3 of 3 people found the following review helpful:
4.0 out of 5 stars
remarkably prescient, August 16, 2004
This review is from: The Tomorrow Makers: A Brave New World of Living-Brain Machines (Paperback)
Considering when it was written, the author of this book should be credit for foresight in taking this supposedly "fringe" stuff seriously. Now that we've got "Wired" into the debate, one might pooh-pooh that, but this was written before the fall of the Soviet Union, much less the rise of the WWW and the dot-com bubble.
I read it in high school, so I don't recall details, but I do remember that it was very entertaining.
Help other customers find the most helpful reviews
Was this review helpful to you? Yes
No
1 of 2 people found the following review helpful:
3.0 out of 5 stars
Technology survives through government funding or open market sales. Open market survival is always better., April 10, 2006
The Thinking Machine had 64,000 processors wired together in a network. Each of the units had a small amount of memory and required a small amount of power. The architecture allowed a great number of small calculations to be done simultaneously. Hillis stated about Thinking Machines, "Basically it is a step toward parallel processing. We know that the human brain is able to do amazing things with switching components that are basically slow-namely neurons. With a switching component that takes millisecond we are able to walk and talk and see pictures and all kinds of things that we can't get ordinary computers to do, even though ordinary computes are made out of components that are thousands of times faster." "The connection machine like the brain, do million of things at once." The connection machine could be rewrited in a millisecond just by writing a command to do so. In one moment it could be using its processors to process vision and the next moment it could be searching a tree analyzing AI logic computations and searches. "There will never be a day that a person turns on a computer and everybody agrees the computer thinks. What is going to happen is that as machines do more and more tasks only humans could do, it will become harder and harder to draw the lines between what's thinking and what's not thinking. And there will become a point that everyone agrees they are thinking". Hillis is not a genius; human equivalent thinking requires for AI research either too write better software; discovery of unknown laws of intelligence and principles of thinking; or more acquire more computation power. Hillis company would rise and fall and he would not be able to build a machine that capable of being proud of him.
Daniel Hillis and Marvin Minsky were determined to build a connection machine as a tool for developing software applications written in LISP too solve AI problems. They went to Sheryl Handler for help. Handler had participated in the startup of Genetics Institute. Handler introduced CBS founder William Paley, who agreed to invest $16 million in venture. Paley believed that Hillis could build a thinking machine. Hillis and Handler wanted to build a machine whose maximum impact would be as a research tool for scientist studying artificial intelligence. Hillis envision a sort of public intelligence utility where people would tap their home PCs and thereby bring AI into their world. Howard Res-nikov argued for a more flexible architecture that could support solving real-world problems. CM-1 was complete but few AI labs could afford the $5 million price tag. Res-nikov was right in his prediction. In 1986, CM-2 arrived and it was able to run Fortran and to do floating-point operations. The CM-2 still needed new special software which meant the programmers needed to learn new programming techniques. The CM-2 floating point operations did not effective take advantage of the 64,000 single-bit processors. Moderately parallel became popular. Moderately parallel uses cheap, off-the-self PCs and clustered them using network software into a single supercomputer that could work with existing software. The functional advantage of Moderately parallel seemed overwhelming and few customers outside the AI community had interest in Thinking machines massively parallel design. Databases became the killer application for data mining, however Hillis and Handler remaining bitter about having to target general scientific computing rather than AI. Oracle would eventual purchase rights to Hillis datamining technology and Sun would buy most of Thinking Machines hardware and massively parallel computing technology.
In 1990, Fujitsu Limit looked forward to bringing to market a 1,000 processor machine. The Japanese Fifth generation intelligent machines threatened to push Japan ahead in electronics. The Western world began shaking at the thought of Japan becoming the market leader in electronic design. Japan threatened to do to computers what it had done to automobiles and consumer electronics.
This would have been the perfect time for Thinking Machines to go public. Instead, Handler felt the company could get a successful teraflop machine on its own. CM-5 arrived but the standard chips it chose were not ready. DARPA purchase of 24 Connection machines totaling $55 million and KSR and IBM criticized the gift. DARPA funding dried up and Thinking machines had to sell machines on their merits on the open market. 1992, CM-5 was not selling and losses for the year totaled $17 million. Mid 1993 Thinking Machines declared bankruptcy.
Thinking Machines needed to solve general computing problems and reduce costs by using machines too assemble machines and machines to design their machines. Government subsidized projects reduced Thinking Machine survivability odds because it lacked stability. Government funding did not become sure money and Thinking Machines lost time too gain market acceptance and the market abandoned Thinking Machines when they saw the government withdraw. The reasoning for market abandonment was the perception Thinking Machines had lost 20% of its revenues and this was too big of a loss, for the company to survive; so, rather than purchasing superior hardware, they looked long-term for support and since the long-term probability seemed low, they did not invest. Thinking Machines would have done better by building a product sellable on the open market and not forming a dependency on any particular market sector. Thinking Machines emphasized superior hardware performance but failed to focus on create software tools, utilities, and languages that would entice developers to create code communities that advocated the thinking machine product. Thinking machine failed to realize that the AI experts would use the connection machine were not in the company and failed to use a DarpNet board too get feedback. Thinking Machines did prove that Hallis Phd thesis was defendable and that connection machine architecture could be designed and assembled creating some of the fastest computers in the world.
Help other customers find the most helpful reviews
Was this review helpful to you? Yes
No
