence

Larry Cauller and Andy Penz, University of Texas at Dallas

It is widely accepted that nanotechnology will help push Moore's Law to, or past, its prediction that the next few decades will witness a truly amazing advance in affordable personal computing power. Several visionary techno-futurists have attempted to estimate the equivalent power of the human brain to predict when our handheld personal computers may be able to convince us that they occasionally feel, well, unappreciated, at least. With the advent of nano-neuro-techniques, neuroscience is also about to gain unfathomable insight into the dynamical mechanisms of higher brain functions. But many neuroscientists who have dared to map the future path to an artificial brain with human intelligence do not see this problem in simple terms of "computing power" or calculations per second. We agree that the near future of nano-neuro-technology will open paths to the development of artificial brains with natural intelligence. But we see this future more in terms of a coming nano-neuro-cogno-symbiosis that will enhance human potential in two fundamental ways: (1) by creating brilliant, autonomous artificial partners to join us in our struggle to improve our world; and (2) by opening direct channels of natural communication between human and artificial nervous systems for the seamless fusion of technology and mind.

Human brain function emerges from a complex network of many billion cooperating neurons whose activity is generated by nanoscale circuit elements. In other words, the brain is a massively parallel nanocomputer. And, for the first time, nanotechnology reveals approaches toward the design and construction of computational systems based more precisely upon the natural principles of nervous systems. These natural principles include: (1) enormous numbers of elementary nonlinear computational components; (2) extensive and interwoven networks of modifiable connectivity patterns; (3) neurointeractive sensory/motor behavior; and (4) a long period of nurtured development (real or virtual). We believe human-like functions will likewise emerge from artificial brains based upon these natural principles.

A simple nanoelectronic component, the resonant tunneling diode, possesses nonlinear characteristics similar to the channel proteins that are responsible for much of our neurons' complex behavior. In many ways, nanoscale electronics may be more suitable for the design of nonlinear neural networks than as simple switching elements in digital circuits. At this NBIC meeting, Phil Kuekes from Hewlett-Packard described a nanoscale cross-link connection scheme that may provide an approach to solving the truly difficult problem of how to interconnect enormous networks of these nanocomponents. But as a beginning, these initial steps to realization of a nano-neuro-computer permit a consideration of the much greater density that is possible using nanoelectronic neurons than has so far been possible with microelectronic solutions, where equivalent chip architectures would need to be millions of times larger. If the size of the artificial brain were small enough to mount on a human-size organism, then it may be simpler to design nurturing environments to promote the emergence of human-like higher functions.

Decades of neuroscience progress have shed a great deal of light upon the complexity of our brain's functional neuro-architecture (e.g., Felleman and Van Essen 1991). Despite its extreme complexity (>100,000 miles of neuron fibers), fundamental principles of organization have been established that permit a comprehensive, although highly simplified sketch of the structure responsible for natural intelligence. In addition, neuroscience has characterized many of the principles by which the network's connections are constantly changing and self-organizing throughout a lifetime of experience (e.g., Abbott and Nelson 2001). While some futurists have included the possibility that it will be possible to exactly replicate the cellular structure of the human brain (Kurzweil 1999), it seems impossible from a neuroscience point of view, even with nanotechnology. But it is not necessary to be too precise. Genetics is not that precise. We know many of the principles of neuro-competition and plasticity that are the basis for the continuous refinement of neural functions in the midst of precise wiring and environmental complexity. But the only test of these far-reaching principles is to construct a working model and learn to use it.

Constrained by the limits of microtechnology, previous attempts to mimic human brain functions have dealt with the brain's extreme complexity using mathematical simplifications (i.e., neural networks) or by careful analysis of intelligent behavior (i.e., artificial intelligence). By opening doors to the design and construction of realistic brain-scale architectures, nanotechnology is allowing us to rethink approaches to human-like brain function without eliminating the very complexity that makes it possible in the first place. The tools of nonlinear dynamical mechanics provide the most suitable framework to describe and manage this extreme complexity (e.g., Kelso 1995; Freeman 2000). But the first step is to recognize and accept the natural reality that the collective dynamics of the neural process responsible for the highest human functions are not mathematically tractable.

