The Philosophy of Deep Learning
Summary
The conference explored current issues in AI research from a philosophical perspective, with particular attention to recent work on deep artificial neural networks. The goal was to bring together philosophers and scientists who are thinking about these systems in order to gain a better understanding of their capacities, their limitations, and their relationship to human cognition.
The conference focused especially on topics in the philosophy of cognitive science (rather than on topics in AI ethics and safety). It explored questions such as:
- What cognitive capacities, if any, do current deep learning systems possess?
- What cognitive capacities might future deep learning systems possess?
- What kind of representations can we ascribe to artificial neural networks?
- Could a large language model genuinely understand language?
- What do deep learning systems tell us about human cognition, and vice versa?
- How can we develop a theoretical understanding of deep learning systems?
- How do deep learning systems bear on philosophical debates such as rationalism vs empiricism and classical vs. nonclassical views of cognition.
- What are the key obstacles on the path from current deep learning systems to human-level cognition?
A pre-conference debate tackled the question “Do large language models need sensory grounding for meaning and understanding ?”.
Speakers
Pre-Conference Debate
- Jacob Browning (New York University)
- David Chalmers (New York University)
- Brenden Lake (New York University)
- Yann LeCun (New York University / Meta AI)
- Gary Lupyan (University of Wisconsin–Madison)
- Ellie Pavlick (Brown University / Google AI)
Lectures
- Cameron Buckner (University of Houston)
- Rosa Cao (Stanford University)
- Grace Lindsay (New York University)
- Raphaël Millière (Columbia University)
- Nicholas Shea (Institute of Philosophy, University of London)
Panel on Deep Learning and Cognitive Science
- Ishita Dasgupta (DeepMind)
- Nikolaus Kriegeskorte (Columbia University)
- Tal Linzen (New York University / Google AI)
- Robert Long (Center for AI Safety)
- Ida Momennejad (Microsoft Research)
Symposia
- Nuhu Osman Attah (University of Pittsburgh)
- Patrick Butlin (University of Oxford)
- Tony Chen (MIT)
- Jacqueline Harding (Stanford University)
- Anna Ivanova (MIT)
- Fintan Mallory (University of Oslo)
- Mitchell Ostrow (MIT)
- Anders Søgaard (University of Copenhagen)
- Philippe Verreault-Julien (Eindhoven University of Technology)
- Cedegao Zhang (MIT)
Poster presentations
- Atoosa Kasirzadeh (University of Edinburgh)
- Wai Keen Vong (New York University)
- Sreejan Kumar (Princeton University)
- Will Merrill (New York University)
- Julia Minarik (University of Toronto)
- Jared Moore (University of Washington)
- Nedah Nemati (Columbia University)
- Emin Orhan (New York University)
- Stephan Pohl (New York University)
- Hokyung Sung (MIT)
- Justin Tiehen (Puget Sound)
Video recordings
Recordings of the whole conference are available online.