October 21, 2005

Bandwidth vs. latency

I recently attended a robotics seminar in which the speaker made an interesting point about machines outperforming humans (in some domain), but unfortunately used the word bandwidth when he really meant latency.

His point was that machines can react much faster than humans. Indeed, humans have relatively long reaction time, ranging in order from a few tens of milliseconds, to a just short of a lifetime, depending on the stimulus, the complexity of the analysis, and the complexity of the action. When it comes to latency, for simple reactive actions, current computers and robots can already outperform humans, sometimes by several orders of magnitude.

On the other hand, humans have extremely large input bandwidth. The amount of sensory data that humans can absorb and process per unit of time is still much larger than what computers can deal with, both in terms of physical bandwidth, and computing bandwidth.

In the world of digital signal processing, where bandwidth directly constrains processing time, latency and bandwidth are--artificially--tied. In the real world, including in our massively parallel brains, bandwidth and latency are not so directly coupled.

No comments: