Philosophers have long thought deeply about many concepts that are used colloquially in the machine learning (ML) community such as epistemology, counterfactuals, explainability, reliability, uncertainty and causality. As ML systems are now embedded in high-stakes decisions across science, industry, and public life, it is urgent that when ML researchers claim properties such as "explainability", "reliability", "intelligence" or "cognition", these claims are made with awareness of what practitioners, policymakers, and affected users mean by those terms. In particular, we argue that the ML community needs to take a step back and review whether the mathematical objectives used in optimisation and evaluation procedures truly take into account how philosophers have analysed them—analyses that explicitly aim to connect notions like explanation, evidence, and uncertainty to human understanding, justification, and use.

Philosophers of science and psychologists are more actively engaged than ever in such questions; however, their interaction with ML researchers remains sparse and fragmented. The goal of the proposed workshop is to facilitate a lively dialogue between the two otherwise largely separate communities, to promote more principled and grounded advances in ML and artificial intelligence.

• 11th May: Paper submission deadline

• 31st May: Notification of decision

• 11th July: Workshop

The workshop will take place from 8am to 5pm on Saturday 11th July, 2026 at Coex, Seoul, South Korea. 

• 8:20-8:30 Opening remarks

• 8:30-9:05 Invited talk: Naftali Weinberger

Title: Surrogate Metric Evaluation is a Causal Inference Problem

Abstract: Many AI systems are evaluated and trained using surrogate metrics that stand in for quantities of ultimate interest. Analyses of surrogate trustworthiness often focus on whether the surrogate and the target quantity are correlated, but this leaves open whether improvements in the surrogate will continue to improve the target under increasing optimization pressure. In my talk, I argue that this question can be conceptualized and addressed using dynamic causal models of feedback control systems. Using the classic example of the Watt governor, I illustrate how such systems exploit feedback loops to create higher-scale relationships by which one quantity adjusts to respond to another (e.g. the governor adjusts steam supply to match steam demand). Surrogate optimization similarly employs feedback in an attempt to make improvements in the surrogate correspond to improvements in the target, and optimization ceases to be useful once proposed improvements to the surrogate fail to match improvements in the target. It is this “matching” relationship, rather than the correlation between surrogate and target, that is relevant for analyzing optimization failures, reward hacking, and Goodhart-style effects.

• 9:05-9:40 Invited talk: Misha Belkin

Title: 75 years of Turing’s AGI

Abstract: I argue that the long-standing problem of building human-level machine intelligence has been solved. 

• 9:40-10:00 Coffee break

• 10:00-10:35 Invited talk: Raphaël Millière

Title: A Bayesian epistemology for LLM evaluation

Abstract: Evaluating the capacities of large language models (LLMs) requires inferring latent abilities from observable task performance. We propose to formalize such inferences within a Bayesian framework, drawing on the epistemology of comparative cognitive science. The framework makes explicit how posterior credences about a system's abilities depend on prior probabilities over hypotheses about computational strategies, likelihood functions shaped by task demands, and marginalization over auxiliary factors that influence performance independently of the target ability. Within this framework, we argue that the relative evidential weight of behavioral versus mechanistic evidence for algorithmic-level hypotheses concerning the representations and computations underlying task performance depends systematically on background knowledge about the target system. For human cognition, decades of research in cognitive science, neuroscience, and evolutionary biology provide strong priors that constrain the space of plausible computational strategies, while well-characterized auxiliary factors (attention, motivation, working memory limits) allow informative marginalization. Behavioral paradigms consequently provide substantial evidential leverage for discriminating between algorithmic hypotheses. For LLMs, this evidential hierarchy shifts: weak priors about learned computational strategies and poorly characterized auxiliary factors (e.g., sensitivity to prompt formatting, tokenization artifacts, training distribution) make behavioral evidence systematically underdetermining across competing hypotheses. Mechanistic interpretability, by contrast, provides more direct access to representations and computations, offering evidence that can discriminate where behavioral evidence cannot.

• 10:35-11:10 Invited talk: Been Kim

Title: Thinking about humans in the era of AI

Abstract: Non-philosopher’s take on what AI era means to humans. 

• 11:10-12:00 Get lunch! 

