A reader has remarked that the essence of Kelly's What Technology Wants is contained in Ray Kurzweil's The Singularity Is Near [1]. He suggested that it be reviewed here.
All I knew about Kurzweil was that he was reported to believe that we are on the verge of eternal life: recent advances in medicine can lengthen life enough that the exponentially accelerating further advances will suffice to further extend our lives, and so on into eternity.
The second Kurzweil blip on my radar was the unfavourable review of [2] by Colin McGinn in the New York Review.
So that was two negative blips. By the time I had finished reading the McGinn review my interest was tweaked. You see, it was clear that McGinn was trying to make the review of the kind that people call "devastating". I became curious why he was trying so hard.
A sample of McGinn's review shows an academic, a world authority on the Philosophy of Mind, confronted with the impertinence of a computer engineer barging in where angels fear to tread:
So he is a computer engineer specializing in word recognition technology, with a side interest in bold predictions about future machines. He is not a professional neuroscientist or psychologist or philosopher. Yet here we have a book purporting to reveal -- no less -- "the secret of human thought." Kurzweil is going to tell us, in no uncertain terms, "how to create a mind": that is to say, he has a grand theory of the human mind, in which its secrets will be finally revealed.It turns out that Kurzweil is remarkably well read, taking into account that he is a mere computer engineer. To me this computer engineer comes across as sufficiently well-read and intelligent for me to suspect he enjoys provoking those across the chasm that separates the "professional philosophers" from people who actually do something. Consider the title of [2] How To Create A Mind: the secret of human thought revealed. After reading some of the philosophers of mind, I came to the conclusion that "mind" is such an elusive concept as to be useless and that the same holds for "thought". Sending your book into the world under this banner looks to me like a tongue-in-cheek provocation, and that is a good start.
It turns out to be a good start for a good read, if one keeps in mind that nobody seems to be able to explain the meaning of "mind" or "thought", even though we all use these words. Kurzweil indicates where he stands with respect to Philosophy of Mind by quoting Marvin Minsky: "Mind is simply what brains do." Don't misread: I don't think that Minsky believes that what brains do is simple.
While academics laboured on dissertations showing that mind is not this and thought is not that, Kurzweil invented, filed patents, and launched and managed companies that took computer applications to new heights of sophistication. For those interested in what has been happening and what might happen next, this book is a good introduction.
So, what has been happening in computers? Here only some remarks about the role of the general-purpose computer, about growth of computers, and about expert systems.
For decades computers have been used to recognize printed characters or spoken words by means of general-purpose hardware running software based on networks of neuron-like data structures. The data structures are abstractions based on what was known about neurons in 1940. Though much progress has been made in this way, it is time we update the tried-and-true paradigm, because by now we know a lot more about neurons.
It has always been tempting to build special-purpose computers by incorporating at the level of the silicon more of the knowledge we have about how neurons work. For example Carver Mead has described a way of using computer technology at the chip level that is more like neurons than simulations on neurons on general-purpose computers [3]. Even though this is a quarter of a century old, it is the most recent that Kurzweil quotes on the possibility of special-purpose hardware. So far artificial intelligence has piggy-backed on the enormous increase in power of general-purpose hardware. It makes sense not to do anything about this until the hardware becomes a limiting factor.
Special-purpose hardware is one direction in which there is room to grow. Beyond that there is a more radical departure from general-purpose computers. Even Mead's radical departure from purely digital assumes top-down control over the layout of the chip. Biological brains achieve their high degree of miniaturization by growing, that is, being built bottom-up at the molecular level. The equivalent of chip layout is achieved by being exposed to an environment that causes an effective pattern of interconnections to form. That is, by learning. Doing this in silicon or in more suitable semiconductors is a possibility when Mead's combination of analog and digital runs out of steam.
By the 1970s computer programs ("expert systems") were built that incorporated the knowledge and performance of an expert in a specialized area, say, gas chromatography. The fact that the effort was successful deservedly received a lot of attention. Yet the practical impact was limited because it was not enough to be an expert in gas chromatography to build such a system. For that a team was needed that not only included such an expert, but also experts in other fields.
