I am an Associate Professor in Philosophy at the University of Massachusetts, Amherst. I work mainly in Formal Semantics, Pragmatics, Philosophy of Language, Cognitive Science and AI/Machine Learning. Most of my research explores the ways in which our general logical and reasoning capacities enter into various aspects of our linguistic competence. I’ve also worked on related topics in the foundations of cognitive science and AI, including the structure of concepts and their relation to categorization, compositionality and decision making.
More recently, my focus has shifted to interpretability techniques for understanding the inner workings of AI models (I believe this work can also she light on human cognition). I’m working on various topics in alignment and fine-tuning of AI models, focusing on improving the transparency of the ethical and scientific decision making of AI models.
Currently, I’m working on training regimes—inspired by certain insights in the philosophy of science—-that nudge AI models to acquire stronger dispositions to generate abstract, human-interpretable and reliable explanations. I’m also working on different ways of tweaking automated techniques for value alignment to steer AI’s decision-making so as to systematically follow the principles of specific ethical systems.
gdelpinal@umass.edu
