I am a Postdoctoral Fellow in Philosophy and Ethics of Artificial Intelligence at the Center for Science and Thought, University of Bonn, and Leverhulme Centre for the Future of Intelligence, University of Cambridge. Previous positions include a stint as Senior Policy Associate at the University of Toronto's Mowat Centre, and Postdoctoral Fellowships at the University of Western Ontario's Rotman Institute of Philosophy, and the University of Tuebingen's Centre for Integrative Neuroscience and Max Planck Institute for Biological Cybernetics. I received my PhD from the University of Pittsburgh in History & Philosophy of Science. I have a MSc in Computer Science from the University of Toronto, where I worked on selection bias in machine learning. I am also active in outreach, including public lectures on AI Ethics, writing for a general audience about data privacy and informed consent, and a long-running collaborative civic engagement project called The Soft City.
My academic research covers topics in the history & philosophy of science, including AI ethics, computational modeling methods,
model organisms, mechanistic explanation, body perception, and classification in psychiatry. I have also done policy research covering big data
and AI in healthcare, informed consent, data governance, and algorithmic decision-making.
Computer Science departments and technology companies are waking up to the dangers that can result from the field's `build first, fix later' ethos, especially when deploying AI systems. I note that many of the strategies being discussed to deal with AI's ethical crisis suffer from the same problems that created the crisis: tech culture's belief that all problems have technical solutions and devaluing of outside expertise. I draw on standpoint epistemology and social epistemology to develop more promising strategies for solving ethical problems in AI.
Philosophers of science have not paid much notice to artificial intelligence. Despite significant interest in computational modeling and simulation methods in a few branches of science like physics, ecology, and climate science, and widespread use of AI tools across many fields, little has been written about AI methods in science. My forthcoming paper in Philosophy of Science clears the ground for a re-integration of philosophy of artificial intelligence into philosophy of science, by revisiting the problem of how connectionist AI can model cognition with unrealistic brain models. I am currently working on papers about the biases inherent in collaborative filtering algorithms, and adversarial examples in Deep Learning.
There are several large scale computational psychiatry projects on the horizon, with goals of discovering risk factors, new treatments, and more effective diagnostic tools, using machine learning methods. A preliminary step is to build an ontology of psychopathology to organize the data. Given that the majority of computational psychiatry studies so far focus on biological risk factors, it is likely that the ontology will be scaled up from existing biomedical ontologies, leaving it open how higher level psychological and social constructs will be integrated. In this paper I lay out some desiderata for how that integration should occur, so as to allow for the accurate representation of cross-cutting, multi-level causes of mental health. I draw on case studies from past ontology projects to illustrate what can go wrong, and how to avoid similar problems.
Historians emphasize how choices of model organisms are made on pragmatic grounds or due to contingent factors, while many scientists and philosophers claim that model organisms ought to be chosen based on phylogenetic relatedness. I examine a variety of examples of model organism, including plants, and argue that i) phylogenetic relatedness is a poor guide for choosing model organisms, and ii) in many cases the apparently pragmatic factors guiding model choice have an epistemic character.
Scientists treat abstractions like network structures or populations as though they are important bearers of causal powers. Philosophers tend to think of abstractions as mere representations, which are not the right kinds of things to bear causal power. This project carves out a metaphysical option that better expresses scientists' views, while not falling into the Third Man problem. This is a joint project with .
Idealization in Computational Models
In a 2018 chapter about Explanation and Connectionist Models in The Routledge Handbook of the Computational Mind, I outline the computational modeling methods used in cognitive science, and contrast them with the simulation methods used in the physical sciences, as well as with the more detailed brain simulations sometimes used in neuroscience. I argue that connectionist models of cognition explain by instantiating idealized neural models, and demonstrating their properties.
Body Perception in Psychiatry
In a paper in Synthese (2019), I demonstrate that treatment and research of Anorexia Nervosa (AN) largely overlook one of its three diagnostic criteria---a disturbance of body perception---despite evidence that this symptom might be central to AN etiology and treatment outcomes. I review the history of revisions to the Eating Disorders category of the Diagnostic and Statistical Manual of Mental Disorders (DSM). This history suggests that the assumption that AN is an Eating Disorder may be responsible for AN's body perception symptoms being overlooked, and perhaps also partly responsible for AN's tragically high relapse and mortality rates. I propose a change to the DSM taxonomy allowing for disorders to be listed under multiple categories.
With several , I worked on a series of experiments investigating whether the Rubber-Hand Illusion (RHI) paradigm might be useful for changing perceptions of the size of one's body. In the first study, using Virtual Reality, we induced a full body RHI on healthy participants, effectively changing their perceptions of the size of their bodies, at least in the short term. This study was published in PLOS One in 2014. Another group recently showed that this effect can be maintained for several hours in both participants with AN and healthy controls, suggesting that it might be a useful paradigm for treating the disturbance of body perception in AN.
Ehrsson, Holmes, and Passingham (2005) introduced a non-visual version of the RHI. I ran an experiment demonstrating that participants can induce a similar tactile version of the RHI on their own body, without guidance from an experimenter. We compare active (subject controlling the brush) and passive (experimenter controlling the brush) touch on objective and subjective messures of the RHI. Preliminary analysis of the results shows i) subjects can independently induce a tactile RHI on their own body, and ii) the active variant of the RHI is stronger than the passive version. These results suggest a way of developing home therapies that might help modify the disturbance of body image characteristic of Anorexia Nervosa.
Mechanistic Explanation in Neuroscience
My dissertation was about how to integrate explanations in cognitive psychology and neuroscience. I defended a non-hierarchical view of the relations between models pitched at different levels of grain and abstraction. On this account, each model might express only some of the causes at play, or part of the mechanism. Multiple models do not fit together neatly like pieces of a puzzle or mosaic, but instead partially overlap in patchy ways.
In a paper published in Synthese (2016), I argued against popular accounts of integration that suggest that psychology and neuroscience can be seamlessly integrated by taking psychological explanations to be sketches of neural mechanisms.
In a paper co-authored with , we used several historical case studies in neuroanatomy, and neurophysiology to illustrate how explanations develop from sketchy, primitive ideas to detailed mechanistic accounts. We highlighted several issues that have not received adequate attention elsewhere, including how top-down and bottom-up methods need to be combined in order to discover how structure and function relate; alternatives to the unrealistic expectation that explanations in neuroscience will all fit into a neat hierarchy of levels; and how the use of model organisms and varying experimental protocols complicates this research. This paper appeared (2017) in The Routledge Handbook of Mechanisms and Mechanical Philosophy.
In a paper for Peter Machamer's Festschrift (2017, Springer), I examine the notion of a mechanism schema. Peter has suggested that Piaget's theory of child development was his inspiration for mechanism schemata. I compare mechanism schemata to Piagetian schemata to draw out some features of mechanism schemata that have been overlooked in the new mechanist literature.
I am currently teaching philosophy and ethics of AI in the University of Bonn's Philosophy Department. I have
also taught in cognitive science, history and philosophy of science, computer science, and neuroscience departments.
Select Past Courses