>On Wednesday I saw Ray Kurzweil give a short talk
>* He graphed the speed of computing devices from 1900 to present, and
> claimed that computing speed is not only increasing exponentially,
> but that the rate at which computing speed is increasing is
> increasing exponentially (the graph on a log scale curves upwards).
Damien Broderick, <damien@ariel.ucs.unimelb.edu.au>, suggested:
>
> Probably this one:
>
> http://www.frc.ri.cmu.edu/~hpm/book98/fig.ch3/p060.html
...which is Hans Moravec's graph.
Note that Moravec comments, "The reduced spread of the data in the 1990s is probably the result of intensified competition: underperforming machines are more rapidly squeezed out." Indeed, the data points in the 1990s are clustered at the upper end of what had been a consistently wide band in previous decades.
But this seems to undercut his thesis of super-exponential growth. We see a wide band getting pruned so that only its upper edge can exist due to competitive pressures. This means that there is not an actual speedup of growth rates, but simply a selection effect, where only the fastest computers of the 1990s are shown on his chart, while a wider range of machines was shown in earlier decades. If we imagine putting back in the full range of 1990s computers which would have been manufactured had the market been less cut-throat, as in previous decades, the extrapolations would be very different. The data would be consistent with a slow growth rate up to 1940, a fast rate from 1940-1960, then a slightly *slower* but constant growth rate from 1940 through the present.
In any case, these kinds of extrapolations are worthless, in my opinion. The fact is that we face significant technological challenges in trying to maintain these rates through the next two decades. Virtually every technology of the past has gone through periods of exponential growth, only to reach some limiting factor. I expect the same thing to happen to electronics, in the pre-nanotech era.