RE: Challenge of Design Complexity

Billy Brown (
Tue, 15 Dec 1998 10:07:53 -0600

Robin Hanson wrote:
> Similar issues have been considered. For example, here is an exerpt from

Yes, I've read the Singularity debate. My argument is a rather different one that your comment about IQ enhancement, and it leads to different conclusions.

If you make an AI that increases its own IQ by 1%, that tells you nothing about how difficult the next 1% improvement will be. It could be 10% harder, or 0.5% harder, or not any harder at all. The only way to find out is to try it and see what happens.

What we can predict is that if you try to repeatedly enhance one narrow ability, you will eventually reach a point of diminishing returns. At some point, you will find that further improvements to ability Y are impractical, because ability X can not handle the increased complexity of your design.

This effect is not unique to intelligence enhancement. Something similar occurs whenever any narrow field of research pulls ahead of the general rate of progress. Eventually it reaches a point where further advances require improvements in some unrelated technology, and you have to wait for those improvements to be made.

Now, this does not mean that the whole thing is a game of diminishing returns. Recent history demonstrates that if you research enough different things, you can create a situation in which the ability of your society to make technological advances increases faster than the difficulty of taking the next step. A reasonable extrapolation of the trend would predict a century or two of steadily-accelerating progress before things begin to change so fast that an unenhanced human can't cope.

Intelligence enhancement (IE) of any kind would, however, add a new dimension to this saga. Roughly speaking, the our rate of progress is determined by:

					    R * P * I
	Progress / unit of time =  -----------
						T * C

Where R represents the resources available to each researcher, P is the population of researchers, I is the average intelligence of the researchers, T is our current level of technological sophistication, and C is a measure of the time and effort required for researchers to communicate. Most of the increasing rate of change in recent times comes from a slow geometric increase in both R and P, and a steady drop in C. Since the changes in R and C are both due to technology, the whole process tends to feed on itself.

Meaningful IE would make I increase in roughly the same fashion as R. Not only would this dramatically speed up our rate of advance, it would also increase the rate at which our rate of advance speeds up.

This post is long enough already, so I'll leave off speculating about the Singularity itself until next time.

Billy Brown, MCSE+I