The workshop aims to establish a foundation for a long-term, interdisciplinary research agenda on the incentives that govern collaborative learning and data sharing. By synthesizing perspectives from machine learning, economics, law, and policy, the aim is to identify the most pressing gaps that limit the practical and ethical integration of distributed data into impactful AI systems. The discussions will center on challenges in valuing and compensating data contributions, designing mechanisms that remain robust in strategic multi-agent environments (including those with autonomous AI participants), and aligning technical solutions with evolving legal and regulatory frameworks. The outcome will be a structured set of research directions that can guide research and policy work over the next decade.
If you have any thoughts on the collaborative research agenda, please let us know by completing this form.
We are currently considering the following directions, but they are by no means exhaustive.
Effective collaboration in machine learning requires reliable methods to quantify the value of data, determine fair compensation, and ensure proper attribution of credit. Existing approaches to data valuation—such as Shapley value–based methods or performance impact metrics—often fail to scale, adapt to non-stationary settings, or account for strategic manipulation. Current frameworks for rewards and attribution are fragmented, leaving open questions about how to fairly recognize contributions in multi-round, multi-party collaborations, and how to balance transparency with privacy. Furthermore, the emergence of data marketplaces presents a challenge in designing pricing and trading mechanisms that encourage high-quality contributions while preventing misuse and preserving ownership rights. Advancing this area requires new theoretical models, scalable algorithms, and robust market designs that work across heterogeneous data, varying trust assumptions, and dynamic participation.
Collaborative data sharing is shaped by the strategic incentives of participants, who may act to maximize their benefit, sometimes at the expense of the collective good. Research has identified risks such as free riding, defections, collusion, and misrepresentation of data quality; however, much of the current theory assumes static or human-only actors. The increasing use of large language model (LLM) agents—capable of autonomous decision-making, negotiation, and coordination—introduces a new strategic layer. These agents can both enable and undermine collaboration, depending on their objectives, capabilities, and alignment. Existing incentive-compatible mechanisms may not be robust to such actors, especially when they can rapidly adapt or form alliances. There is a need for more sophisticated models of agent behavior, novel mechanisms that are resilient to both human and AI strategies, and empirical studies that capture the evolving dynamics of data sharing in environments where humans and autonomous AI systems coexist.
Technical designs for data sharing and incentives operate within legal and policy frameworks that are often jurisdiction-specific, evolving, and incomplete. Regulatory regimes govern privacy, intellectual property, data ownership, and liability, yet the application of these laws to modern AI systems—particularly LLMs—remains unsettled. Questions persist around the status of training data in copyright law, the handling of personal data under “right to be forgotten” provisions, and the accountability for AI-generated content derived from shared datasets. The rapid deployment of LLMs has exposed gaps between technical feasibility and legal compliance, especially when models can memorize sensitive information or generate outputs that qualify as derivative works. Bridging this gap requires frameworks that integrate legal constraints into the design of incentive mechanisms, while also informing policymakers about the technical possibilities and limitations of privacy-preserving and attribution-aware methods. Interdisciplinary research is essential to ensure that solutions are both implementable and adaptable to future regulatory changes.