John Pollock

Practical reasoning aims at deciding what actions to perform in light of the goals a rational agent possesses. This has been a topic of interest in both philosophy and artificial intelligence, but these two disciplines have produced very different models of practical reasoning. The purpose of this paper is to examine each model in light of the other and produce a unified model adequate for the purposes of both disciplines and superior to the standard models employed by either.The philosophical (decision-theoretic) model directs activity by evaluating acts one at a time in terms of their expected utilities. It is argued that, except in certain special cases, this constitutes an inadequate theory of practical reasoning leading to intuitively incorrect action prescriptions. Acts must be viewed as parts of plans, and plans evaluated as coherent units rather than piecemenal in terms of the acts comprising them. Rationality dictates choosing acts by first choosing the plans prescribing them. Plans, in turn, are compared by looking at their expected values. However, because plans can be embedded in one another, we cannot select plans just by maximizing expected values. Instead, we must employ a more complex criterion here named coextendability.

The objective of this book is to produce a theory of rational decision making for realistically resource-bounded agents. My interest is not in “What should I do if I were an ideal agent?”, but rather, “What should I do given that I am who I am, with all my actual cognitive limitations?” The book has three parts. Part One addresses the question of where the values come from that agents use in rational decision making. The most comon view among philosophers is that they are based on preferences, but I argue that this is computationally impossible. I propose an alternative theory somewhat reminiscent of Bentham, and explore how human beings actually arrive at values and how they use them in decision making. Part Two investigates the knowledge of probability that is required for decision-theoretic reasoning. I argue that subjective probability makes no sense as applied to realistic agents. I sketch a theory of objective probability to put in its place. Then I use that to define a variety of causal probability and argue that this is the kind of probability presupposed by rational decision making. So what is to be defended is a variety of causal decision theory. Part Three explores how these values and probabilities are to be used in decision making. In chapter eight, it is argued first that actions cannot be evaluated in terms of their expected values as ordinarily defined, because that does not take account of the fact that a cognizer may be unable to perform an action, and may even be unable to try to perform it. An alternative notion of “expected utility” is defined to be used in place of expected values. In chapter nine it is argued that individual actions cannot be the proper objects of decision-theoretic evaluation. We must instead choose plans, and select actions indirectly on the grounds that they are prescribed by the plans we adopt. However, our objective cannot be to find plans with maximal expected utilities. Plans cannot be meaningfully compared in that way.

The question addressed in this paper is how the degree of justification of a belief is determined. A conclusion may be supported by several different arguments, the arguments typically being defeasible, and there may also be arguments of varying strengths for defeaters for some of the supporting arguments. What is sought is a way of computing the “on sum” degree of justification of a conclusion in terms of the degrees of justification of all relevant premises and the strengths of all relevant reasons.

OSCAR is a fully implemented architecture for a cognitive agent, based largely on the author’s work in philosophy concerning epistemology and practical cognition. The seminal idea is that a generally intelligent agent must be able to function in an environment in which it is ignorant of most matters of fact. The architecture incorporates a general-purpose defeasible reasoner, built on top of an efficient natural deduction reasoner for first-order logic. It is based upon a detailed theory about how the various aspects of epistemic and practical cognition should interact, and many of the details are driven by theoretical results concerning defeasible reasoning. The architecture is easily extensible by changing the set of inference schemes supplied to the reasoner. Existing inference schemes handle many kinds of epistemic cognition, including reasoning from perceptual input, causal reasoning and the frame problem, and reasoning defeasibly about probabilities. Work is underway to implement a system of defeasible decisiontheoretic planning.

It’s morning. You sit down at your desk, cup of coffee in hand, and prepare to begin your day. First, you turn on your computer. Once it is running, you check your e-mail. Having decided it is all spam, you trash it. You close the window on your e-mail program, but leave the program running so that it will periodically check the mail server to see whether you have new mail. If it finds new mail it will alert you by playing a musical tone. Next you start your word processor. You have in mind to write a paper in moral philosophy about whether people who send spam

Probability is sometimes regarded as a universal panacea for epistemology. It has been supposed that the rationality of belief is almost entirely a matter of probabilities. Unfortunately, those philosophers who have thought about this most extensively have tended to be probability theorists first, and epistemologists only secondarily. In my estimation, this has tended to make them insensitive to the complexities exhibited by epistemic justification. In this paper I propose to turn the tables. I begin by laying out some rather simple and uncontroversial features of the structure of epistemic justification, and then go on to ask what we can conclude about the connection between epistemology and probability in the light of those features. My conclusion is that probability plays no central role in epistemology. This is not to say that probability plays no role at all. In the course of the investigation, I defend a pair of probabilistic acceptance rules which enable us, under some circumstances, to arrive at justified belief on the basis of high probability. But these rules are of quite limited scope. The effect of there being such rules is merely that probability provides one source for justified belief, on a par with perception, memory, etc. There is no way probability can provide a universal cure for all our epistemological ills.