• 12:00-13:10 Poster session (Hall A, 3201–3217)

• 13:10-13:25 Break

• 13:25-14:00 Invited talk: Cameron Buckner

Title: “Reasoning” models, philosophy of inference, and artificial epistemic agency

Abstract: The frontier of AI is being pushed forward now by “Large Reasoning Models” (LRMs) that self-generate long textual “Chains-of-Thought” (CoTs) before answering user queries. While there is already a cottage industry investigating whether these chains of thought are “faithful” to underlying computations producing answers, existing faithfulness research has largely analyzed faithfulness in purely behavioral terms, as a particular input-output profile on reasoning problems. More broadly, there has been little attempt in this literature to define reasoning or to distinguish rational inference from other forms of decision-making. By contrast, philosophical work on rational inference has tended to focus on internal psychological connections between evidence and conclusions—in the philosophical tradition, rational inference involves the exercise of a distinctive form of epistemic agency to construe evidence in particular ways and to draw conclusions because of these construals. In this talk, I will contrast the behavioral and philosophical approaches to rational inference. I will argue that while the philosophical tradition runs the risk of overintellectualizing inference, there are some pragmatic reasons for engineers to review the richer philosophical notion of reasoning when designing the next generation of artificial agents with more powerful forms of epistemic self-calibration and agency.

• 14:00-14:50 Oral presentations (6 x 7 minutes, followed by 7 minutes of joint Q&A)
  1. Phongsakon Mark Konrad, Toygar Tanyel, Serkan Ayvaz
      Self-Reports Do Not Identify Self-Models: An Identifiability Test for Counterfactual Reports
  2. Kola Ayonrinde, Raphaël Millière
      Getting Monosemantic About Monosemanticity
  3. Amine M'Charrak, Thong Pham, Thomas Lukasiewicz, Yuxiao Dong, Shohei Shimizu
      Before Normative and Moral Alignment: Causal Contract Faithfulness as a Precondition for Trustworthy AI
  4. Gilad Landau, Aviv Keren
      The Concept of Representation in ML: Beyond Plato and Aristotle
  5. Louis Mahon, Elliot Ford, Callum Hackett
      A Definition of Good Explanations and the Challenges Explaining LLM Outputs
  6. Joseph Keshet
      Why Sampling Is Not Choosing: Intentionality, Agency, and Moral Responsibility in Large Language Models

• 14:50-15:25 Coffee break

• 15:25-16:00 Invited talk: Alison Huang

Title: (How) Does Accountability Require Understanding ML Models?

Abstract: Concerns about artificial intelligence—particularly systems based on machine learning models whose internal operations are opaque to human understanding—are frequently framed as concerns about accountability. Yet the concept of accountability itself is often left underspecified, encompassing a heterogeneous set of issues that call for distinct responses. This paper focuses on a specific subset of accountability concerns: the “who” questions. We distinguish between two questions: When an AI system causes harm, who is to blame, and who bears the obligation to provide compensation? We argue that the answers to these two questions about accountability require different information and do not entail one another. Then we investigate whether the opacity of contemporary machine learning models undermines our capacity to hold the relevant actors accountable in these two different senses of accountability, and how to rethink efforts aimed at increasing understanding.

• 16:00-17:00 Panel discussion & Closing

• Self-Reports Do Not Identify Self-Models: An Identifiability Test for Counterfactual Reports
  Phongsakon Mark Konrad, Toygar Tanyel, Serkan Ayvaz

• Getting Monosemantic About Monosemanticity
  Kola Ayonrinde, Raphaël Millière

• Before Normative and Moral Alignment: Causal Contract Faithfulness as a Precondition for Trustworthy AI
  Amine M'Charrak, Thong Pham, Thomas Lukasiewicz, Yuxiao Dong, Shohei Shimizu

• The Concept of Representation in ML: Beyond Plato and Aristotle
  Gilad Landau, Aviv Keren

• A Definition of Good Explanations and the Challenges Explaining LLM Outputs
  Louis Mahon, Elliot Ford, Callum Hackett

• Why Sampling Is Not Choosing: Intentionality, Agency, and Moral Responsibility in Large Language Models
  Joseph Keshet

• Proleptic Epistemology for Societal Impacts of AGI
  Priyansh Singhal, Sandeep Kumar, Piyush Joshi

• The Opacity of Descent: Optimization, Epistemic Asymmetry, and the Semantics of Convergence in Deep Learning
  Mahdi Ghaznavi

• Taxonomizing Arguments for Language Model Skepticism in the Language of the Theory of Computation
  Michael Guerzhoy

• Fair Learning with Biased Labels: When Observed Accuracy Is the Wrong Target
  Heng-Chien Liou, I-Hsiang Wang

• Reliability, Faithfulness, and the Limits of Post-hoc Explanations of Opaque Scientific Models
  Nick Oh, Helen Jin

• From Observation to Intervention: A Causal Audit of Expert Importance in Mixture-of-Experts Models
  Leonard Engmann, Christian Medeiros Adriano, Holger Giese

