home | sought

Monotonicity is Not Enough

Maverick runs after a squirrel through the wet dew grass, and it skuttles up a tree, but the legend forgets the prize when some small brown bird splashes in the bird bath. Soon, the bath is on the ground, and Maverick is tongue-out panting to relax on the patio. She laughs at her dog's absolute stupidity as we pick back up the yard. We are always chasing something, but unlike the dog, we see when we are thrashing. Maverick doesn't need to worry (he needs the exercise), but it got me thinking about this emphasis on monotonicity.

I wrote recently about using monotonicity to structure tasks towards goals. However, in a changing world, we often find ourselves with a new gear shifter for a car that was stolen. Tickets for a band we are no longer interested in. Code for a product requirement that doesn't exist anymore. Sunk cost for a goal that doesn't make sense. Using monotonicity means that we are sure where we are going. However, if we haven't found our footing. If we are surveying new grounds. If we want to account for our own inability to predict the world (thanks Taleb), then it's not a framework we can use. We need to hedge our bets. Why tie yourself to a goal if it's likely to change?

So, instead, do things that are locally useful, without regard to a specific goal:

• meet new people (contrast my recommendation with Seneca Letter VII: be weary of their vices. I say you want to establish in yourself first the ability to avoid vices. Acting like Diogenes, giving up the need with the benefits. It is impressive, but I am not into it)
• be friendly to people
• eat healthy food
• exercise
• get enough sleep
• do not worry about something you cannot change
• do your work
• do not lie to yourself
• change something, and see if you understand the result

If you are unaccustomed to a problem domain, and you pick the wrong goal, there are problems you might hit:

• you might Goodhart's law yourself into a bunch of other problems
• your monotonic path turns out to be a sinuous river in a more appropriate domain representation (waste of time, regardless of the quality of a goal)
• the problem space is too complex, and goal-change thrashing doesn't converge (need to play the meta)

Nowak (SuperCooperators again) points out that the biblical "what goes around comes around" encodes this idea. He talks about how these strategies are very useful at the group level. Given individual survival tied up in the group, selfish strategies are like a cancer. A locally selfish decision towards any goal is not repeatedly maintainable (this is not to agree with Kant's universal categorical imperative).

What do you do if you are tied to the goal? You can view addiction in this framework. Oneitis in this framework. The flavor of Coca-Cola in this framework. The goal (that high) eats value. The goal is a lie. Tell yourself the truth. Then when you do see a mountain, and you have the gear, and you have the vision, climb to the peak, but generally hedge by picking a path which takes locally useful edges (when viewing them from a variety of life goals or general principles).

So let me try to live this domain-independent for a while.