The Singularity is Near: When Humans Transcend Biology, by Ray Kurzweil

I should start by saying that I tried with this book. I really did. I tried 231/487 pages of text and 66/100 pages of notes worth. But I couldn't finish it. Normally I would have written off a book that I disliked much earlier, but I persevered. I was actually pre-disposed to like it: not only had Kurzweil referenced one of my articles in an earlier book (which I actually never read, but was flattered by), but this tome came highly recommended by a friend of mine. Joe said he liked it because it allowed him to stretch his imagination. He found it fun to read the way he finds science fiction fun to read. I found the book unbearable for more-or-less the same reason.

I don't see the point in spending too long reviewing what, for me, was not a useful read. Also, I can't claim to be authoritative, because the book may have got better after I gave up. But here's why I personally wouldn't recommend it to anyone.

The beginning of the book was about the exponential growth of technology, and particularly computing power. This didn't bother me too much, though I did have some grumbles. Funnily enough, I saw MIT's Rod Brooks give a talk about robotics riding on the coat-tails of electronics at around the same time I finished this section, which helped make me willing to read on. However, the supposed end-point of Kurzweil's argument (transhumanism through downloading people into machines) already did not seem credible to me (or, for that matter, terribly interesting). Little dialogues between Molly in 2004 (the year he wrote the book), her 2104 self, and other characters were patronizing, charmless, and a very poor homage to those in Douglas Hofstadter's much better book Gödel, Escher, Bach: An Eternal Golden Braid.

However, the material that bothered me most was in Chapter 3: on achieving the computational capacity of the brain in a machine. This subject is my passion: it's the thing that pulled me away from the optics and optoelectronics into neuromorphic engineering, robotics, and artificial intelligence. I've read a lot about this, thought a lot about it, heard arguments from every side, and my PhD thesis was related to the subject. From that perspective, I don't believe he knows his stuff: enough to be an interesting conversationalist, yes; but to write an authoritative book chapter on the subject, absolutely not.

Just a few specific things that make me question his knowledge and judgement. First, I think he really misunderstands nanotechnology (for what it is, not what Eric Drexler said it would be). If anyone's interested I can expand on this. Second, I think Kurzweil deeply misunderstands the importance of the dense interconnectivity of the brain (which I mentioned in an earlier post) and the difficulty of achieving this in electronics. Finally, the following sentence shows a staggering level of misunderstanding about the nature of analog and digital computing:

"We should keep in mind, as well, that digital computing can be functionally equivalent to analog computing—that is, we can perform all the functions of a hybrid digital-analog network with an all-digital computer. The reverse is not true: we can't simulate all of the functions of a digital computer with an analog one."

Given that digital computers are implemented using analog components and given that we know there are functions that Turing machines cannot perform that analog machines can, this seems a breathtakingly ignorant statement. It is the exact opposite of the truth. And written without the slightest hint that it might be controversial.

Rather than go on in this vein, let me say again that I did go through the notes for all the chapters I read, not just the text. That's one of the other things that lost him credibility. He actually cites Wikipedia as if it were an authoritative source. (I love it, and it's great for quick reference, but not if you need to really know about something). He uses press releases (from Eureka Alert and others) and journalist-written articles as if they were primary sources. He uses genuine technical papers too, but not enough. At times, it feels like he's got some intern to go on the web to get random references to justify what he's said. With about 100 pages of notes, he's clearly trying to impress us with his research: yet, next to statements as strong (and wrong) as the one I quoted above, there are no notes at all.

It's not just that I disagree with him. Given the number of back-of-the-envelope calculations that Kurzweil relies on, you have to trust that he really understands the issues deeply enough to be competent to rise above the details and abstract out what's important and what's not.  By the end of the third chapter my trust was completely gone.

This was close to being the last straw for me, but my friend convinced me to keep going. I finally put the book down (or actually, found myself reading the pointless free newspapers in the Tube night after night rather than subject myself to more of Singularlity) when we started getting into stuff about living for ever and he told us about the hundreds of food supplements (was it 250?) and daily injections that he was taking. Enough.

My copy has a quote on the cover by New York Times' Janet Maslin: "Startling in scope and bravado". I think this is pretty accurate. It's both startling that Kurzweil is willing to claim that he really understands such a broad and diverse range of subjects (when he doesn't), and that he expected people to believe this claim. Maybe some do, but I'm not one of them.

 

Addendum: 12 September 2007

Above I said that I felt the author's ideas about nanotechnology were unconvincing. I've now written elaborating on why. If you're interested, please check out my post in the main blog.

Comments

[Scarybug]

"If anyone's interested I can expand on this."

Please do!

My father-in-law gave me this book, but I haven't read it yet. I'm very interested in certain aspects of transhumanism like networking people with wearable or implanted computers, sensory augmentation, etc, but I find a lot of what people focus on is immortality and uploading your mind, which I'm not convinced is anywhere near happening.

[Todd T]

Refreshing review. Thanks.

[Barak A. Pearlmutter]

> ... and given that we know there are functions
> that Turing machines cannot perform that analog
> machines can [Siegelman 1995] ...

This is not true in any meaningful sense. The machines proposed by Siegelman can perform super-Turing computation only if they are pre-loaded with uncomputable numbers, eg in the weights of the network. Similarly, if the numbers they manipulate are limited in precision, say to only 500 digits of accuracy, then they are no longer super-Turing.

The work by Hava Siegelman was of importance for some of the mathematical insights it led to, but not for actually describing a realizable super-Turing machine.

[Sunny Bains]

I didn't go into this in any great detail here, but the super-Turing nature of analog computation is definitely true in a meaningful sense. Please read my PhD thesis on the subject (Physical computation and embodied artifical intelligence) and I'd be happy to debate it with you from there... You can download it at:

http://www.sunnybains.com/index.php?page=publications.htm