I want to be fair to this book before criticising it, because the criticism is easy and there is something real underneath. The author has been making dated public predictions about computing for decades, in a field where almost everyone else hedges, and he has been directionally right about the thing that matters most, which is that computation would keep getting radically cheaper for far longer than the sceptics expected. The chapters presenting cost per unit of computation over a very long period, and showing that the trend survived multiple changes of underlying technology, are the strongest material here. The observation that the trend is a property of the whole enterprise rather than of any one manufacturing approach is a genuinely useful way to look at it, and a lot of people who dismiss this author have not engaged with that argument.
Committing to dates in public is also a real virtue and it is rarer than it should be. Most technology commentary is constructed so that it can never be wrong. This book says a thing will happen by a particular year. That is a service, and it makes the author checkable in a way that his critics mostly are not.
The material on brain interfaces, on biotechnology and on the medical applications is the most grounded part of the book, probably because it engages with specific technologies and specific obstacles rather than with curves. Now the problem, which is fundamental rather than incidental. The self assessment of the 2005 predictions is not credible. Predictions are reinterpreted generously, partial developments are scored as hits, and the standard applied to a claim after the fact is looser than the claim as originally written.
If you go back to what was actually said and read it plainly, the record is much more mixed than the accounting here suggests. An author who asks to be judged on his forecasts, and then marks his own homework this leniently, has undermined the very thing that made him worth reading. The methodological issue runs deeper. A trend line is not an explanation.
Showing that a quantity has grown exponentially for fifty years tells you that something has sustained it, and it does not tell you what, and therefore does not tell you whether it will continue. The book repeatedly treats the extrapolation as though it carried causal force, and moves from a curve about computation to conclusions about intelligence, consciousness and civilisation without doing the work of connecting them. The step from more operations per second to general intelligence is the entire question and it is assumed rather than argued. The social and economic chapters are the weakest.
They are essentially optimistic assertion. Abundance arrives, work is transformed benignly, the distributional problems resolve themselves. Anyone who has read seriously on the political economy of technological change will find nothing here that engages with why the benefits of previous transitions were distributed the way they were. The whole argument depends on these chapters being right and they are the least examined part of the book.
Repetition is a practical complaint. If you read the 2005 book, a large amount of this will feel familiar, and the update is less thorough than nineteen years of intervening evidence would justify. The rise of deep learning, which is the single most important thing to have happened in this field in the interval, is folded into the existing framework as confirmation rather than treated as something that might have required the framework to change. And there is a tonal problem that will lose readers.
The confidence is uniform. Difficult problems and easy ones are described with the same certainty. There is no visible sense of anything that might not work out, which for a book about the most consequential transition its author can imagine is a strange absence. My two point nine is for genuinely interesting long run data on computation, for the honesty of making dated public claims at all, and for value as a clear statement of a worldview that has real influence inside the industry even where it is not openly credited.
Marked down heavily for self assessment that does not survive checking, for mistaking extrapolation for mechanism, for social analysis that amounts to hoping, and for recycling a great deal of a previous book. Read it to understand a position that matters, and do not read it as an account of what is going to happen.