Research

I am a philosopher of science specializing in the philosophical foundations of cognitive science. My research focuses on two sets of related issues. The first concerns the nature of scientific evidence and how cognitive scientists can develop high-quality evidence for cognitive models. Much of my work focuses on scientists using assumptions about the rationality of a system to provide evidence for cognitive models—specifically, in an approach known as resource rational analysis (Lieder and Griffiths, 2020). My work analyzes how this strategy has worked, how it should work, why it will work, and why it can work better with the conceptual foundations I develop. Toward this end, the second set of issues I am focused on concerns normativity and cognition. My research develops a framework of resource rationality to answer questions such as how to evaluate how well cognitive systems make use of limited resources, how cognitive limitations affect epistemic norms, the extent to which coming to know a system’s intentional states requires attributing rationality to that system, and related questions.


Resource Rationality

Resource rationality is rationality relative to constraints. The framework of resource rationality that I have developed involves a maximally broad notion of what can count as a constraint. This makes all systems trivially resource rational, as all systems can be understood to be doing their best relative to a sufficiently expansive set of constraints. Much of my research involves showing why this previously undefended and prima facie strange view is both well-motivated and provides a useful foundation for evaluating, prescribing, and studying psychological beings.  See my dissertation. 


Cognitive Limitations and Epistemic Norms (under review)

I argue that there is no principled way to distinguish between agents being rational relative to their limitations and agents being irrational. Making this distinction requires differentiating between cognitive constraints that lead to irrationality and those that do not, and this cannot be done in a satisfactory way. Similar lines of thought have previously been used to argue for the primacy of ideal epistemology (Carr, 2022). But I argue that the distinction between ideal and non-ideal epistemology falls apart for the same reason; supposedly ideal (inductive) epistemic norms subtly invoke cognitive limitations, and these limitations surprisingly cannot be disentangled from other kinds of cognitive limitations. Thus no line can be drawn between ideal agents, resource rational agents, and irrational agents. I then develop a framework for how to epistemically evaluate agents in light of these conclusions. An agent's rationality is not fully brought into view until it is shown how the agent is doing the best they can relative to their constraints. This includes specifying all psychological facts that prevent them from doing better and characterizing how mutable these facts are. Comparing agents’ rationality is done as follows: one agent is more rational than another if the former agent’s constraints are a proper subset of the latter’s. Agents subject to qualitatively different constraints are not comparable ceteris paribus. This rational comparison relation induces a partial order—a feature that is independently motivated and proves useful for resource rational analysis.


The Logic of Theory-Testing In Resource Rational Analysis (draft available)

In this paper, I look at resource rational analysis as a methodological strategy in cognitive science. Resource rational analysis develops and tests models of cognition by assuming agents are as rational as possible given their limited cognitive resources. Psychologists start by comparing observable human behavioral data to theoretical models of rational performance. Discrepancies between the theoretical and the observed behavior indicate cognitive resource constraints. Constraints get incorporated into a new theoretical model of (resource) rational performance. New discrepancies emerge and the process iterates.


Resource-Rational Analysis:


The main purpose of this paper is to analyze the logic of theory-testing instantiated by this methodology and provide an epistemic justification for such a testing strategy. The idea is that in early stages of inquiry into a complicated black box such as human cognition, problems of underdetermination of theory by evidence are exacerbated, and this situation makes hypothetical induction (confirmation of theories via agreement of predicted and observed observations) particularly ineffective. I argue that resource rational analysis avoids such difficulties by instantiating a logic of theory-testing distinct from hypothetical induction—namely a dynamic testing strategy that Smith (2014), looking at the history of Newtonian gravity research, has called “closing the loop.” 


In addition to establishing the descriptive claim that resource rational analysis instantiates the closing the loop logic, I offer an epistemic justification for this research strategy. I do so by analyzing closing the loop as a synergistic combination of demonstrative induction and neo-Popperian falsification. Demonstrative induction is deductive inference from observable data to theoretical claims (Norton, 1993). In resource rational analysis, one deduces claims about cognition from characterizations of laboratory task environments, mediated by the assumption that humans are resource rational (note that adopting the trivial notion of resource rationality allows this mediating assumption to be true and not an empirical question). Neo-Popperian falsification involves using topology to characterize degrees of falsifiability and justifying methods that result in progressive falsification (Kelly, 1996; 2024). Using formal learning theory, I show that models that suppose humans are more rational (that is, models that posit fewer constraints) are more falsifiable and that iteratively de-idealizing rational models of cognition constitutes a neo-Popperian progressive falsification method. This shows that psychological facts qua constraints are learnable in the formal sense. In closing the loop, demonstrative induction and neo-Popperian falsification are both necessary to manage the problems of undetermination and avoid the problems of hypothetical induction. This analysis provides a justification for resource rational analysis and shows why it is poised to overcome the unique obstacles to providing high-quality evidence that cognitive scientists face.


