Choiciness

What makes a situation/context/event seem to us like a "genuine choice"?

Why talk about choiciness?

• The concepts of "choice" and "preference" are strongly linked.
  • Preferences of an agent (at least "revealed" preferences) are specified in terms of their behavior in situations of choice.
• There seems intuitively to be something to the notion of genuine choice, openness of what can be done.

Choiciness nominalism

Choiciness nominalism — Taking choiciness to be merely a feature/device/construct of our minds that is useful for modeling some other entitites, without commiting to anything "real" corresponding to this construct of our minds.

A concept defined in terms of a nominal construct is also (mostly?) nominal. Therefore, assuming choiciness nominalism implies preference nominalism.

Let's start with some illustrative examples. In all of these, one human, Bob, is observing/modeling/predicting/[trying to understand] another human, Alice.

Complete constraint / strong default

Alice is in a situation that completely constrains her possible behavior, relative to Bob's model of Alice. Examples:

• Alice is locked in a prison cell and put in a straitjacket.
• Alice is commutting to her work place and taking the same route that she always takes.
• Alice finds a ripe apple when she's dying of hunger. She eats the apple.
• Alice is fully immersed in reading a book. She doesn't feel tired, doesn't need to go to the toilet at the moment, is not hungry or thirsty. She doesn't have other higher-priority stuff to do. So she just keeps reading.
• Alice really needs to use the toilet.
• Alice is asleep.
• Alice is running away from a rabid dog.
• Alice is solving a routine problem, like writing monkey code for the one-hundredth boring Django app in her life, and going through the same standard chain of testing, deploying, debugging, etc.

Some important things to notice here:

• What constrains Alice (relative/according to Bob's model of Alice) is not just the physical/causal constraints (physical impossibilities) but also telic constraints (social expectations, her own purposes/goals), as well as Bob's expectations of her tendencies based on Bob's observations of Alice so far (along with Bob's theory of mind/knowledge of human psychology, etc).
• It feels a bit fraught to say that in these cases Alice doesn't have a choice. It is possible for her not to go to the workplace this one day. Maybe she would risk getting fired but something else than work might take priority. Perhaps it's better to say that what all these cases say is that Alice has a strong default and, correspondingly, needs a strong reason to deviate from the default.
• The default is specified on some level of detail. The default includes Alice going to work at approximately the same time, taking the same route, but it doesn't include her listening to the same playlist while doing so or sitting in the same seat on the bus.

Incomplete constraint / no strong default

Sometimes Alice is not fully constrained. There is no strong default. For all Bob knows, Alice may do X or she may do Y and none of them would be surprising to Bob.

• Alice is given a choice between two kinds of ice cream: vanilla and chocolate. Bob's model of Alice predicts that Alice will choose one of them (rather than choosing not to eat ice cream) because he knows that Alice is slightly hungry and also that she likes sweets in general. However, his model is inadequate for predicting her behavior in the context of ice cream.
  • Alice chooses vanilla. On further occasions she also chooses vanilla. Until at some point she chooses chocolate and keeps choosing chocolate forever.
  • Alice chooses vanilla. On further occasions she also chooses vanilla. Until at some point she chooses chocolate but only once. And then she goes back to choosing vanilla forever.
  • Alice's choices seem to be completely random, as if each were drawn from $\text{Bernoulli}(0.5)$.
• Alice gets a new job. She tries ten different ways of getting there from her home, testing each at least five times. Finally, she sticks to cycling there through such-and-such way, switching to something else only temporarily when her bike breaks.

Bob's model of Alice is underconstrained in particular contexts (relative to a given level of detail or interest). If Alice starts behaving reliably in similar contexts, then this reliable behavior gets incorporated into Bob's model of Alice as another constraint. This seems like a canonical case of the concept of preference?

Violation of strong default

Suppose Alice (surprisingly! (to Bob)) doesn't go to her workplace even though there's no apparent (to Bob) reason/explantion for her not going to her workplace. Bob starts thinking about explanations for her aberrant behavior. He starts postulating additional constraints, reinterpreting Alice and the relevant context(s).

Unless Bob has some straightforward/effective procedure for infering these additional constraints in such cases, it's now his behavior that is underconstrained. There is no strong default way for him to refactor his model of Alice.

This is also the case in the previous section on incomplete constraint. However, then Bob doesn't choose; he just waits for more information. He might also do that here: just observe Alice for longer.

Is the choiciness of the modeler/subject/registrar entangled with the choiciness of the modeled/object/registered?

General caveats

• The discussion so far elides the problems of ontology/parsing/thingness/individuation:
  • Why is the space of possible choices/behavior parsed into such-and-such discrete choices?
  • Why is the world-flux parsed in such-and-such contexts?
• Is the talk of constraints appropriate?
  • Bob's model may use stuff that is more rich/elaborate than constraints, like probabilistic models. Ditto the determinants of Alice's behavior that Bob is trying to model.
  • One advantage of constraint-speak is that it may make reasoning more tractable, e.g. constraints may compose more neatly than more elaborate components.
• Why is the talk of choiciness/defaults useful for Bob's purpose of modeling Alice?
  • Presumably, this should tell us something about Alice as agent?

Some tidbits to retro-motivate the discussion so far

We often don't know what the human will do and don't see this as "choice". For example, epileptic seizure. The seizure is strictly speaking "internal to the human" but not in the sense we are interested in (similarly e.g. toxoplasma influencing the behavior of a human/cat/rat). So what is the sense we're interested in?

We postulate (hallucinate? construct?) some internal space of the human such that outputs coming out of this space are generally viewed to result from human will/[genuine choice] (due to V-stuff). It's not a subregion of matter because however you draw the material boundary (e.g. "just PFC+hypothalamus" or whatever), the epileptic seizure can be within it and we still don't perceive it as "choice".

Choiciness = underdetermination by environment?

What even is the environment? The simplest hypothesis is that Env is World (set)minus InternalMindSpace. What is InternalMindSpace though?

In choicy contexts, the human is constrained but underconstrained. Many alternatives are valid. Telic constraints therefore need to be incorporated. What are «telic constraints»?

Single-situational choiciness (preference) versus extended choiciness (wanting, goal-directedness, aspiration). The latter seem to manifest themselves in the former (though maybe not always?).

What distinguishes V-things from other things?

Applying the choicy perspective to oneself

We're not only applying this perspective to other agents. We apply this view to ourselves as well.

Do we perceive ourselves as underconstrained/unpredictable, then notice that we behave reliably and infer that we have a preference?

Choiciness realism

The discussion so far only talked about choiciness from the perspective of Bob (the modeler/subject/registrar), not Alice (the modeled/object/registered). We would like a stronger notion of choiciness and defaults, one that makes some non-trivial claims about Alice herself. The #-/TODO next is to do just that.