• Mistakes as Epistemic Signatures: An Efficiency-Modulated Cumulative Error Framework for Comparison and Diagnosis of AI Errors
  Darshini N

• Lifted Representation Hypothesis in Language Models
  Bumjin Park, Jaesik Choi

• Explanation in an Emerging Science of Large Language Models
  James Ming Liang Ang

• On Epistemic Diversity in Large Language Models
  Elisabeth Kirsten, Nicole C. Krämer, Muhammad Bilal Zafar

• Dignity as Answerability: How World-Model AI Reframes Human Moral Standing
  Junghoon Justin Park, Jiook Cha

• Operative Contexts: Belief Revision and Memory in Agentic AI
  Emma Cabalé, Philippe Beraud, Philippe Limantour

• A Relativistic Perspective of Reliability in Machine Learning
  Rajeev Verma

• When Belief Bends to Belief: Sycophancy as a Single-Layer Truth–Compliance Tension in LLMs
  Valentin NOËL

• Can LLMs Navigate Beliefs and Facts? Depends on How You Phrase It
  Quang Minh Nguyen, Luis Frentzen Salim

• Beyond Accuracy: Epistemic Justification in Trustworthy Machine Learning
  Poojak Patel, Maneth Perera

• An Evolutionary Epistemology of Post-Training
  Nicholas Clark, Bill Howe

• Trust as Predictive Precision: Reliability and Influence in Representation Alignment
  Hidenori Tanaka

• Explanation for Whom? Hospitable Interpretability for Machine Learning
  Abutalib Namazov

• Constituting What Counts: A Phenomenological Approach to Human-AI Ontological Translation
  Prerna Luthra, Manojshyaam C J

• Online Boundary-Aware Memory for Case-Based Reasoning Agents
  Zheng Dong, Luming Shang

• When Do Transformer Components Compose? Validating a Log-Pool Decomposition Criterion
  Junyu Ren, Su Hyeong Lee, Risi Kondor

• From Prompts to Proof Obligations: Formal Sidecars as an Epistemic Interface for Trustworthy ML
  Junyu Ren

• Savage Without Monotonicity
  Shuo Li Liu, Jingni Yang

• Explanations are a Means to an End: Decision Theoretic Explanation Evaluation
  Ziyang Guo, Berk Ustun, Jessica Hullman

• Interpretability Should Prioritise Use-Inspired Basic Research for AI Safety
  Kola Ayonrinde

• Fictionalism about Personas: Folk Psychology as an Interpretability Strategy
  Weiming Sheng

• AI Review Is a Systemic Risk to Peer Review: Toward a Blockchain-Supported Claim-Level Ledger for Accountability
  Yibo Miao, Yichi Zhang, Yinpeng Dong

• On the Detectability of LLM-Generated Text: What Exactly Is LLM-Generated Text?
  Mingmeng Geng, Thierry Poibeau

• AI Wellbeing: Measuring and Improving the Functional Pleasure and Pain of AIs
  Richard Ren, Kunyang Li, Mantas Mazeika, Wenyu Zhang, Yury Orlovskiy, Rishub Tamirisa, Wenjie Jacky Mo, Thuy Dung Nguyen, Long Phan, Steven Basart, Austin Meek, Aditya Mehta, Oliver Ingebretsen, Alice Blair, Brianna Adewinmbi, Vy Phan, Alice Gatti, Adam Khoja, Jason Hausenloy, Devin Kim, Dan Hendrycks

• Vision-Language Asymmetry in Bistable Image Captioning
  Arohan Agate

• Measuring the Ruler: Reading Benchmark Saturation as Evidence
  Sebastian M Schmon

• Do LLMs Really Represent the World? A Teleosemantic Assessment of Pre-Trained, Fine-Tuned, and Agentic LLMs
  Eliot Du Sordet

• Where Does Prediction Error Come From When the Data Is Perfect? A Decomposition of the Model–World Gap in Predictive Uncertainty
  Johanna Einsiedler, Rosa Lavelle-Hill, Constantin T. A. Wiegand

• Reconciling Causality and Non-Equilibrium Thermodynamics with Hamiltonian Causal Models
  Dario Rancati, Max Welling, Francesco Locatello

• Towards Automated Evaluation of Socio-Technical Harms in LLMs: A Normative Taxonomy and Multi-Turn Red-Teaming Framework
  Byeongho Lee, Hyundeuk Cheon

• Explaining What Machine Learning Learns through Explainable AI
  Jinyeong Gim

• Procedural Generalization: A Resource-Sensitive Account of Knowing-How
  Tomer Galanti, Saharsh Koganti, Priyadarsi Mishra, Pierfrancesco Beneventano

