Lyle N. Long+ , Troy D. Kelley* , and Michael, J. Wenger§
+ Moore Distinguished Scholar, California Insitute of Technology and Distinguished Professor of Aerospace Engineering, Bioengineering, and Mathematics The Pennsylvania State University
* Engineering Psychologist U.S. Army Research Laboratory § Associate Professor of Psychology The Pennsylvania State University
§ Associate Professor of Psychology The Pennsylvania State University
While there have been discussions of computers reaching human leve ls of intelligence (, ), we should not over-simplify the issue of building intelligent or conscious machines. We should also not conflate intelligence and consciousness. Many computational problems may involve trade-offs of intelligence and consciousness. The largest current computer is the 212,992 processor IBM BlueGene (74 terabytes memory and 596 teraflops peak speed). It could probably simulate 10e13 synapses , while humans have about 10e15. There are many unknowns, however, related to wiring diagrams, software, hardware, algorithms, learning, sensory input, and motor-control output. A machine that combines intelligence and consciousness cannot just be an isolated computer. It will need to be a complex system of systems and be capable of learning and understanding real world situations. The key, however, is emergent behavior development through a variety of algorithmic techniques including: genetic algorithms, machine learning, cognitive architectures and connectionist methods. Humans will not be capabl e of completely specifying and programming the entire system; learning and emergent behavior  will be a stringent requirement for development. Conscious machines will need to be embedded in the real world with significant input/output capabilities and the ability to learn from people and experience. They will also need to be able to use context to modulate the expression of learning. The human sensory systems use hundreds of millions of cells, and there are roughly 600 muscles in the human body. The fascinating robotic vehicles in the DARPA Urban Challenge have very few sensor systems (e.g. lasers, cameras, and radar) and very few motor-control output channels. They also required complex software and teams of engineers. Cognitive architectures (e.g. Soar, SS-RICS, and ACT/R), have been implemented on mobile robots (, ), but these too are not very capable yet. Biological systems and computers can be compared in terms of in terms of memory and speed, but these are only two of the requirements for an intelligent machine. Evolution is basically an optimization program, and the human brain has been evolving for a t least 4 million years. Genetic algorithms and evolutionary techniques can be used to simulate human evolution; however, duplicating the conditions that led to the evolution of the human brain would be difficult, if not impossible. Symbolic A.I. will not lead to machines capable of duplicating human behavior. Connectionists and subsumptive architectures will not, by themselves, lead to the development of human-level intelligence nor the functional characteristics that define consciousness. Rule-based systems and cognitive architectures require humans to program the rules, and this process is not scalable to billions of rules (a.k.a. the Frame problem). The machines will need to rely on hybrid systems and emergent behavior; and they will need to be carefully taught and “mothered” by teams of engineers and scientists. In conclusion, human-level intelligence and consciousness might be possible as an emergent property of a massively parallel learning machine using a hybrid system of algorithms, architectures, and computational mechanisms.
1. Kurzweil, R., The Age of Spiritual Machines: When Computers Exceed Human Intelligence 2000: Penguin.
2. M oravec, H., Robot: Mere Machine to Transcendent Mind. 1998: Oxford University Press.
3. Long, L.N. and A. Gupta, Biologically-Inspired Spiking Neural Networks with Hebbian Learning for Vision Processing, in 46th AIAA Aerospace Sciences Meeting. 2008, AIAA: Reno, NV.
4. Koch, C., The Quest for Consciousness: A Neurobiological Approach. 2004: Roberts and Company.
5. H anford, S.D., O. Janrathitikarn, and L.N. Long. Control of a Six -Legged Mobile Robot Using the So ar Cognitive Architecture. in 46th AIAA Aerospace Sciences Meeting, AIAA Paper No. 2008-0878. 2008. Reno, NV.
6. A very, E., T.D. Kelley, and D. Davani, Using Cognitive Architectures to Improve Robot Control: Integrating Production Systems, Semantic Networks, and Sub-Symbolic Processing, in 15th Annual Con ference on Behavioral Repre sentation in Modeling an d Simulation (BRIMS). 2006: Baltimore, MD.
Viens no labākajiem rakstiem par mākslīgo inteliģenci. Minēts galvenais, kas vēl pagaidām nav atrisināts, pietrūkst mākslīgai inteliģencei – emergent behavior. I.V.