Simon P Stevens

I've recently been reading Ray Kurzweil's The Singularity is Near. Kurzweil is a futurist and he most well known for his work on a prediction made of the near future called the technological singularity. First named as such by Vernor Vinge, and combined with something Kurzweil calls "the law of accelerating returns" he believes that technological progress is going to continually increase it's rate of improvement until sometime in the mid 21st century we will create an intelligence superior to our own. Following this we will experience a rate of technological progress so explosive that it can be considered a singularity. Borrowing the term from physics a technological singularity, like it's physical counterpart, has an event horizon beyond which the rules change so drastically predictions become near impossible. Kurzweil considers this most likely to be triggered by an artificial intelligence surpassing human capability. Such an intelligence would have the knowledge and ability to design improved versions of itself at an ever increasing rate.

In the earlier chapters Kurzweil looks at the history of technological progress. When he graphs the rate of progress against a logarithmic time axis the major paradigm shifts of history appear to be happening closer and closer together. Kurzweil uses this graph to predict future innovation.

The major problem I have this is that all of Kurzweil's predictions are based on extrapolating exponential growth. Extrapolation is all very well within a short time frame, but exponential growth in physical systems is usually restricted within limits. In a such a system the negative feedback may also be growing exponentially which means although initially it may be too small to be noticed, after the growth passes some boundary the negative feedback becomes relevant and the overall growth is no longer exponential. It's far more likely that the growth curve is sigmoid, and that we are simply at the early growth phase of the graph. Unfortunately it's impossible to tell where we are on the growth graph. Kurzweil makes the implicit assumption we are at the beginning of the growth curve and we have a significant distance to go before growth starts to slow. We could be at the centre where the negative feedback is about to overtake.

There will be limits. The speed of light could be a hard limit on computing speed, or ultimately heat-death could be the hard limit, but there is a limit somewhere, and it will likely come a lot sooner that the heat-death of the universe. At some point we will reach a level of complexity where the negative feedback inherent in the complexity itself will overwhelm the forward growth. The question is how close are we to the limit and that is something we are only likely to know when we reach it.

Kurzweil does actually discuss negative feedback in chapter two of his book and counters with the suggestion that it is not just one technology that is experiencing exponential growth, but many. As each old technology follows a sigmoid growth pattern and it's growth starts to slow, multiple new ones are ramping up to take it's place. He claims that this allows the growth of the overall system to remain exponential. What he doesn't consider however is that while he has taken feedback into account on a single tech scale, he has ignored the likelihood of feedback on a universal system scale. As the number of related technologies increase so do the number of interconnected nodes between these technologies. As areas of research become more related the complexity of the progression as a whole will start to experience feedback.

Lets be clear on this, I'm not saying that Kurzweil's predictions are wrong, or that they won't happen, what I'm saying is that exponential growth cannot and will not continue indefinitely. If you make predictions about the future based on exponential growth you are also making an implicit prediction that the exponential growth in your subject will continue at least to the point of your prediction. In The Singularity Is Near Kurzweil makes a prediction of 2045 for the creation of true, strong, human level AI. The prediction itself describes explosive growth following the date of the singularity trigger event. I concur that following development of strong AI it does indeed seem likely that there will be a period of increased growth rate as the AI builds and expands on what humans were previously capable of, but the assertion that exponential growth will continue from now up until the singularity event seems founded on nothing more than Kurzweil's desire to witness this event for himself.

The problem I have with Kurzweil is that he touts his prediction of the technological singularity as a scientific and accurate prediction, but it's not. It's scientific up to the point of making predictions from a trending system, but as to whether that trend will continue or not is entirely down to guess work. Extrapolating vastly beyond your data points is not valid science. The further you move from your measured data the more uncertainty you bring.

Of course, just because a prediction isn't based on sound science doesn't mean it won't happen. I truly hope to see the development of strong AI, and to witness a technological singularity unfolding would certainly be an astounding event to live through even if it does mark the end of the human era as Vinge predicts.

A friend has posted a well informed rebuttal of my criticism which you can read on his website Lewy Land.

2/22/2011 12:30:00 PM [permalink]