‘In the year 2525/ if man is still alive/ if woman is to survive/ they may find…’ — Zager and Evans.
So Ray Kurzweil has given his keynote address at SLCC. High time I delivered an essay on his ideas concerning the radical future…
AN ICONIC IMAGE
What would be a good visual image for the 21st century?
One candidate might be what Damien Broderick called ‘the spike’; a chart that shows exponential growth over time. The person most commonly associated with such charts is Ray Kurzweil. His lectures and books are full of them, and there has been much debate over their implications.
But I think a better image is ‘Mount Improbable’. Invented by Richard Dawkins, the mountain’s appearance depends on your point of view. Dawkins wrote, “dwarfed like insects, thwarted mountaineers crawl and scrabble along the foot, gazing hopelessly at the sheer unobtainable heights”. This image was intended to show the sheer improbability of random chance assembling something as complex as an eye.
The sheer cliff face of ‘technological mount improbable’ stands, instead, for lack of knowledge. We do not know how the mind works. We do not have a concise, scientific definition of life. We have yet to work out all the processes that result in us aging as we grow older. Therefore, voices have spoken sceptically about our chances of building conscious machines, halting or reversing senescence, and other marvellous breakthroughs Kurzweil expects sci-tech to achieve. To say ‘we do not know how to achieve X today, and we probably never will know’ is to gaze at the vertical cliff of technological mount improbable and declare it unclimable.
But there is another side to mount improbable. Dawkins described it as “gently inclined grassy meadows, graded steadily and easily towards the distant uplands”. This is a reference to how evolution actually works. That is, not by random chance but by cumulative selection. Genes which happen to build useful adaptations (useful in the sense that they slightly increase the chances of being passed on to the next generation) are retained, while none useful genes which lower the chances of surviving long enough to reproduce are eliminated. And so, step by cumulative step something relatively simple like light-sensitive cells can evolve into something as complex as the eye.
The aspects of Kurzweil’s work that gains the most attention are his seemingly incredible claims regarding massively intelligent machines, uploading the mind into cyberspace, nanomachines that will reverse senescence and so on. But I think his most interesting point was made during an interview with John Brockman: “The kinds of scenarios I’m talking about 20 or 30 years from now are not being developed because there’s one lab that’s sitting there creating a human-level AI in a machine. They’re happening because it’s the inevitable end result of thousands of little steps. Each step is conservative, not radical, and makes perfect sense. Each one is just the next generation in some company’s product”.
Michael Shermer of the sceptics society described scientific progress as the cumulative growth of knowledge over time, in which useful features are retained and none useful features are rejected, based on the rejection or confirmation of testable knowledge. Just as the cumulative steps of evolution conquers the seemingly unscalable face of mount improbable, so perhaps the accumulation of thousands of conservative steps in R+D labs the world over can scale up to the peaks of technological mount improbable.
According to Kurzweil, each step is not at all radical. I suspect most people think truly dramatic consequences can only result from similarly dramatic steps. If such transformative technologies are only a few decades away, where is all the mind-blowing R+D that would point to that fact? Actually, it has often been the case that the most transformative technologies had humble beginnings. Michael Faraday showed how a loosely dangling wire will twirl round and round, dragged by a magnetic field, if you run a current through the wire. So what? Well, that’s the principle of every electric motor, from the featherweight spinning plates of a disk drive, to the mighty pumps pouring tons of fuel into a jet engine. Needless to say, few had the foresight to see that Faraday’s humble proof-of-principle would one day lead to so many kinds of electric motor.
So what about today? Some futurists talk about the transformation of the Web into something like a global brain. That’s not the explicit goal of most R+D today. Semantic Web tools are mostly used for the more conservative purpose of coding and connecting companies’ data so that it becomes useful across the organization. IBM, for instance, offered a service that finds discussions of a client’s product on message boards and blogs, drawing conclusions about trends. Nothing radical here, just ways to improve market research. Semantic tools enable companies to benefit from the metadata they obtain about users, while users benefit from the increased efficiency brought about by the companies’ semantic tools. Such relatively mundane tools and the conservative steps they enable will lay down the foundations upon which the next generation of applications will be built and so it goes on step by cumulative step. Through work like the semantic web, we are able to tie together increasingly large and diverse areas of scientific knowledge, and turn this mess of information into easily-searchable and accessible data. Apply cumulative knowledge to something like Google, and perhaps you ultimately get smart software tools that can read billions of journals and blogs, listen in on all the chatter of the scientific community, and in microseconds see in that complex web answers to intractable problems no human could ever perceive, even if they spent a lifetime studying the data on the Web.
Another point to consider is that this might be the general state of things at any particular time. Look ahead into the future, or back into the past and the accumulation of many little steps amounts to significant change. But how many people concern themselves with technologies from way back, or the shape of things to come? If people make choices based on a comparison between new technologies and what was available recently (a few years ago, say) and they look ahead only as far as a couple of years, then all that should be apparent is a conservative step from a previous technology, to a current capability, to a future possibility (one not too radical, considering what is possible ‘now’). Look far enough ahead and tremendous change seems apparent. But the people who live in a the society in which such technology is commonplace will almost certainly not get there in a single bound. Instead, as per usual, the ascent will be achieved via the accumulation of many little steps. A capability like mind uploading will almost certainly be achieved only when a succession of enabling technologies leading step by step from current R+D and commercially available technology, has been established.
