The prediction that did not survive time

In 1965 the British mathematician Irving John Good, a former colleague of Alan Turing from the codebreaking team at Bletchley Park, wrote that it was more probable than not that a machine would be built before the end of the twentieth century that would far surpass all the intellectual activities of any human being. Almost thirty years later, in 1993, the mathematician and science fiction writer Vernor Vinge predicted in a lecture for NASA that within thirty years we would have the technical means to create superhuman intelligence, after which the human era would be over. Both deadlines have passed. Neither prediction came true as it was written. That is a fact, not an opinion. And it deserves more attention than it usually gets.

Why clever people get this wrong

The easy answer, "they were simply mistaken," misses the more interesting question. Good and Vinge were no fantasists; Good helped build the first generation of computers, Vinge taught mathematics at a university. Their mistake was not in the logic of their argument, if a system can improve itself, improvement accelerates itself, that still holds as reasoning. But in the leap from "this process exists" to "and therefore it happens around this year." A curve that has climbed steeply for twenty years feels irresistibly predictable. Whether it will keep on doing so, nobody knows in advance, and yet people write down years all the time as though they do.

A curve that did hold, and one that did not

To be fair: not every technology curve fails this way. The number of transistors on a chip doubled roughly every two years for decades, the rule of thumb known as Moore's law. And that extrapolation held up remarkably long, although it too is now slowing noticeably. The difference with Good and Vinge is that transistor density is a technical, measurable process with a clear manufacturing problem behind it. "When will a machine surpass humans in every conceivable field" is not a comparable measurable process, it is a bundling together of dozens of separate developments that advance at unequal speeds, and that is precisely why there is never a single curve to extend.

What this means for a builder

For anyone deciding today what to invest in, the lesson is not "ignore predictions." It is: build in such a way that you do not depend on the year in which someone else turns out to be right. Set aside money and attention for the scenario that arrives tomorrow and for the scenario that arrives in ten years, instead of staking all your choices on one deadline that comes from a lecture or a book. That is not caution born of fear. It is simply acknowledging that Good and Vinge, for all their knowledge, could not see the timeline. And neither can you, however convincing the curve may look at this moment.

"A thought to take with you: a good analysis and a hard date are two different claims, so treat them that way."