Workshop on Functions and Representations in AI Systems

16th–17th November 2026
Copenhagen, Denmark (Carlsberg Academy)

The concepts of function and representation are widely invoked in the cognitive and life sciences. An influential philosophical tradition, teleosemantics, proposes to ground representation partly in biological functions and ground functions in historical selection processes. While this remains a leading approach, the account has faced challenges, ranging from the notorious "swampman" objection, to more recent disputes about the nature of functional and representational explanation in scientific practice.

Independently, the recent wave of developments in artificial intelligence has sparked questions about the representational capacities of AI models, e.g. what, if anything, can large language models represent? The emerging field of mechanistic interpretability is replete with function- and representation-talk. But is this talk justified? Are AI models subject to selection processes, in the proper sense? And are the concepts of representation and function deployed here the same as those invoked in the life sciences? Are standard philosophical frameworks adequate for the case of AI? Or should we look elsewhere, e.g. to theorising about artefacts? And what role, if any, do human norms and intentionality play in grounding AI functions and content?

A workshop on these issues, co-funded by the Pioneer Center for Artificial Intelligence (https://www.aicentre.dk/p1-programs/functions-and-representations-in-ai-systems) and the Center for Philosophy of Artificial Intelligence (University of Copenhagen) (https://cpai.ku.dk/), will be held in Copenhagen, November 16th–17th, in the beautiful Carlsberg Academy building (https://www.carlsbergfondet.dk/en/about-the-foundation/our-buildings/the-carlsberg-academy/).

Preliminary program:

Day 1: Monday 16th November

9:00–9:30

Registration & morning coffee

9:30–10:30

Keynote talk – Marc Artiga (Valencia) – “What grounds functions in Large Language Models?”

10:30–10:45

15 minute break

10:45–11:30

Fabian Hundertmark (Valencia) – AI Frogs, Proper Functions, and Internal Representations - with James Turner (Duisburg-Essen)

11:30–12:15

Iwan Williams (Copenhagen) – Doing without historical functions for representational explanations of AI

12:15–13:45

Lunch

13:45–14:30

Eliot Du Sordet (Neuchâtel) – Selection Histories and Representational Targets: Pre-Trained, Fine-Tuned, and Agentic LLMs

14:30–15:15

Louise Daoust (Eckerd) – Ascriptions of Representation and Function in Artificial Intelligence Research

15:15–15:30

15 minute break

15:30–16:15

Jumbly Grindrod (Reading) – Sparse Auto-Encoders and Holism about LLMs

16:15–16:30

15 minute break

16:30–17:30

Keynote talk – Rosa Cao (Stanford) – Alien meanings and functions without history

Day 2: Tuesday 17th November

9:15–9:45

Registration & morning coffee

9:45–10:30

Annika Schuster (TU Dortmund) & Nina Poth (Radboud) – Deflating Artificial Representations

10:30–11:15

Augustas Pinkevičius (Vytautas Magnus) – Large Language Models and the Limits of Teleosemantics in the Grounding Problem

15 minute break

11:30–12:15

Michela Ghezzi (Eindhoven/Cambridge) – The epistemic basis of AI evaluation: from behavior to mechanism

12:15–13:45

Lunch

13:45–14:30

Alessandro Corona Mendozza (Aarhus) – Severe Testing for Representational Claims in Neural Networks

14:30–15:15

Cameron Yetman (Toronto) – The Platonic Representation Hypothesis: Variations on a theme

15:15–15:30

15 minute break

15:30–16:15

Fintan Mallory (Durham) – (Title TBD)

16:15–16:30

15 minute break

16:30–17:30

Keynote talk – Dimitri Coelho Mollo (Umeå) – Learning for machine learning systems - with James Turner (Duisburg-Essen)

Please feel free to forward this information to anyone who you think might be interested in attending the workshop.

Attendance is free but registration is required. You can register here: https://docs.google.com/forms/d/e/1FAIpQLScVu6hdujp-jZ1UH7-DuccfEQOorf-icTIGn1roPJjHhzVULQ/viewform