Monday 18 August 2008

Why Questions and Rational Agents

In this post I continue to limn a novel theory of action. I hold that an intentional agent counts as a rational agent if and only if that agent is responsive to should-questions. Examples of should questions include: “Should Sue have a third glass of wine?”, “Should I believe in God?” and “Should I desire revenge?” To ask if Sue should have a third glass of wine is to ask if having a third glass of wine is good. This gives rise to the following question. If having a desire already involves an appearance of the good (as I have claimed), then does not Sue’s desire for a third glass of wine already entail an affirmative answer to the question “is having a third glass of wine good?” I hold that in the case of animals that are unable to deliberate and thereby arrive at judgements about what is good this is in fact the case. However, agents with the capacity for rational deliberation may arrive at judgements about the good that are at odds with their desires. In such agents what appears to be good a la desire may be different from what is judged to be good.

What is judged to be good reflects the perspective of the agent. What appears to be good a la desire reflects the perspective of some subsystem of the agent. On this score, the analogy with theoretical reasoning is rather straightforward. For example, we may say that from the perspective of the visual subsystem it is true that a partially submerged stick is bent. Even so, from an agent-level perspective, it may be false that the partially submerged stick is bent. Thus, both theoretical and practical cognitions exhibit two distinct levels of evaluation—namely, sub-agential evaluations and agent-level evaluations. The sub-agential evaluations in theoretical and practical reasoning are (perceptual) appearances and desires respectively:
(T2): Appearances are sub-agential perspectives on the true

(P2): Desires are sub-agential perspectives on the good
The agent-level evaluations in theoretical and practical cognitions are theoretical and practical judgements respectively:
(T3): Theoretical judgements are agent-level perspectives on the true

(P3): Practical judgements are agent-level perspectives on the good
Sub-agential evaluations, (T1) and (P1), may be had by both rational and proto-rational animals alike. However, in the case of proto-rational animals, an appearance of the good a la desire, leads to intentional action without agent-level evaluations. For such agents, there is only one level of evaluation—namely, the sub-agential. However, for rational animals there are two levels of evaluation: the sub-agential (as represented by desires) and the agent-level (as represented by practical judgements).

Rational deliberation is the process by which competing sub-agential evaluations are weighed against each other in order to arrive at a unified agent-level evaluation of the true/good. The motivation for rational deliberation comes from a desire, inherent in all rational agents as such, for a unified perspective on the true/good. In cases in which all the relevant sub-agential systems agree on what is true/good (which may very well be the majority of cases) a unified agent-level perspective is easy to come by. In such cases, the need for deliberation is minimal and there may be very little difference in the cognitive activity of a rational and proto-rational animal. However, in cases in which two (or more) sub-agential evaluations are at odds with each other, a unified agent-level evaluation can only be arrived at by endorsing one and rejecting the other. In such cases, the advantages of having a capacity for rational deliberation become most apparent.

In cases in which a proto-rational animal experiences two competing sub-agential evaluations, which ever exerts the greater volitional force ultimately determines what the animal believes (in the case of appearances) or which action is performed (in the case of desires). For such animals there is no agent-level evaluation or all-things-considered judgement about what should be believed or done. For such animals, should-questions are simply not applicable. This imposes a significant limitation on the ability of proto-rational animals to achieve right belief or action. Which of the two competing sub-agential evaluation happens to exert the greater motivational force may not be the one that actually gets things right. Consider the partially submerged stick example. Suppose that while it appears true that the stick is bent, from the perspective of the visual sub-system, it also appears true that the stick is straight from the perspective of the tactile sub-system. We can imagine a proto-rational animal for which the deliverances of the visual system exerts greater motivational force vis-à-vis its beliefs than the tactile system. Such an animal would form the belief that the partially submerged stick is bent when it is in fact straight.

Contrast this with rational agents that have the capacity for agent-level evaluations. Confronted with competing appearances (each relative to a particular perspective), the rational agent can only arrive at a unified agent-level assessment by endorsing one of the sub-agential evaluations and rejecting the other. It decides which of the sub-agential evaluations to endorse based on which best coheres with its beliefs, values and commitments. For example, we can suppose that the rational agent has observed the stick being submerged and removed from the water and that it also has the belief that sticks are not the kind of things that can bend and straighten themselves out at will. Moreover, it may also believe that optical illusions are far more common than tactile illusions. When the deliverances of the visual and tactile subsystems are considered in the light of these beliefs, the rational agent may be led to see the tactile subsystem as a more reliable guide to the true, in the particular case. This may then prompt the rational agent to judge that the partially submerged stick is actually straight. I will refer to beliefs that are the product of rational deliberation as judgements. Judgements represent a special subset of beliefs that only animals with the capacity for rational deliberation can have. Judgements have a greater chance of getting things right than beliefs that are not judgements since they are not simply the product of what sub-agential evaluation happens to be stronger.

The upshot of a capacity for judgements is that an animal is more versatile in forming true beliefs in novel circumstances. What appears to be true/good is simply a product of the type of subsystems we have, given our evolutionary history. The fact that it appears to be true that a partially submerged stick is bent is simply an upshot of the type of visual system we happen to have. If instead of light waves, which are subject to refraction, our visual system actually detected delta-waves, which are not subject to refraction, then our visual systems would not tell us that partially submerged sticks are bent. Likewise, the fact that I may desire sweets (perhaps contrary to my better judgement) is also primarily the product of my evolutionary history.

Given how natural selection works, our subsystems offer the types of evaluations that they do because they tended to get things right in the environment of our biological ancestors. However, when we are confronted with unfamiliar circumstances (ones that our ancestors may not have frequently encountered) there is a greater chance that one or more of our subsystems would get things wrong. Rational deliberation serves as a corrective against the fallibility of our subsystems which (though open to modification via training) cannot be adapted at the rate that our agent-level evaluations can. In brief, agent-level evaluations increase our chances of getting things right in novel circumstances since they may serve as a corrective against the errors of our less plastic sub-agential evaluations. This gives rational animals a clear advantage over proto-rational animals in arriving at true beliefs and good actions.