The trade-off between speed and complexity

The hypothesis that there has been selection pressure for mechanisms which enable us to perceive the present tends to be conflated with the hypothesis that there has been selection pressure for mechanisms that compensate for inevitable neural delay. The relationship between the two is more subtle, because increases in neural delay can be advantageous for building more useful perceptions. Proponents of the hypothesis that the brain has mechanisms for perceiving the present (i.e., mechanisms designed to generate a perception at time t that is representative of the scene at time t) typically say that the advantage is that it helps overcome inevitable neural delays. That is, “latency compensation” and “perceiving the present” have gone hand in hand. I, too, have made this equation in my own articles on how these ideas may be integral to a unified account of illusions (Changizi 2001; Changizi & Widders 2002; Changizi et al., in press; although see Changizi 2003, pp. 75–77). The implicit assumption can often seem to be that natural selection has attempted to minimize neural delays – by shortening wires, speeding up signal propagation, and using rapid computational algorithms for generating a visual percept – and whatever latency between retina and perception is left is handed over to the compensation mechanisms to deal with. Although this is an open possibility, the hypothesis that we perceive the present is not committed to this possibility; it is only committed to the idea that perceptions belong to the present. What is left open is how long the delay is, and whether it is all “inevitable” or whether the delay may be much longer than it would be if selection for short processing times trumped all other selection pressures. Consider computer software as an analogy. Computer processing speed has risen by many orders of magnitude over the course of the last 20 years, but you may have noticed that many of your programs still take considerable time to start up. Computer designers know how long a wait we are willing to endure, and use that time to carry out fancier computations. That is, when faster computers arrive, computer designers do not appear to be saying, “Now we can compute the same old things nearly instantaneously!” Instead, they seem to be saying, “Now think about how much more we can compute while the user waits!” Just as computer software delay is a consequence of a trade-off between shorter delay and more complex computations, our perceptual delay is a trade-off between shorter delay and fancier visual computations. For example, if evolution can find a new clever trick for extrapolating farther out into the future – say from 30 msec to 120 msec – then it could utilize this trick and allow itself four times the amount of computation time to build sophisticated useful perceptions. The resultant latency of 120 msec would not be understood as an inevitable delay left over after trying to reduce it as much as possible. Instead, it would be better to say that there is selection pressure to maximize the delay for which the nervous system is able to compensate, thereby buying more time to prepare the perception. Counterintuitively, then, it may well be that the slower-to-react brains are the “smarter” ones. Visual prediction as indicated by perceptual adaptation to temporal delays and discrete