Daniel Grimmer

I am a Postdoctoral Associate and Lecturer in Philosophy at Yale University. My research trajectory spans across disciplines, beginning with Physics (Ph.D., Waterloo) and expanding into the Philosophy of Physics (DPhil, Oxford), Cognitive Science, and Artificial Intelligence.

Currently, my work focuses on Evolutionary Epistemology in silico (recent talk). Remarkably, we can efficiently and faithfully simulate Darwinian evolution using the machine learning technique of Meta-Learning (see this paper). Artificial neural networks can thereby become good analogical models for exploring the evolutionary dynamics which shaped human cognition, including its innate faculties. These new tools should help shed light on the age-old philosophical debate between Nativism and Empiricism as well as the concept of innateness itself. In aid of this program, I have recently derived a suite of advanced optimization algorithms directly from evolutionary first principles (recent paper).

Primary Research Interests

1. Artificial Intelligence & Evolutionary Epistemology. The dominant training regime for neural networks is broadly Empiricist: a general-purpose learning method (gradient descent) extracts patterns and associations from massive amounts of data, beginning from an effectively tabula rasa initialization (i.e., random weights and biases). Rejecting hand-coded innate structures, the Bitter Lesson says: scale, scale, scale! But Nativism is Scalable too: we can faithfully and efficiently simulate Darwinian evolution using meta-learning. We can thus redeploy these former paragons of Empiricism to Nativist ends. In principle, it should be possible to evolve a wide range of different innate structures within neural networks, everything from Kant’s categories, to Jung’s archetypes, and Chomsky’s Universal Grammar. See this paper for further details.

2. Metaphysics of Space and Time. What if we could remove and replace the topological underpinnings of our spacetime theories just as easily as we can switch between different coordinate systems? I claim that we can, by using the ISE Method of topological redescription which I developed in my DPhil thesis. (See this video and this paper.) I claim that these new topological redescription techniques lead us to a conventionalist/neo-Kantian view of spacetime topology which I call the Dynamics-First View of Spacetime Topology (see this paper). For instance, in cases of spacetime-dualities (e.g., AdS-CFT) two different species might evolve radically different spatial intuitions and, relatedly, different ontologies/mereologies. Who then is right about the world’s fundamental metaphysics? My duality-inspired boardgames and videogames build upon this theme.

3. Measurement in Quantum Field Theory (QFT). How should we model quantum measurement processes which involve quantum fields? How must our characterization of QFT’s observables differ from how we characterize the observables of non-relativistic quantum mechanics (NRQM)? Can we model QFT-involved measurement using PVMs and POVMs as we are used to in NRQM? Perhaps surprisingly, we cannot. This gives rise to what I call the Pragmatic QFT Measurement Problem (paper, video abstract).