How accurately can artificially intelligent systems predict the future?
In a very meta sense, I am interested in predicting how far computational models for prediction can go. The geometric figure known as Gabriel’s horn refers to a three dimensional figure with infinite surface area but finite volume. Much in the same vein, predictive models in general may reach a convergence point. Despite the fact that time may continue endlessly, the accuracy of computational models may reach a maximum. This course itself follows the history of prediction, I am interested in its cloudy future.
There already exist certain models that seek to chart the course of predictive models, the earliest of which in modern society were concerned with the capabilities of computers. In a 1965 paper on computer science, Gordon E. Moore described arguably, the most important and accurate trend in computer performance. Only about 20 years after the inception of the first modern computer, Moore estimated that approximately every two years, the number of transistors in a dense integrated circuit doubles, and thus the performance of the computer will double. The correlation between transistors and time has so accurately followed this trend, that Gordon’s prediction has been coined as “Moore’s law”, after its certitude. Not surprisingly, the capabilities of artificial intelligence have followed a very similar trend of exponential growth, that shows no sign of slowing yet.
This system of predicting prediction is wholly observational. Moore, as well as other computer scientists who attempt to chart the course of developments in computing, looked at trends in computing to divine the future of the industry.
In addition to this observational system, other prominent figures in the industry have imparted their own predictions. A prime example of this has been Ray Kurzweil, who is somewhat considered an oracle within the field of artificial intelligence. His predictions seem to have no basis, but he has successfully predicted many of the computing innovations of the past 25 years with startling accuracy. In 1990 he successfully predicted that by 1998 a computer could defeat a world chess champion. In 1997, IBM’s Deep Blue defeated then-world-chess-champion Garry Kasparov. In 1990 he also made the prediction that by 2010, PCs would be capable of answering questions by accessing information via the Internet. Needless to say, he was correct.
Kurzweil’s predictions for computing for the next 25 years are eerie to say the least. Among them are the ability to upload a human consciousness to a machine, and non-biological intelligence becoming a billion times more intelligent than natural intelligence. While it is easy to look at his prior success and be fearful of an impending Terminator-esque future, his predictions for the future have been decried as being unoriginal and light on empirical backing.