An important module in the Belief-Desire-Intention architecture for artificial agents (which builds on Michael Bratman's work in the philosophy of action) focuses on the task of intention reconsideration. The theoretical task is to formulate principles governing when an agent ought to undo a prior committed intention and reopen deliberation. Extant proposals for such a principle, if sufficiently detailed, are either too task-specific or too computationally demanding. I propose that an agent ought to reconsider an intention whenever some incompatible prospect is (...) sufficiently valuable along some dimension that can be assessed at zero or near-zero computational cost. (shrink)

It would be good to have a Bayesian decision theory that assesses our decisions and thinking according to everyday standards of rationality — standards that do not require logical omniscience (Garber 1983, Hacking 1967). To that end we develop a “fragmented” decision theory in which a single state of mind is represented by a family of credence functions, each associated with a distinct choice condition (Lewis 1982, Stalnaker 1984). The theory imposes a local coherence assumption guaranteeing that as an agent's (...) attention shifts, successive batches of "obvious" logical information become available to her. A rule of expected utility maximization can then be applied to the decision of what to attend to next during a train of thought. On the resulting theory, rationality requires ordinary agents to be logically competent and to often engage in trains of thought that increase the unification of their states of mind. But rationality does not require ordinary agents to be logically omniscient. (shrink)

Information is a force multiplier. Knowledge of the enemy's capability and intentions may be of far more value to a military force than additional troops or firepower. Situation assessment is the ongoing process of inferring relevant information about the forces of concern in a military situation. Relevant information can include force types, firepower, location, and past, present and future course of action. Situation assessment involves the incorporation of uncertain evidence from diverse sources. These include photographs, radar scans, and other forms (...) of image intelligence, or IMINT; electronics intelligence, or ELINT, derived from characteristics (e.g., wavelength) of emissions generated by enemy equipment; communications intelligence, or COMINT, derived from the characteristics of messages sent by the enemy; and reports from human informants (HUMINT). These sources must be combined to form a model of the situation. The sheer volume of data, the ubiquity of uncertainty, the number and complexity of hypotheses to consider, the high-stakes environment, the compressed time frame, and deception and damage from hostile forces, combine to present a staggeringly complex problem. Even if one could formulate a decision problem in reasonable time, explicit determination of an optimal decision policy exceeds any reasonable computational resources. While it is tempting to drop any attempt at rational analysis and rely purely on simple heuristics, we argue that this can lead to catastrophic outcomes. We present an architecture for a ``complex decision machine'' that performs rational deliberation to make decisions in real time. We argue that resource limits require such an architecture to be grounded in simple heuristic reactive processes. We thus argue that both simple heuristics and complex decision machines are required for effective decision making in real time for complex problems. We describe an implementation of our architecture applied to the problem of military situation assessment. (shrink)

Recent claims, mainly from computer scientists, concerning a largely automated and model-free data-intensive science have been countered by critical reactions from a number of philosophers of science. The debate suffers from a lack of detail in two respects, regarding the actual methods used in data-intensive science and the specific ways in which these methods presuppose theoretical assumptions. I examine two widely-used algorithms, classificatory trees and non-parametric regression, and argue that these are theory-laden in an external sense, regarding the framing of (...) research questions, but not in an internal sense concerning the causal structure of the examined phenomenon. With respect to the novelty of data-intensive science, I draw an analogy to exploratory as opposed to theory-directed experimentation. (shrink)

Cognitive Science, Volume 46, Issue 5, May 2022.

Marr's levels of analysis—computational, algorithmic, and implementation—have served cognitive science well over the last 30 years. But the recent increase in the popularity of the computational level raises a new challenge: How do we begin to relate models at different levels of analysis? We propose that it is possible to define levels of analysis that lie between the computational and the algorithmic, providing a way to build a bridge between computational- and algorithmic-level models. The key idea is to push the (...) notion of rationality, often used in defining computational-level models, deeper toward the algorithmic level. We offer a simple recipe for reverse-engineering the mind's cognitive strategies by deriving optimal algorithms for a series of increasingly more realistic abstract computational architectures, which we call “resource-rational analysis.”. (shrink)