We should also be aware of a possible illusion, which is to think each little step is the same size, when in fact they are getting bigger. This illusion happens because we expect more and more from our technology. Consider the work that Weta Digital was expected to do for Peter Jackson’s three ‘Lord Of The Rings’ movies. They had to deliver 540 shots for the first film. When you consider a major blockbuster normally has 200 CG effects shots, that was a massive undertaking. But, for the second film Weta was expected to deliver 799 shots, and for the third film the number had risen to a staggering 1488 shots. But, it is not the case that the challenge went from hard to infeasible to ludicrous, because the technology and knowledge grew. As Peter Jackson said, “the infrastructure, the organization, the software that was written was all helping the following year as a stronger base”. Put it this way. If Weta had been able to use the tools and knowhow that existed for ‘Return Of The King’ when working on ‘Fellowship Of The Ring’, the 540 shots of the first film would have been turned over without much bother.
If increases in the power of technology and knowledge generally results in us expanding our horizons, that could well mean that the leaps we can take get larger, while the peak we are trying to climb seems as far away as it ever was.
THE UNCERTAINTY OF CONQUERING THE PEAKS
If there is one word that bothers me in Kurzweil’s quote, it is ‘inevitable’. Max More has spoken out against the perception of a technological singularity as an inevitable future event, fearing this is a meme too prone to being hijacked by our tendency to believe in a higher power that will solve our problems for us. Let us therefore adapt technological mount improbable to take into consideration the uncertainty inherent in current transhuman speculations.
Technological mount improbable is a mountain range covered by clouds. We cannot see the peaks that stand for cures for aging, artificial general intelligence, or productive nanosystems. We suspect such peaks exist, others claim they do not. One thing is for sure, even if they do exist our vision is not yet clear enough to allow us to see how to climb up there.
There are many paths winding their way along the mountain range. Some may allow us to reach a peak in a shorter time than you might have thought possible. They stand for new technologies such as improved gene-sequencing, or improved knowledge like an overall theory of how the mind works. With luck and effort we may find ourselves on such paths and ascend to the peaks of artificial general intelligence (AGI) etc. in timeframes measured in years or decades rather than centuries. But there are also paths that lead to dead ends, by which I mean hypotheses regarding the way to treat aging which are incorrect, theories of how to code artificial intelligence that are wrong and so on. The mountain range has peaks that we should perhaps not climb. There are peaks for bio weapons, runaway self-replicating nanobots, genetic experiments that would be outrageous if performed on human beings. We may have to take longer, more winding paths as we navigate the moral and ethical questions that advances in genetics, nanotechnology, information technology and cognitive science will invariably raise.
We can discuss the probability of a particular peak being successfully scaled. The chances increase as more and more R+D seek different ways to ascend it, and as the justifications for attempting the climb increase in number. Smart robots would be useful for industry, convenient as home appliances, tactically decisive in military conflicts. Reverse-engineering the functions of the human brain could well lead to insights into how to build artificial general intelligence, but it would also provide clues on how best to deal with neurological disorders. Many groups from various fields are pursuing their own ideas of how the mind works, how to encode that into AGI and have different reasons for doing so. Perhaps most are on the wrong path, but some may be right, or at least will acquire knowledge that will point us in the right direction.
We should also consider convergent knowledge, whereby a seemingly unconnected area of research comes to our aid. A recent example of this comes from optogenetics, a technique combining lasers, neurology, surgery, and genes taken from certain microorganisms, which all together produces a direct control mechanism for neurons. Team leader, Dr Karl Deisseroth, commented, “these microorganisms were studied for decades by people who just thought they were cool. They didn’t have a thought for neurology, much less neuroscience… [but] without that, we would not be able to build what we did”. Remember, that something like human-level AI may not happen just because some robotics lab is trying to build such a thing, but because many seemingly unrelated areas of research converged on the solution to this notorious problem. Vernor Vinge well understands the potential of cumulative and convergent knowledge: ‘We need to extend the capabilities of search engines and social networks to produce services that can bridge barriers created by technical jargon and forge links between unrelated specialities, bringing research groups with complimentary problems and solutions together’.
People often ask me for my timeframe of when such things as cures for aging, productive nanosystems and AGI will arrive. Unlike Kurzweil, who boldy declared ‘I set the date for the Singularity… as 2045’ (mark that in your calendar, folks) I refuse to speculate. We are not on a highway heading toward a clearly marked destination. We are in a mountain range, dwarfed by its peaks, dimly glimpsing shadowy summits through a fog of ignorance and presented with a bewildering array of twisting paths, most of which probably lead nowhere or to places we would be wise to avoid. We may conquer the peaks within decades, or we may find our efforts thwarted for centuries to come. But one thing I feel is certain. If it can be done, we shall not rest until we have conquered the peaks of mount technological improbable. To turn away from the challenge would be contrary to the inquisitive mind that is our species’ defining characteristic.