Published on

Miller's Law and its AI implications

Useful chunks and other tradeoffs

Miller's Law:

"The Magical Number Seven, Plus or Minus Two: Some Limits on Our Capacity for Processing Information"

The exact numbers and types of conceptual chunks are not as important as the point that evolution has made some trade-offs when it comes to what the brain is built to handle.

Too few chunks as a limit, and a person's ability to juggle multiple simultaneous ongoing situations would be impaired : There are obvious ways in which this could be harmful for survival.

Too many chunks as a limit, assuming that access speeds and detail are not impaired, then more brain space must be given over to keeping the various chunks active.

Overcoming limitations

In everyday life, people have strategies to overcome these limitations : For instance they might write information down, completely side-stepping having to keep all the information 'current' (or 'top of mind').

Why have limitations ?

Suppose that general AI was attainable, and the mechanism for the retention of a single idea at the 'top of mind' was understood (i.e. the implementation of the brain was done in a way that exposed some of the key 'modules' - rather than being an all-or-none black box). Then the capabilities of the brain could be extended simply by expanding the 'top-of-mind' capacity to (say) 20 items.

It would be difficult not to grant that the AI was now superior in some respects to a regular human brain. Maybe this superiority would not give it advantages in 'human' situations (but humans may have already self-selected the situations in which we find ourselves to avoid 'getting in too deep'). The machine would simply be able to juggle more mental balls simultaneously.

So : Here's an example of how a superior intelligence might be created, once the basic AI model was completed.

Not too threatening, but undeniably better.