• The Wrong Question? Artificial Consciousness and the Politics of AI Agency
  Thierry Poibeau

• Unsafe Consensus in Diagnostic Deliberation
  Yuting Yan, Yinghao Fu, Haozhou Gao, Tianjian Zhang, Aoxi Liu, Shuang Li

• Uncertainty as Perceptual Testimony in Vision-Language Models
  Ahmad A Rushdi

• Factuality Beyond Reference in LLMs
  Thierry Poibeau

• Reality and Practice: A Relational Reading of the Platonic Representation Hypothesis
  Sebastian M Schmon

• Privileged Self-Access Matters for Introspection in AI
  Siyuan Song, Harvey Lederman, Jennifer Hu, Kyle Mahowald

• Epistemic Misalignment in Human-AI Systems: A Four-Quadrant Taxonomy of Uncertainty
  Mayank Kejriwal

• Can Standard MARL Metrics Distinguish Communicative from Strategic Action?
  Majid Ghasemi, Mark Crowley

• Efficient Counterfactual Reasoning in ProbLog via Single-World Intervention Programs
  Saimun Habib, Vaishak Belle, Fengxiang He

• DeepSWIP: Single-World Counterfactual Semantics for DeepProbLog
  Saimun Habib, Vaishak Belle, Fengxiang He

• The Hawk Effect: Why We Need a Two-Dimensional Measure of Machine Intelligence
  Fryderyk Kuzma

• Counterfactuals Without Worlds: When ML Counterfactual Explanations Are Ill-Posed
  Muhammet Anil Yagiz

• Reliable for Whom? Directional Reliability in AI-Mediated Political Dialogue
  Jaeyoun You

• Noticing the Watcher: LLM Agents Can Infer CoT Monitoring from Blocking Feedback
  Thomas Jiralerspong, Flemming Kondrup, Yoshua Bengio

• Trustworthiness and co-cognition in artificial intelligence systems
  Silvère Gangloff

• Articulate Intuition or Genuine Analysis? Benchmarking Epistemic Reliability in LLM-as-a-Judge Peer Review
  Nuo Chen, Bingsheng He

Submissions are now closed. Thank you for your interest in PhilML workshop! 

We invite short paper submissions (up to 4 pages, excluding references and appendix) from both philosophers and ML researchers on the following topics:

• Epistemology of learning systems: knowledge, belief, evidence, justification, understanding, etc.

• Uncertainty: interpretations of probability and credence; confidence, ignorance, ambiguity, etc.

• Counterfactual reasoning: when counterfactual questions are well-posed, and what makes counterfactual answers meaningful.

• Foundations of causal modelling: in particular, links between causal formalisms used in ML and philosophical accounts of causation.

• Explainability and interpretability: explanation vs. prediction; understanding as a cognitive and social achievement; what counts as an explanation for whom, and why.

• Reliability, robustness, and generalisation: principled notions of “reliability” beyond accuracy, statistical/philosophical perspectives on “reliable” scientific or societal use.

Submissions should be made by 11th May (anywhere on earth) on openreview. 

Author guidelines

• Format: All submissions must be in PDF format. Submissions are limited to four content pages. Unlimited additional pages are allowed for references and supplementary materials. Reviewers may choose to read the supplementary materials but will not be required to. Camera-ready versions may go up to five content pages. 

• Style file: You must format your submission using the ICML 2026 LaTeX style file. Please include the references and supplementary materials in the same PDF as the main paper. 

• Double-blind reviewing: The reviewing process will be double blind. As an author, you are responsible for anonymizing your submission. In particular, you should not include author names, author affiliations, or acknowledgements in your submission and you should avoid providing any other identifying information (even in the supplementary material). 

• LLM policy: The use of LLMs are permitted only as a writing assistance tool. 

• Dual-submission policy: We welcome ongoing and unpublished work. We will also accept papers that are under review at the time of submission, or that have been recently accepted for publication at a non-ML venue (i.e., any venue that is not ICML, NeurIPS, ICLR, or a similar conference or journal). Submissions published in venues for related fields (in particular, philosophy) are welcome.

• Non-archival: The workshop is a non-archival venue and will not have official proceedings. Workshop submissions can be subsequently or concurrently submitted to other venues.

• Visibility: Submissions and reviews will not be public. Only accepted papers will be made public.

• Reciprocal reviewing: Authors of submitted works are encouraged to volunteer as reviewers for other submissions, to ensure a fair and high-quality review process. 

For questions, please contact philml.icml26@gmail.com.