29 May 2017

On Time-Free Theories of Decision Making

"Rationality" is often espoused as a method of understanding what is true, and as a method of making decisions on how to act. As the former, it has (in many respects) been very successful. On the latter front, as a program for an agent, it is often somewhat less successful. In areas where decisions can be made with substantial delay, there is a finite problem-space, and the ability to systematically evaluate options, rationality works fine.

When relaxing these constraints, things get dicier. Given an unbounded possibility space for solutions, and a lack of self-evidently perfect solutions, the time to come up with an optimal response can easily be exponential in both the number of concerns involved, which is to say that assuming P≠NP, we must choose between thinking until the heat death of the universe, or caring about the time it takes to formulate a response. This is not to say that rationality consists solely of exhaustively searching problem spaces; only to point out that almost any decision-making problem operating on open problem-spaces must contend with the speed at which it's able to produce answers.

However, if you ask rationality how to hold a conversation, or design something in an unbounded problem-space, you will likely have issues. In regular conversations, as opposed to what happens on Twitter dot com, it is imperative to produce replies in real time, usually beginning sentences before knowing exactly how they end. Given that even slight delays can ruin a response, conversations are hard to explain as revolving around rational methods of becoming right and using that knowledge to take the correct action.

There are many domains that straddle the line between the improvisation of conversation, and the slow decision-making of, say, Chess. For these, we may make plans. Rational types particularly like making plans. However, they sometimes fail to acknowledge that carrying out a plan is seldom an act that looks the same as their plan. They will often cling to the Mohist delusion that, if they create the best plan, and follow it exactly, they can have an ideal result. Some will allow that their plans will not totally solve the issue, and some improvisation will be necessary. However, it is rare to see them consider that their plan provides little or no value, and the activity's success is dominated by their ability to act in a moment.

I don’t want to say that correct knowledge is an impediment to solving these unbounded problems, although it can certainly be a distraction. There are certain ways in which it can be applied—to eliminate sets of possibilities that are impossible, or to suggest directions for further movement. However, beyond the low-hanging fruit, seeking correct knowledge comes with an obvious time cost. If you are sitting in an endless ocean of possibilities, is it better to try to systematically light the areas around you, or shine your light near and far for glimpses of bright reflections?