The additional expertise concerned how to transfer knowledge from the expert to a form digestible by computer software. This makes demands on both the software and the form of the knowledge. This is a far cry from the expert verbally expressing her knowledge to a class of students. So far it has not been possible for a computer system to pick up anything from the unstructured informal kind of information that humans learn from.
This is why a system such as "Watson", designed to compete in the game show "Jeopardy", is so intriguing: not all of its information needed to be explicitly entered by humans. A significant part came from "Watson" being set loose to read massive amounts of text aimed at human readers, including in this case the reading of the entire Wikipedia. Of course "reading" in this case only meant to pick up lots of associations; not what we would call "knowledge" [4].
The Singularity is Near is the earlier of the two Kurzweil books that I'm reporting on. It is the one with the wider scope: Chapter Four is a preview of How To Create a Mind. "Singularity" has a wide scope indeed: it begins long before humans, or even life, existed and projects to a future in which intelligence has evolved beyond recognition. In Chapter One Kurzweil distinguishes the following six epochs in cosmic history:
In Chapter Two Kurzweil starts by remarking on the acceleration of development that is already apparent in the six epochs: evolution spent more time in epoch 1 than in epoch 2 and so on down the list. We are in epoch 4 and within this epoch acceleration is apparent. By the time we reach the late 20th century we have Moore's law predicting the increase in performance of computer chips remarkably well for the past decades. The increase is exponential: it is not that a constant amount of performance has been added, but that performance has been multiplied by a near-constant factor year over year. In Chapter Two we find an intriguing variety of graphs showing that exponential increases in performance have been found in a variety of other technologies.
Will a computer be able to achieve human intelligence? This would be an important milestone in the development of technology. A simple-minded approach is first to get a computer to reach the computational capacity of the human brain (for now, no computer does) and by that time to make sure we can suitably program such a computer. An extrapolation of Moore's law says yes, computer power will grow to equal that of the human brain, and says that we don't have to wait very long. But it is also likely that Moore's law will break down, as exponential laws always do, and that this will happen before that time.
It is wise to remember that Moore's law has applied to a narrow paradigm: a general-purpose processor with von Neumann architecture on a single chip with a two-dimensional layout. Chapter Three explores several alternative paradigms according to which computer hardware can develop in the future. These are likely to pick up where the current paradigm leaves off. It is also wise to remember how the human brain was designed, or, rather that it wasn't designed at all, but got assembled randomly. With a modicum of design human intelligence will be achieved with less computing power.
With inexorable logic Chapter Four considers how to program a computer with the computational capacity of the human brain. This is a crucial consideration: where technology has been extremely successful in hardware, the opposite has been the case in software.
Chapter Five concerns three overlapping technologies: genetics, nanotechnology, and robotics. If we substitute computing for robotics and forget about nanotechnology, then we have just two overlapping technologies with the striking phenomenon that the same rapidly increasing performance holds in two different domains.
The thesis of the book is that the various exponential rates of improvement lead to an imminent discontinuity in history. But could it be that the exponential phenomenon is restricted to the high-technology world? What's it going to do about the fact that a large proportion of humanity suffers from a lack of food and other essentials?
Kurzweil does address this question and points out that genetics has the potential to drastically improve food production. He blithely ignores that science is not enough. For example, in 2002 a drought in Africa threatened the lives of 15 million people. A 15,000 ton aid shipment of US corn, about one third of it genetically modified, was turned away by Zimbabwe on the grounds that some kernels might be planted and thus endanger future exports to the European Union, where sale of GM corn is forbidden. Part of the shipment was offered to neighbouring Zambia, which had accepted such shipments in several earlier years. In 2002 it was rejected even though three million Zambians were facing famine. President Levy Mnanawasa declared: "Simply because my people are hungry is no justification to give them poison, to give them food that is intrinsically dangerous to their health" [5]. Did anybody tell the President that Americans have been eating that stuff for years with no apparent ill effects?
I recommend The Singularity is Near: when humans transcend biology and wonder what it takes for humans to transcend stupidity.