Projectible conditions are (roughly) those whose universal generalizations are con firmed by their positive instances. This paper proposes certain modifications to the above definition in order to capture the pre-analytic notion it is supposed to explicate. Then we investigate what logical operations, when performed on projectible conditionals, yield new projectible conditionals. A number of surprising theorems are proven, and these theorems indicate that few conditionals having complex antecedents and consequents are projectible. It is also shown that projectibility is not closed under contraposition, and this is proposed as a solution to the Paradox of the Ravens. Finally, a general conjecture is made concerning just what conditionals are projectible, and it follows from this conjecture that although most conditionals are not projectible, they can still be dealt with inductively, but in ways more complex than recognized by classical confirmation theory.

In its present incarnation, OSCAR is a fully implemented programmable architecture for a rational agent. If we just focus upon the epistemic reasoning in OSCAR, we have a powerful general-purpose defeasible reasoner. The purpose of this paper is to describe that reasoner. OSCAR's defeasible reasoner is based upon seven fundamental ideas. These are (1) an argument-based account of defeasible reasoning, (2) an analysis of defeat-status given a set of interrelated arguments, (3) a general adequacy criterion for automated defeasible reasoners, called “d.e.-adequacy”, (4) the claim that a generalpurpose defeasible reasoner must have the kind of structure I have called “flag-based”, (5) an algorithm for computing defeat-status, (6) an interest-driven monotonic reasoner, and (7) an account of degrees of justification, degrees of interest, and their interactions. These will each be discussed in turn.

Probabilities are important in belief updating, but probabilistic reasoning does not subsume everything else (as the Bayesian would have it). On the contrary, Bayesian reasoning presupposes knowledge that cannot itself be obtained by Bayesian reasoning, making generic Bayesianism an incoherent theory of belief updating. Instead, it is indefinite probabilities that are of principal importance in belief updating. Knowledge of such indefinite probabilities is obtained by some form of statistical induction, and inferences to non-probabilistic conclusions are carried out in accordance with the statistical syllogism. Such inferences have been the focus of much attention in the nonmonotonic reasoning literature, but the logical complexity of such inference has not been adequately appreciated

To summarize, in order for rational agents to be able to engage in the sophisticated kinds of reasoning exemplified by human beings, they must be able to introspect much of their cognition. The problem of other minds and the problem of knowing the mental states of others will arise automatically for any rational agent that is able to introspect its own cognition. The most that a rational agent can reasonably believe about other rational agents is that they have rational architectures similar to its own, and that they have thoughts related to their rational architectures in certain ways. This leads to Rational Functionalism as an account of what it is to be a cognizer having mental states. That in turn entails that a computer can be a person in precisely the same sense as my next door neighbor if it can appropriately mimic my rational architecture. There is nothing I could know about my neighbor that I could not believe with equal justification about the computer. Rational Functionalism also makes it reasonable to define the narrow content of a thought to be its overall place in the agent's rational architecture, that is, its conceptual role. This is, however, avery narrow notion of content. For practical purposes, we are not interested in knowing the narrow contents of other people's thoughts. We are only interested in rather general properties of those narrow contents. This is what is expressed by the use ofthat-clauses in public language

As anyone who is familiar with the literature knows, there is a great deal of controversy concerning which, if any, of the extant theories of propositional modal logic correctly formalizes the logic of certain logical concepts such as analyticity and logical necessity. Most of the controversy concerns certain principles that involve iterated modalities (where one modal operator occurs within the scope of another). For example, there is considerable disagreement about whether the principle(□p⊃□□p)should be considered valid. However, when philosophers and logiciansapplymodal logic to concrete problems, they rarely need principles which involve iterated modalities. For most practical purposes, principles involving only one layer of modalities are all that are needed. This suggests that if we try to construct a theory of modal logic in which there are no iterated modalities, we can avoid most of the controversy and still have a theory that is strong enough for all of the normal uses to which modal logic is put.

A theory of rational choice is a theory of how an agent should, rationally, go about deciding what actions to perform at any given time. For example, I may want to decide whether to go to a movie this evening or stay home and read a book. The actions between which we want to choose are perfectly ordinary actions, and the presumption is that to make such a decision we should attend to the likely consequences of our decision. It is assumed that these decisions must be made in the face of uncertainty regarding both the agent’s initial situation and the consequences of his actions.

Cognitive agents form beliefs representing the world, evaluate the world as represented, form plans for making the world more to their liking, and perform actions executing the plans. Then the cycle repeats. This is the doxastic-conative loop, diagrammed in figure one.1 Both human beings and the autonomous rational agents envisaged in AI are cognitive agents in this sense. The cognition of a cognitive agent can be subdivided into two parts. Epistemic cognition is that kind of cognition responsible for producing and maintaining beliefs. Practical cognition evaluates the world, adopts plans, and initiates action. There is a massive literature both in philosophy and artificial intelligence concerning various aspects of epistemic cognition, and large parts of it are well understood. Practical cognition is less well understood. We can usefully divide practical cognition into five parts: (1) the evaluation of the world as represented by the agent’s beliefs, (2) the adoption of goals for changing it, (3) the construction of plans for achieving goals, (4) the adoption of plans, and (5) the execution of plans. There is a substantial literature in AI concerning the construction and execution of plans, and I will say nothing further about those topics here. This paper will focus on the evaluative aspects of practical cognition. Evaluation plays an essential role in both goal selection and plan adoption. My concern here is the investigation of evaluation as a cognitive enterprise performed by cognitive agents. I am interested both in how it is performed in human beings and how it might be performed in artificial rational agents.