Instead, higher functions of the brain are emergent properties of its neuro-interactivity between neurons, between collections of neurons, and between the brain and the environment. While purely deterministic, it is no more possible to track the cause-effect path from neuron activity to higher functions such as language and discovery than it is to track the path from an H2O molecule to the curl of a beach wave. Unfortunately, appeals to emergence always leave an unsatisfying gap in any attempt to provide a complete explanation, but nature is full of examples, and classical descriptions of human intelligence have depended strongly upon the concept of emergence (i.e., Jean Piaget, see Elman et al. 1997). But modern emergent doctrine is gaining legitimacy from the powerful new tools of nonlinear dynamical mathematics for the analysis of fractals and deterministic chaos. Instead of tracking cause-effect sequence, the new paradigm helps to identify dynamical mechanisms responsible for the phase shifts from water to ice, or from exploring to understanding.

From the perspective of neuro-interactive emergence, brain function is entirely self-organized so it may only be interpreted with respect to the interactive behavior of the organism within meaningful contexts. For instance, speech communication develops by first listening to one's own speech sounds, learning to predict the sensory consequence of vocalization, and then extending those predictions to include the response of other speakers to one's own speech. This natural process of self-growth is radically different from the approaches taken by artificial intelligence and "neural net" technologies. The kernel of this natural process is a proactive hypothesis-testing cycle spanning the scales of the nervous system that acts first and learns to predict the resulting consequences of each action within its context (Cauller, in press; see also Edelman and Tonomi 2001). Higher functions of children emerge as a result of mentored development within nurturing environments. And emergence of higher functions in artificial brains will probably require the same kinds of care and nurturing infrastructure we must give our children.

So the future of the most extreme forms of machine intelligence from this neuroscience perspective differs in many respects from popular visions: (1) "artificial people" will be very human-like given their natural intelligence will develop within the human environment over a long course of close relationships with humans; (2) artificial people will not be like computers any more than humans are. In other words, they will not be programmable or especially good at computing; (3) artificial people will need social systems to develop their ethics and aesthetics.

An optimal solution to the problem of creating a seamless fusion of brain and machine also needs to be based upon these neurointeractive principles. Again, nanotechnology, such as minimally invasive nano-neuro transceivers, is providing potential solutions to bridge the communication gap between brain and machine. But the nature of that communication should be based upon the same neural fundamentals that would go into the design of an artificial brain.

For instance, sensory systems cannot be enhanced by simply mapping inputs into the brain (e.g., stimulating the visual cortex with outputs from an infrared camera won't work). The system must be fused with the reciprocating neurointeractivity that is responsible for ongoing conscious awareness. This means that brain control over the sensory input device is essential for the system to interpret the input in the form of natural awareness (e.g., there must be direct brain control over the position of the video source). In other words, brain enhancements will involve the externalization of the neurointeractive process into peripheral systems that will respond directly to brain signals. These systems will become an extension of the human mind/body over a course of accommodation that resembles the struggle of physical therapy following cerebral stroke.

Fusion of artificial brains into larger brains that share experience is a direct extension of this line of reasoning. This also would not be an immediate effect of interconnection, and the fusion would involve give and take on both sides of the connection over an extended course of active accommodation. But the result should surpass the sum of its parts with respect to its ability to cope with increasing environmental complexity.

Speculation leads to the next level of interconnection, between human and artificial brains. On the face of it, this appears to be a potential path to cognitive enhancement. However, the give and take that makes neurointeractive processes work may be too risky when humans are asked to participate.

Figure C.15. Neurointeractive artificial brain/human brain interface for neuroprosthesis or enhancement. References

Abbott, L.F., and Nelson SB. 2000. Synaptic plasticity: taming the beast. NatNeurosci 3:1178-83

Cauller, L.J. (in press). The neurointeractive paradigm: dynamical mechanics and the emergence of higher cortical function. In: Theories of Cerebral Cortex, Hecht-Neilsen R and McKenna T (eds).

Edelman G.M. and G. Tonomi. 2001. A Universe of Consciousness: How Matter Becomes Imagination, Basic Books.

Elman, J.L., D. Parisi, E.A. Bates, M.H. Johnson, A. Karmiloff-Smith. 1997. Rethinking Innateness: A Connectionist Perspective on Development, MIT Press, Boston.

Felleman, D.J., and Van Essen D.C. 1991. Distributed hierarchical processing in the primate cerebral cortex. Cereb Cortex 1(1):1-47.

Freeman, W.J. 2000. Neurodynamics: An Exploration in Mesoscopic Brain Dynamics (Perspectives in Neural Computing). Springer Verlag.

Kelso, S. 1995. Dynamic Patterns (Complex Adaptive Systems). MIT Press, Boston, MA.

Kurzweil, R. 1999. The Age of Spiritual Machines. Viking Press, New York, NY.

0 0

Post a comment