The connection machine is a supercomputer with thousands of interconnected computers. Its design allows scientists to at least partially emulate the processes in a human brain. Using parallel computing, the connection machine implements artificial intelligence. Some of these areas include face and other graphic recognition, applications for complex problem solving in various fields such as medicine and cryptology, and coding and decoding of sensitive documents.
In 1981, Danny Hillis wrote down the first description of the architecture of the connection machine. He was a student at Massachusetts Institute of Technology (MIT) who worked at MIT’s Artificial Intelligence Laboratory. During the late 1970s, human cognition research, which included a study of how humans think, made it necessary to seek processing capability beyond the so-called sequential computers. In 1983, Danny Hillis also helped found the Thinking Machines Corporation, where the connection machines CM-1, CM-2, and CM-5 were built in 1985, 1987, and 1993, respectively.
The connection machine makes use of fast parallel processors. When a connection machine is presented with an input, such as a picture of a face for recognition, it delegates the recognition task to a hierarchy of thousands of computers. This is analogous to a chief executive officer executing huge and complicated tasks by delegating to a small set of people who then each further delegates to a set of people, and so on. As a result, a huge task is done in relatively short time by “parallel-acting” individuals, similar to parallel processors in a connection machine.
Home computers are sequential computers with limited parallel-processing capabilities. For instance, graphic processors in home computers are parallel processors that prevent the slowing down of the main processor so it can interact with the user in real time. Full-pledged sequential computers execute one instruction at a time based on the programmer’s interpretation of the solution to a problem. The home computer is useful for relatively simple applications that do not require very complicated processing under time pressure, and not equipped to carry out connection machine functions.
There are various types of computers depending on processor speed, data word size, and architecture. Processor speed is usually in cycles per second wherein a processor is clocked or timed, and sometimes it is described as number of floating point instructions per second. Data word size is the number of bits that a processor is able to work on in one machine instruction, typically 32, 64, or 128 bits or more for bigger computers. Architecture is the way parts of processors and computers are interconnected together. There are simple parallel computers and massively parallel computers, such as the connection machine.
