25 February 1999

Nature


Nature 397, 663 - 664 (1999) Macmillan Publishers Ltd.


Exit Homo sapiens, stage left

JOHN L. CASTI

The Age of Spiritual Machines: When Computers Exceed Human Intelligence
by Ray Kurzweil
Viking: 1998. 376 pp. $25.95, 15.69

Robot: Mere Machine to Transcendent Mind
by Hans Moravec
Oxford University Press: 1998. 220 pp. $25, 15.12

CORBIS/BETTMANN


Man made machine:
super-intelligent robots 'will
be our heirs, sharing our goals
and values'.
According to evolutionary biologists and palaeontologists, about 99 per cent of all species that have ever inhabited the Earth are now extinct. If you believe the thesis of these two books, Homo sapiens' day in the sun as the leading intellectual force on this planet is also just about over. In fact, Kurzweil and Moravec both turn out the light for humans somewhere in the middle of the twenty-first century, at what Kurzweil would call the "dawning of the age of spiritual machines''.

Kurzweil, one of the great innovators in computing technology, predicts that computers exceeding the memory capacity and computational power of the human brain will be available within the next 20 years. In the scenario sketched out in The Age of Spiritual Machines, well before the year 2050, information will be fed directly into our brains via neural connections with machines, and the distinction between humans and machines will be totally blurred. At that point, computers will have acquired consciousness, human-style, and will embark on an evolutionary pathway that diverges from our own in almost every way that counts -- physically, emotionally and cognitively.

In the world of artificial intelligence, such behaviour would be termed 'strong AI'. Kurzweil's vision is the strongest of strong AI -- with a vengeance. Lest you fear that machines will then just keep humans around for laughs and/or as pets, Kurzweil predicts that humans will gradually co-evolve with these machines via neural implants that enable us to 'upload' our carbon-based neural circuitry into whatever kind of hardware machines are using in the late twenty-first century. At this point, there will be no clear distinction between humans and computers, and life expectancy will no longer be a term that pertains to intelligent beings. So humans won't exactly disappear; they will simply merge with the machines. Isn't that comforting?

The rationale underlying the arguments of both books is what Kurzweil has termed "The law of accelerating returns'', which applies specifically to evolutionary processes like the development of computer technology. Basically, this states that as order exponentially increases, time exponentially speeds up. In other words, the time interval between notable events gets shorter as time passes. The crucial implication is then that the valuable products of the evolutionary process also accelerate. Since computer technology is an evolutionary process that builds on its own progress, the time required to accomplish a fixed objective gets exponentially shorter as time goes on. So, for instance, it took 90 years for the first million instructions per second (MIP) to be achieved versus one day for an additional MIP now. If you buy into this law -- and all empirical evidence currently available supports it completely -- then the replacement of humans by machines as the primary intellectual force on Earth is indeed imminent.

A crucial part of the argument is that the computational process is evolutionary, so it will not crash into the same kind of barriers faced by other sorts of exponentially growing processes, such as Moore's law (which states that the surface area of an integrated circuit chip is halved every 12 months).

How will computing power continue to accelerate as Moore's law dies out? According to Kurzweil, it will use the third dimension of space in semiconductor design -- including circuits that don't generate heat. And as this technology tops out, others like DNA and quantum computing will step in to take its place. Thus, the entire evolutionary process of computing is literally unbounded. So goes Kurzweil's argument, anyway.

The Moravec volume draws essentially the same conclusions, emphasizing the emergence of robots embodying the greatly enhanced computing technology of the twenty-first century. But his vision of the "robotic takeover'' is not an apocalyptic one; rather, he takes the startling view that "Intelligent machines, which will grow from us, learn our skills, and share our goals and values, can be viewed as children of our minds''. In other words, these super-intelligent robots will be our heirs, and as such we will want them to outdistance us, much as parents want their biological heirs to have better, more productive lives than they have themselves.

Moravec then goes on to argue that these intelligent machines offer us lowly carbon-based forms the best chance we'll ever get for immortality by uploading ourselves into advanced robots -- the very same process that Kurzweil emphasizes. At this point, we will become our children and live for ever. This is either a science-fiction nightmare or a utopian fantasy: take your pick. In either case, the story Moravec weaves is fascinating.

While each of these books tells much the same tale, they do it in very different ways. Kurzweil's account focuses mostly on the intelligence of machines and the way humans will interface with them via direct neural connections to our brains. The writing is always lively and the arguments are well presented and amplified by numerous boxed cut-outs, tables and graphs showing the ever-accelerating pace of computing technology. The Moravec volume, on the other hand, spends a considerable number of pages on a detailed discussion of the evolution of robots, before moving on to some of the same types of speculations about machine intelligence as Kurzweil. The writing is equally lively, bolstered with diagrams and charts to support the arguments.

I recommend both books to any reader who wants a mind-expanding account of the rise of the age of intelligent machines. These two volumes are nothing less than a blueprint for how to shove Homo sapiens off centre-stage in evolution's endless play.

John L. Casti is at the Santa Fe Institute, 1399 Hyde Park Road, Santa Fe, New Mexico 87501, USA.


  Nature Macmillan Publishers Ltd 1999 Registered No. 785998 England.