Normative Commitments In Artificial Intelligence (under review)

This paper explores how to empirically investigate and engineer normative commitments in AI systems.  I express skepticism toward probing AI systems' internal states to identify mental states and propose that the best approach is to articulate behavioral conditions for the presence of normative commitments. Normative commitments are rich behavioral capacities that, following Bilgrami (2008) and others, are not reducible to dispositions or higher-order dispositions. I use my account of resource rationality to characterize a notion I call meta-reflection (related but importantly different from meta-induction or meta-cognition), which is a property of a system such that intervening on some of its constraints results in the system maintaining itself as resource rational. Where resource rationality is trivial, meta-reflection is highly non-trivial. I argue that such a capacity provides necessary and sufficient conditions for the presence of normative commitments in a system and also provides a way to characterize different “levels” of commitment. Since meta-reflection can be empirically investigated, it provides a way to behaviorally study normative commitments in AI systems and non-human animals. 


Intentionality and the Rationality Assumption (draft available)

This paper looks at intentionality and the rationality assumption—in particular, the view that attributing intentional states to a system requires attributing rationality to that system (e.g., Dennett, 1989; Davidson, 1995). I argue that the rationality assumption for intentionality should be understood in terms of my notion of resource rationality. Arguments against the rationality assumption, such as Stich’s (1985), do not defeat the reasons for thinking intentionality presupposes rationality—they just show that attributions of irrationality to intentional systems are possible. Thus intentionality does not presuppose ideal rationality. I argue that irrationality can be attributed by specifying constraints on rationality in the form of psychological details. However, the rationality presupposed by intentionality cannot be minimal rationality. I argue that until an agent's resource rationality is brought into view—that is, how they perform optimally relative to their constraints—it is not possible to fully specify the extent to which they possess intentional states. Thus, intentionality presupposes specifically resource rationality.


Isaac Newton's Interdisciplinary Research Methods (under review)

This paper examines Newton’s evidential reasoning in his chronological studies.  I identify a pattern of evidential reasoning in his chronological works, in which Newton “exports” inductive risk to areas outside of chronology in order to arrive at more certain results in his historical studies. I look at Newton’s various sources for his chronological research, including primary and secondary historical documents, biology, social science, and astronomy, and analyze how Newton used these domains to turn historical data into evidence for his chronological scheme.  I argue that Newton employed what has been called demonstrative induction in chronology.  In demonstrative induction, the inductive risk is confined to the premises.  The strength of this kind of inference therefore depends on the strength of the premises.  I further argue that Newton placed restrictions on the kinds of premises that could enter into his demonstrative inductions. In particular, Newton required that premises needed to be supported by inductive generalization.  Further, Newton relied as much as possible on premises that were not hypotheses about chronology—e.g., matters of dating and ordering events in the past—but pertained to other domains of inquiry, such as textual interpretation and astronomy.  The result is an analysis that shows in what way Newton avoided hypotheses in chronology.


History of Mathematical Psychology Book Project

I am currently writing a book with Colin Allen, Nuhu Osman Attah, Mara McGuire, Dzintra Ullis on the history of this research program that took formation in the 1950's and 1960's and continues today, in part in the form of the Society and the Journal of Mathematical Psychology. We have been performing interviews with important figures from mathematical psychology and reading historical materials. Watch our keynote address at MathPsych/ICCM 2021: Three Questions about Mathematical Psychology


ALIUS Research Group

I am one of three coordinators of ALIUS. ALIUS is an international and interdisciplinary research group dedicated to the investigation of all aspects of consciousness, with a specific focus on non-ordinary or understudied conscious states traditionally classified as altered states of consciousness. One of the main outputs of ALIUS Is the annual ALIUS Bulletin. See my interview with Daniel Dennett that I did with Daniel Friedman.


References

Bilgrami, A. (2008). Intentionality and norms. In M. De Caro & D. Macarthur (Eds.),Naturalism 

in Question (pp. 125-152). Harvard University Press.

Carr, J. R. (2022). Why ideal epistemology?. Mind, 131(524), 1131-1162.

Davidson, D. (1995). Could there be a science of rationality? International Journal of 

Philosophical Studies, 3(1), 1-16.

Dennett, D. C. (1989). The intentional stance. MIT Press.

Kelly, K. (1996). The logic of reliable inquiry. Oxford University Press, USA.

Kelly, K. (2024, February). The topology of scientific inquiry [Unpublished manuscript], 

Carnegie Mellon University.

Lieder, F., & Griffiths, T. L. (2020). Resource-rational analysis: Understanding human cognition 

as the optimal use of limited computational resources. Behavioral and brain sciences, 43.

Norton, J. D. (1993). The determination of theory by evidence: The case for quantum 

discontinuity, 1900–1915. Synthese, 97, 1-31. 3

Smith, G. E. (2014). Closing the loop. Newton and empiricism, 262-352.

Stich, S. P. (1985). Could man be an irrational animal? Some notes on the epistemology of rationality. Synthese, 

115-135.