1986: First commercial AI supercomputer modelled after the human brain

The Connection Machine CM-1 (1986), and its faster, more performant successor the CM-2 (1987) were the first commercial supercomputers designed expressly for problems of Artificial Intelligence (AI). A massively parallel supercomputer with 65,536 processors, it was the brainchild of Danny Hillis, conceived in the early 1980s as his doctoral thesis under Marvin Minsky at the MIT Artificial Intelligence Lab, and built at his start-up Thinking Machines Corporation.

At the time, conventional wisdom held that massively parallel processing was difficult or even impossible. Serial supercomputers however, which executed one instruction at a time, had run up against an insurmountable barrier to faster performance: the speed of light, and thus the speed of signals in wires and traces. Danny surmised that if intelligence could arise out of the 86 billion very, very simple but richly interconnected neurons in the human brain, AI should be able to arise out of a huge number of richly connected but very simple 1-bit processors. The data for the problem to be solved was distributed over the tens of thousands of processors. All processors then simultaneously performed the same calculation, each on their own data, an architecture known as Single Instruction Multiple Data (SIMD).

What enabled the processors to communicate faster than previous SIMD designs was the internal network, a 12-dimensional boolean n-cube structure suggested by Nobel Prize physicist Richard Feynman, who spent his summers working with us. Within this hardwired physical structure, the software data structures for communication and transfer of data between processors could change as needed depending on the nature of the problem. The connections between processors were more important than the processors themselves, hence the name "Connection Machine."

In 1987 the CM-1 was superceded by the more powerful CM-2 in the same package and architecture, but with  additional floating-point hardware. In 1991 the CM-1/CM-2 hypercube design was superceded by the CM-5, with a very different form and architecture. Both the CM-2 in 1989 and CM-5 in 1993 won the prestigious Gordon Bell Prize as the fastest supercomputers in the world.  

Visual design of the Connection Machines CM-1/CM-2:

From 1983-1985 I led the packaging and industrial design team of the Connection Machine CM-1 at Thinking Machines Corporation, working with industrial design consultants Allen Hawthorne and Gordon Bruce, and mechanical engineering consultant Ted Bilodeau. (The CM-2, released in 1987, was a more advanced successor but used the same physical package.) 

The form of the machine was to express both its function and the passions of its creators: the dream of producing a "Machina Sapiens," a new genus of living, thinking machines. For a description of the design history, theory and concepts behind the visual design, see my article:

• Thiel, Tamiko. "The Design of the Connection Machine," (pdf or html) in DesignIssues, MIT Press, Cambridge, MA, Vol. 10, No. 1, Spring 1994. pp. 5-18.
  
  Republished in the book "The Designed World: Images, Objects, Environments." Editors: Richard Buchanan, Dennis Doordan and Victor Margolin, 2010.
  
  

How the Connection Machine inspired Steve Jobs:

My old friend Joanna Hoffman was marketing manager for the first Macintosh and the NeXT computer, and Steve Jobs' "right hand," as depicted by Kate Winslet in the film "Steve Jobs." In 1986, when the Connection Machine came out, Jobs asked her to get its designer to create his new NeXT computer as well. To which Joanna replied, "Too late, Tamiko has gone to Europe to become an artist!" But the machine made a lasting impression on him, as is obvious from the perfect black cube housing for the NeXT computer. From that point on, Jobs' designs were not merely useful - they were visually sublime.

See "CM-1/CM-2 Design Legacy" >>

Yale Professor David Gelernter describes the CM as sculpture:

"There have even been heroes of computer design: The CM-1 and CM-2 parallel supercomputers of the 1980s, designed by a team headed by Tamiko Thiel, were elegant and fascinating. At Yale, visitors stopped by the machine room often to admire the thing just as sculpture." Gelernter, David. "Beyond the Grey Box," ID Magazine, Vol. 45 #2, March/April 1998, page 60.

Connection Machines Legacy - an Untold Story in the History of AI

(See "Connection Machine Tech Legacy" interviews >>)

In the early/mid 1980s both AI and parallel processing were considered of interest only for ivory tower academics, not for real world computing and problem solving. Danny Hillis' radical new approach to computing, however, attracted scientists to Thinking Machines who saw the value and the promise of a computing paradigm that focused on the structures required by the data itself, rather than faster and faster serial processors.

Today's History of AI timelines omit the work that Danny and other scientists did at Thinking Machines and on the Connection Machines. Their contributions have been forgotten. But my interviews with several colleagues reveal how the ideas, the programming techniques, the software and the hardware shaped the technology of the world today. A few examples:

• Danny Hillis talks about how Google became the world's premiere search engine and AI company, first using parallel programming techniques that Sergey Brin learned on a Connection Machine, and then with the acquisition of Danny's subsequent company MetaWeb - whose CTO John Giannandrea then became Google's head of AI (he is now at Apple).

• Brewster Kahle talks about his term "Big Data" and how he used the Connection Machine to develop the first natural language search engine.

• Nobel physicist Richard Feynman used the Connection Machine to look at his speciality quantum chromodynamics, and also ran the first simulations of his seminal ideas for quantum computing on a Connection Machine.

• Even the hardware enabling Machine Learning today consists of chips that are in essence miniature Connection Machines, such as the GPUs built by NVIDIA.

Existing machines:

At its height, I have heard there were Connection Machines at 70 installations around the world. These included the models CM-1, CM-2, CM-2a, CM-200 and the successor CM-5 (in a completely different package design). The following machines have been preserved in museum collections. Machines are not always on exhibit - see notes.

• CM-1: Computer History Museum, Mountain View, Silicon Valley, CA. (Permanent exhibit)
• CM-1: Museum of American History, Smithsonian Institution, Washington DC.
• CM-2: Museum of Modern Art, New York. ("Thinking Machines" exhibit 2017-2018; "Energy" exhibit 2019-2020.)
• CM-2: Living Computer Museum, Seattle, WA.
• CM-2 and DataVault: Computer Museum of America, Roswell, Georgia, USA
• CM-2a: Musée Bolo, Lausanne, Switzerland (Permanent exhibit)
  ( The CM-2a is a single cube package by Mark Harrison.)