In this third part of my ongoing review of “The Singularity is Near” by Ray Kurtzweil (here are parts one and two) I will be focusing on his core thesis and my main substantive objection: the pace of technological progress over the next 30 years. Kurtzweil believes, based on an expectation of general exponential growth of all information-based technologies, that by the time we’re a third of the way through this century we will have reached a rate of change that is incomprehensible to us now. Major new categories of technology will have initial laboratory demonstrations in January, product rollouts in February, and worldwide adoption by March. Because of this ongoing rate of change, the species will have undergone a complete transformation. We will live among, and many of us will be, virtual and/or fully synthetic humans.
Count me among the skeptics. I think that this theory of an exponentially increasing rate of change is untenable for many, many reasons. Let’s start with the simplest. Kurtzweil projects that the time it takes for new technology to be adopted is halving every decade. So while it took ten years for worldwide penetration of a new technology in the decade of the 2000’s (the “double-ohs” as I like to call them), it will take only 5 years to develop new technology in the 2010’s, two and a half years in the 2020’s, a little over a year in the 2030’s, seven months in the 2040’s, and so on. This would be an exciting trend, if true. However, if we project backward we should conclude that it took 20 years for new technologies to be adopted in the 1990’s, 40 years in the 1980’s, and so on. So apparently in the 1820’s it should take well over TWO MILLION years for any single technology to get from invention to adoption. How long was it for the steam engine?
So we have developed our first law of debunking the Law of Accelerating Returns: project any overly optimistic expectation of the future backward, and see if it corresponds with reality. If not, it’s wrong.
This simplistic projection also fails to take two other issues into account. The first is that it takes a certain minimal time to develop an idea into a useful product. Even if adoption was instant, the time to convert a basic concept into a tested, marketable invention has not been shortened significantly. I have been in the software business for 20 years, and I have seen management fads come and go. If there had been any advancement in the field I would been right at the front of the line, applying everything I could to my own products. The simple fact is that there is no “silver bullet.” In the last decade no one has managed to significantly shorten the product cycle in a repeatable manner. Perhaps there are a few people doing it but I have never met them, and more importantly I have never had to compete against them.
The second and more serious problem is that it assumes a regular schedule of major scientific and engineering breakthroughs. Kurtzweil assumes that to fit his exponential projections researchers and technologists will oblige him with new theories and new creations that suit his schedule. How, you might wonder, will we manage to see such rapid development and worldwide adoption of new technologies in a few tens of years? Kurtzweil’s answer is illuminating. By that time, he foresees, we will all be mentally enhanced so that subjective time is magnified. In a few months of real time our computer-assisted selves will be able to do so much more that it will be as if we have spent years evaluating and testing each new advance.
I don’t think this is credible. There is a limit to how much we can expect the considerable genius of our species to be able to create in the decades ahead. While I have tremendous confidence in our ability to solve problems and create new ways of thriving, innovation cannot be churned out on an assembly line basis. New theories and ways of thinking do not now and have never appeared as convenient solutions to pressing problems. More often than not changes of this type come unbidden, appearing like the surprising beauty of a butterfly in summer – anticipated but never fully predicted.
Kurtzweil believes that these breakthroughs will arrive with plodding efficiency. His roadmap he calls GNR: Genetics, Nanotechnology, Robotics. Let’s take each in turn.
Genetics. I have no doubt that the accelerating research in genetics, proteomics and biotechnology will yield fabulous benefits. However the growth of the field is not particularly fast, let alone unbounded. It is predicated on the increasing number of trained people doing research (which is itself controlled by the number of students entering the field), and the increasing body of knowledge and practices that they draw upon. In addition the very necessary and appropriate regulations that govern medical research require all potential human therapies to undergo thorough testing which can take many years to complete. If the sum of these is an exponential, it is a very gradual one.
Nanotechnology. I have always been skeptical of the magical predictions of nanotechnology proponents, but a full critique would require a separate post. So let’s leave that aside for a moment and assume they can ultimately do what optimists predict. How long will it take? Kurtzweil’s timeline in particular depends upon micron-scale autonomous robots infiltrating our body tissues in order to prolong life and monitor and alter our neuronal activity. While this is possible in theory, I do not believe that this technology can be developed in the very short window that he requires. First of all consider that this assumes advanced robotics, which even Kurtzweil himself places later in his future history. We currently have at best meter-scale robots that can operate semi-autonomously in a natural environment. Reducing that scale by 1000 as well as increasing the degree of autonomy will take decades – looking only at the development and test cycle, never mind the technology breakthroughs required – and that would only get us millimeter-scale robots. (To deploy medical devices in only ten years, with all the testing required, practical devices would have to be in the prototype stage today. My reading of the literature suggests to me that we are probably a decade from that point.) Miniaturization by another factor of 1000 would take the same number of years again assuming exponential increases in technological capability, and possibly many more depending on how much harder the problem is, to get to the micron-scale devices needed for the neural interface which does a lot of the heavy lifting for Kurtzweil’s next phase.
Robotics. Robotics is a hobby of mine. I go to the expos and read the magazines as well a dabbling in the technology myself. Kurtzweil is not interested in robots as such but in robot minds – i.e. machine intelligence. Most of what we think of as “robots” today, like underwater research devices or BattleBots, are actually mindless tele-operated vehicles. In terms of brainpower, most autonomous robots today hover somewhere between a dishwasher and a hand calculator. I have a robot that mows my grass, for example, and although I would never give it up I would hardly call it intelligent. It follows a wire painstakingly installed around the perimeter of my lawn and often gets stuck or confused and calls for help. It is neither fully autonomous nor does it interact with the even semi-natural environment of a landscaped yard without the aid of artificial landmarks. But it’s still better than changing clothes and doing the job by hand. I bought a vacuum robot but it couldn’t cope with our house’s open floor plan or the one inch riser between the tile and the hardwood. Eventually the mechanism got jammed and died.
These are mere examples, and they measure results on the early “linear” part of the exponential curve that we would expect based on the law of accelerating returns. Many people (and not just people who hope to live forever) have argued that robotics is right now at the verge of exploding into a range of practical applications. But I’ve been to the conferences, and I’ve watched people try to give live demos, and I have developed my own first law of robotics. “To a first approximation, all robots are broken.”
To his credit, Kurtzweil has a game plan for so-called strong artificial intelligence – i.e. computer software that can pass an empirical test that puts it on a level playing field with human intelligence. First we deploy legions of micron-scale autonomous robots in our nervous systems (see Nanotechnology); then we “reverse-engineer” (it’s hard to overemphasize how many times he says this) the neuronal interaction inside the human brain in response to any number of possible stimuli; then we encode those results into computer programs that emulate the key processes to get the same results. Voilà – artificial minds.
I agree that we have a working example of a system – the human brain – that we can use as a starting point for understanding that property that people in the humanities call intelligence. Where I demur is the idea that this is a matter of “reverse-engineering.” That term is ordinarily used when you come into possession of a human artifact which you do not understand, but which despite its complexity was designed using commonly-understood principles that you can flesh out by analyzing its function within a known set of possible parameters. The brain is nothing like that.
The human brain is both enormously complex and a product of the chaotic process of evolution. This makes it less engineering than a matter of the messy, historical, and incompletely understood mechanisms of biology. It also has a number of aspects which are essentially mysterious, and although they are clearly the result of brain processes right now it seems inimical to suggest that these ineffable qualities of our subjective experience have essentially mechanical roots. Someone will have to bridge the gap. Whoever manages to do that will be a genius, and genius cannot be scheduled.
- jack*
Comments