Prof. de Garis creates the Planet's First Brain Building Center (BBC) (at USU)

During the fall semester of 2002, I invited M.Sc. and Ph.D. computer science students to collaborate with me on research into building artificial brains. The response was excellent, to the tune of some 30 students. So we now meet weekly on Tuesdays for lectures, brain storming sessions, handouts, etc. We felt that 30 was too large to be labeled a group, so we got ambitious and called ourselves a center.

The first research project we will undertake is called "e-LEM" which stands for "electronic - Learnable Evolution Model". My friend and colleague at George Mason University (GMU) in Virginia, Prof. Ryszard Michalski, devised a machine (concept) learning algorithm called LEM (Learnable Evolution Model) which replaces the blind operators of evolutionary computation with a concept learning approach (the high fitness chromosomes are treated by the concept learning algorithm as +ve examples, and the low fitness chromosomes are treated as -ve examples - the conept rule thus found is then used to generate the next generation of chromosomes). This LEM algorithm is hundreds of times faster than blind evolutionary algorithms. I spent a week with Ryszard over the summer of 2002 and had the idea to implement LEM directly in programmable hardware. (Xilinx recently gave the BBC $20k worth of programmable hardware (Virtex chip and board and SW)).

Since LEM speeds up EC by hundreds, so too do hardware speeds compared to software speeds, therefore it should be possible to perform a complete run of a genetic algorithm in about 1/100th of a second. This is incredibly fast compared to the half hour or so to evolve a neural network with a PC.

The BBC is submitting research grant proposals to the usual sources (Darpa, NSF, DoD (army, navy, airforce, etc) hoping to get a fat brain building grant comparable to the $0.4M given for this work by the Japanese when I was at ATR in Kyoto, and the $1M given by the Brussels Government before the bankruptcy of my previous lab Starlab.

We plan to design and build a second generation brain building machine we call BM2 with a billion artificial neurons. We find in practice that we get best evolvabilities when we have no more than 20 neurons per neural net using our latest neural net models, so that means 50 million neural nets or modules as we call them. It will be totally impractical to evolve individually 50 million modules, so somehow multi-module evolution will need to be automated.

We are hoping that e-LEM may help as a component in multi-module evolution. It will be so fast that we will have the luxury to play "what-if" experiments by changing individual modules within an algorithm (that the user does not know about).

The BBC is collaborating with other professors in other universities, e.g. Dean Prof. Dr. George Holling of Utah Valley State College, a professional electronic engineer, Prof. Dr. Garry Greenwood of Portland State College, also a professional electronic engineer, and Prof. Dr. Alan Shaw, of USU (the local hardware design guru, with a Stanford PhD). Prof. Vladimir Kulyukin (a "cognitive roboticist") of USU is also collaborating. So there is no lack of qualified personnel.

There are also commercial links. The Salt Lake City company Senapps, aims to get seed money to build an artificial brain building machine. If this effort is successful, then students trained within the BBC could get jobs with Senapps.

The greatest need now for the BBC is to get decent funding, preferably a million dollars or a least several hundred thousand to pay for the construction of the BM2. With a BM2 it will be possible to build the world's first artificial brain, a long term dream.