Location: 909 Renaissance Park, Northeastern University (1135 Tremont Street, Boston, MA, 02120)
This conference will be held at Northeastern in 909 Renaissance Park on April 27 & 28. The meeting will bring together researchers in philosophy and cognitive science who are interested in both formal and experimental methods and their applications to descriptive and normative modeling of cognitive phenomena (broadly construed). Featured speakers at this inaugural FAX meeting will include Tania Lombrozo (Princeton Psychology), Laurie Paul (Yale Philosophy), Miriam Schoenfield (MIT Philosophy), Josh Tenenbaum (MIT Cognitive Science), Fiery Cushman (Harvard Psychology), and Thomas Icard (Stanford Philosophy & Cognitive Science).
Abstract: “Reverse engineering the self” Where is the self in a machine that can “think for itself”? Is such a concept necessary for independent intelligence? We think it is, and argue that a model of the self is necessary both for full computational accounts of the human mind, and for any machine that can be claimed to learn and think for itself. Forging links between contemporary AI, cognitive science, and philosophy, we lay out a roadmap for how to reverse-engineer the capacity for self representation in human cognition into an intelligent machine. At the heart of our framework is the ability to self-center: the ability of the agent to represent itself in the world using indexical properties of time, space, environment, and agent. We develop several thought experiments demonstrating the importance of self-centering for flexible thinking and self-correction, and show how it underlies intelligent self-awareness and agency.
Abstract: “How we know what not to think” A striking feature of the real world is that there is too much to think about. This feature is remarkably understudied in laboratory contexts, where the study of decision-making is typically limited to small “choice sets” defined by an experimenter. In such cases an individual may devote considerable attention to each item in the choice set. But in everyday life we are often not presented with defined choice sets; rather, we must construct a viable set of alternatives to consider. We will present several recent and ongoing research projects that each aim to understand how humans spontaneously decide what actions to consider—in other words, how we construct choice sets. A common theme among these studies is a key role for value representations. We will present a conceptually similar series of studies on which objects and events people choose to represent, and consider points of convergence between these case studies.
Abstract: “Can imprecise probabilities be practically motivated?” It is often claimed that in response to certain types of evidence we are rationally required to adopt belief states that are best represented by imprecise probabilities: sets of probability functions, rather than single ones. Imprecise probabilities, or ranges of probabilities (like 60-80%), have also been advocated in the
reporting of scientific information to policy-makers, in engineering, and in the design of artificial intelligence systems. The question motivating this paper is this: should we expect the usage of imprecise probabilities to result in better decision-making than the usage of precise probabilities in these contexts? I will argue that we should not expect imprecise probabilities to do better by undermining the only decision-theoretic motivation for imprecision that has been offered in the literature: ambiguity aversion. If we can’t expect imprecision to help us make better decisions, we need to rethink whether there are good reasons to use imprecise probabilities in contexts in which good decision-making is what’s of primary concern.
Abstract:“Explaining for the best inference” Advocates for inference to the best explanation (IBE) typically argue that explanatory considerations are relevant in evaluating whether a given hypothesis should be accepted over alternatives. I’ll suggest that explanatory considerations do in fact affect human reasoning, and that this influence is often beneficial, but not (necessarily)
because it leads to a preference for hypotheses that are more likely to be true. Instead, I’ll suggest that the practice of seeking and favoring explanations with particular “virtues” is often a good epistemic policy because it supports subsequent learning. This idea is closely related to what Daniel Wilkenfeld and I call “explaining for the best inference” (EBI), an alternative to IBE that focuses on the downstream (epistemic) consequences of the explanatory process.
Abstract: “Causal simulation” Mental simulation is widely implicated in causal cognition. But how should we understand appeal to simulation in accounts of causal reasoning? On one view, it should be construed as part of a “process model” or an algorithmic-level account, specifying details omitted from a proper computational-level account, e.g., one based on structural equations. I propose instead that causal simulation be understood as a distinctive alternative to familiar formal treatments of causation. The proposal is supported by a combination of formal and experimental observations.
Abstract: “Consensus as evidence in folk metaethics” Recent work in folk meta-ethics finds a correlation between perceived consensus about a moral claim and metaethical judgments about whether the claim is universally or only relatively true. We argue that consensus can provide evidence for meta-normative claims, such as whether a claim is universally true. We then report
several experiments indicating that people use consensus to make inferences about whether a claim is universally true. We also examine different forms of relativism. One extreme, if familiar, form of relativism is simple subjectivism, according to which the truth of a claim depends only on the attitudes of the person making the claim. We define a notion of objectivism in contrast to this kind of subjectivism. According to “minimal objectivism”, an objective property is one that is either entirely mindindependent or depends on standards or sensibilities that are shared by some group. Across several studies, we find that people use consensus as evidence that evaluative properties like funny and delicious tend to be treated as minimally objective rather than simply subjective. For instance, people tend to think it’s possible to be mistaken about whether a joke is funny or a dish is delicious. At the same time, people do not regard funniness as a universally objective property. Many of the studies rely on the idea that information about consensus provides evidence regarding minimal objectivity. We show that participants are appropriately sensitive to evidentially relevant characteristics of the sample.