Software Is Most Significant Loss in Firm’s Demise

Michael Schrage is a writer, consultant and research associate at the Massachusetts Institute of Technology. He writes this column independently for The Times

The impending bankruptcy of Thinking Machines Corp., the technically brilliant supercomputing company, recalls venture capital’s marvelously cynical cliche: “The earliest Christians got the best lions.” By that standard, the Cambridge, Mass.-based parallel-processing pioneer qualifies as a particularly tasty martyr.

Founded by an MIT prodigy, weaned on defense funding and championing a radically new architecture for high-performance computers, Thinking Machines was the first company to demonstrate the viability of massively parallel computing. Founder Danny Hillis’ essential idea was that a computer with thousands of processors could, like a swarm of hungry piranha, nibble a large problem to death far faster than a giant single processor.

Of course, getting tens of thousands of piranha processors to work in parallel is a tremendous feat of computer architecture. But Thinking Machines did it--and redefined fundamental ideas on how computers should be designed. As technology, it was revolutionary. As an entrepreneurial business, it was a flop. The bankruptcy lions got their snack.

It’s now popular in the Thinking Machines eulogies to comment on the irony of the technology succeeding even as the company failed. What’s more, there’s a consensus that Thinking Machines was an important company whose legacy lies in the parallel-processing paradigms it leaves behind.


But that may not be the case. As much of a breakthrough in hardware architecture that a massively parallel computer represents, the real legacy of Thinking Machines may rest in the software it created for its Connection Machines. A decade hence, people will look back at Thinking Machines and see the birth of a whole new set of approaches to advanced concepts of computer science such as genetic algorithms and artificial life.

Danny Hillis launched Thinking Machines saying he wanted to build a computer “that would be proud of me.” And for nearly 10 years he has been using his machines as a medium to model artificial life. He’s developed software that in turn grows other software through the process of Darwin-like natural selection.

With thousands of “piranha” processors and brilliant software, the piranhas can be given their own “genetic” characteristics and then set off to live with one another. Over time, perhaps, the traits of individual piranhas will change, or maybe the way piranhas act as a group will change. The computer software--indeed, the computer itself--can thus evolve.

It’s impossible to conceive of traditional supercomputers, even the most powerful of them, acting in this way. Until you had the massively parallel infrastructure of a Thinking Machines computer, you simply couldn’t conceive of supporting a large population of replicating software life forms.


“It’s an idea whose time is actually overdue,” Hillis said recently. “It’s scary for a lot of people, particularly in programming. . . . This whole technique goes against the notion that we design and build our software. It’s like we turn the keys of the car over to nature.”

Artificial life has become one of the hottest and most important research areas in all of computerdom. It’s also evolving into a business as Wall Street companies and traders try to evolve trading strategies by breeding them on the computers through genetic algorithms. The fittest software survives to trade another day; the rest are “selected out.” Other industries are using genetic algorithms to design product parts and grow new patterns of useful data.

Much of the so-called A-life community today either uses Connection Machines as the media for genetic software engineering or the software concepts originally tested by Hillis. In “Out of Control,” A-life commentator Kein Kelly describes the work of Thinking Machines’ Karl Sims in his efforts to breed new kinds of pictures and images:

“The mathematical logic of breeding pictures is indistinguishable from the mathematical logic of breeding pigeons,” Kelly wrote. “Conceptually, the two processes are equivalent. Although we may call it artificial evolution, there is nothing about it that is more or less artificial than breeding dachshunds. Both methods are equally artificial (of the art) and natural (true to nature).”


Will the legacy of Thinking Machines lie in new ways of designing computers it inspired--or in the new ways of creating artificial life? If we just look at an architecture without evaluating the way it influences creativity and behavior, we risk missing the point.

Consider Thomas Alva Edison, America’s greatest single inventor. People remember Edison for the invention of the light bulb. In reality, the Wizard of Menlo Park’s most significant breakthrough was the invention of the electric utility grid that lights up all those light bulbs and electrical devices. A light bulb without the infrastructure is just a technical novelty.

Similarly, the most obvious part of Thinking Machines is its thinking machines. But what we also have is the obituary of the company whose software will eventually create multibillion-dollar industries in artificial life.