Book review of Ray Kurzweil

Ray Kurzweil:
"The Age of Spiritual Machines" (1999)
(Copyright © 2015 Piero Scaruffi | Terms of use )

Unlike his previous book "The Age of Intelligent Machines" (1990), which was a serious A.I. book, this new book is simply a self-congratulatory book, an exercise of relentless self-glorification with very little substance. He congratulates himself with having been right with most of his predictions (a trick that always works when you tweak the meaning of your predictions accordingly to what happened after you made the prediction, especially if the predictions were so obvious that just about everybody else would have made the same predictions except that most of us have don't think of predicting that tomorrow the Sun will rise or that next year we'll have to pay taxes). In actuality, none of the predictions listed in this book came true so far, except the ones that were already true when he wrote it.

Kurzweil argues that there exists a general law, the "Law of Accelerating Returns" that transcends Moore's Law. Order causes more order to be created and at a faster rate. Order started growing exponentially millions of years ago, and progress is now visible on a daily basis. This echoes science-fiction writer Vinge who wrote "The Acceleration of technological progress has been the central feature of this century" (1993). They both base their conclusions on the ever more frequent news of technological achievements. In my opinion they are confusing progress and the news cycle. Yes, we get a lot more news from a lot more sources. If the same news and communication tools had been available at any time in previous peacetime periods, the people alive back then would have been flooded by an equal amount of news.

His familiarity with contemporary A.I. (Hava Siegelmann's and Eduardo Sontag's Recurrent Neural Networks of 1992, Geoffrey Hinton's Helmholtz machine of 1995, Vladimir Vapnik's "Support-Vector Networks" of 1995, David Field & Bruno Olshausen's sparse coding of 1996, Sepp Hochreiter's and Jeurgen Schmidhuber's Long Short Term Memory (LSTM) model of 1997, Yann LeCun's second generation Convolutional Neural Networks of 1998) seems rather limited. In general, Kurzweil sounds out of touch with reality, unaware of all the technical details that need to be worked out for any of his "visions" to come true.

This book can be badly misleading about the history of technology, of AI and of philosophy. Read for what it really was: a provocative and enthusiastic reaction to IBM's Deep Blue beating the world chess champion in 1997.

Very outdated.