mable computer can be constructed. Theoretical possibility was one thing, the practical demands of money and time another.

The demands were met in a rather roundabout manner through Hillis's interest in robots. From time to time he had mused openly about building a robot. Word of his idea somehow reached the ear of Harry Loucks, then director of the Mid-America Center in Hot Springs, Ark. Would the students like to construct a robot as a display in the center's museum? The students agreed in principle, but the project seemed too complicated. Just then the old Tinkertoy dream resurfaced. WouId the center like a computer made out of Tinkertoys instead?

Hillis and company set out to assemble the first Tinkertoy computer in a laboratory at M.I.T. The first model, unlike its successor, was a bulky cube with sides about one meter long. It was impressively complicated. Packed with logic devices made entirely of wooden sticks and spools, the machine signaled its moves by waving nine flags from the top of the framework. The prototype Tinkertoy computer had to be taken apart for the trip to Hot Springs, and once it was reassembled on site, the machine never quite worked properly again. On the other hand, it did make an intriguing exhibit. (It is currently on display at the Computer Museum in Boston.)

In 1979 Loucks contacted the group again. Could they make a new Tinkertoy computer, one that worked? By this time Silverman was in Ottawa and Hillis in Boston, each pursuing a new career. Over the telephone Hillis and Silverman worked out an improved design. It was to be reliable, and that meant simple. They decided to lay out all the possible tic-tac-toe boards in a row and devise some kind of reading mechanism that would move from row to row until it found a pattern matching the current board. The very act of recognition could trigger a preset response.

While Hillis contemplated ways to represent tic-tac-toe boards with digital Tinkertoy components, Silverman analyzed the game. To appreciate the complexities involved even in this childhood pastime, readers might consult the game tree shown on the opposite page. In the middle of the tree sits the initial board, a three-by-three grid empty of X's and O's. From this initial board nine new ones can arise, depending on which of the nine squares X plays. The figure shows just three possibilities; the remaining six are rotated versions. Each of the three