drupal statistics module

Machines Like Us

Machines Like Us interviews: Johnjoe McFadden

Monday, 26 November 2007

MLU: What are the consequences of your cemi field theory for artificial intelligence?

JM: One of the proposals of the cemi field theory is that the cemi field performs field computation in the brain (a process very similar to quantum computing) and this is the major advantage of consciousness that has been selected by natural selection. Computers currently lack this level of interaction and thereby lack the cemi field mediated general intelligence that is provided by field computing. I therefore predict that computers that compute only through wires will never acquire general intelligence and will never be aware.

However, there is nothing magical about cemi field awareness: it could be simulated by a computer with an architecture that allowed computations to take place through field interactions. Such computers would acquire natural intelligence and awareness.

MLU: Sounds like a marvellous post graduate student project. Do you know of any efforts to build a computer in this way?

JM: Bruce MacLennan at the The University of Tennessee has proposed that field-level information processing might be able to perform some computational manipulations, such as Fourier transforms and wavelet transforms, linear superpositions or Laplacians, more efficiently than digital computers. Efforts to design optical computers through -- for instance, the use of Vertical Cavity Surface Emitting Laser arrays (VCSEL) to interconnect circuit boards and thereby exploit field-level information transfer and processing -- is also ongoing.

An intriguing experiment performed by the School of Cognitive & Computing Sciences (COGS) group at Sussex University that appears to have (accidentally) evolved a field-sensitive electronic circuit. The group used a silicon chip known as a field-programmable gate array (FPGA), comprised of an array of cells. Electronic switches distributed through the array allow the behaviour and connections of the cells to be reconfigured from software. Starting from a population of random configurations, the hardware was evolved to perform a task, in this case, distinguishing between two tones. After about 5,000 generations the network could efficiently perform its task. When the group examined the evolved network they discovered that it utilized only 32 of the 100 FPGA cells. The remaining cells could be disconnected from the network without affecting performance. However, when the circuit diagram of the critical network was examined it was found that some of the essential cells, although apparently necessary for network performance (if disconnected, the network failed), were not connected by wires to the rest of the circuit. According to the researchers, the most likely explanation seems to be that these cells were contributing to the network through electromagnetic coupling -- field effects -- between components in the circuit. It is very intriguing that evolution of an artificial neural network appeared to capture field effects spontaneously as a way of optimizing computational performance. This suggests that natural evolution of neural networks in the brain would similarly capture field effects, precisely as proposed in the cemi field theory. The finding may have considerable implications for the design of artificial intelligence.