Collective Bargaining in the Information Economy
Can Address AI-Driven Power Concentration
Read the full paper on NeurIPS website or PDF via OpenReview .
1Simon Fraser University 2RadicalxChange 3University of Texas at Austin
NeurIPS Position Papers 2025
What is CBI?
Collective Bargaining for Information means data creators negotiate with AI builders over data use
researchers, writers, artists, coders...
Intermediary
labs/tech companies
- Data intermediaries represent anyone who creates valuable data for training, evaluation, retrieval (can be cooperatives, nonprofit, companies, public bodies...)
- AI operators build and deploy models using data acquired by a variety of data creators
- Negotiations cover terms of data use, compensation, limits, transparency
The Problem
Information Markets Fail
- Information is cheap to copy, hard to exclude
- Competitive markets drive price toward zero
- Producers struggle to capture value -- think piece work for 1 cent instead of well paying, stable content creation jobs
- Humans have centuries of experience iterating on imperfect solutions: IP rights, secrecy, monopolies
AI Makes It Worse
- Stronger models extract value more efficiently
- Likely to challenge copyright and erode the effectiveness traditional protections
- Risk: extreme power concentration--a potential "capital singularity"
The Solution
CBI Creates Friction To:
- Prevent unchecked extraction of the value of information by powerful actors
- Maintain incentives to produce high-quality information
- Distribute bargaining power more evenly
Benefits
- Higher quality data will create better, safer AI models
- Prevents collapse of information ecosystems (journalism, Wikipedia, etc.)
- Can support data provenance and transparency -- benefits entire research community
- "Natural source" of accountability for AI development
The Stakes
Political & Moral Risks
- Democratic instability from labor disruption
- Narrowed creative and cultural diversity
- Homogenization of moral and social norms
- Feedback loops that accelerate concentration
Technical Risks
- Fragile information ecosystems
- Less reliable, lower-quality models
- Unpredictable harms because of data opacity
- Loss of diverse training data sources (performance issues, moral issues)
Immediate Actions
Information Creators
Form and join data intermediaries that handle data flow to AI builders
ML Community
Ship attribution, consent, and robustness into model pipelines; do data valuation to make negotiations credible.
HCI Community
Design usable consent, opt-out, and bargaining interfaces that keep humans in the loop.
Urgent Policy Actions
Provide antitrust safe harbors and clarity so creator organizations can participate in CBI now.
Long-term Policy Framework
Set durable rules for data use, audits, and recourse in AI supply chains.
AI Safety Community
Consider CBI as core power-balancing work
Tech Companies
Engage in CBI on the buyer side to get better data; some potential increase in short term costs, but large potential gains -- both in terms of capabilities that arise from better data and maintaining/regaining public trust.
Technical Foundations
Research areas that support CBI:
Data Valuation & Attribution
- Influence functions, data Shapley, etc.
- Attribution methods
- Scaling studies
Data Control & Robustness
- Federated learning, consent infrastructure, licensing schema and protocols
- Provenance tracking
- Data poisoning and adversarial research
CBI Is Already Emerging
- Data licensing deals (e.g., sellers like Reddit, news publishers, etc. are contracting with buyers like OpenAI, Google)
- Labor negotiations centered on AI (e.g., SAG-AFTRA)
- Artist coalitions and opt-out movements
Not a cure-all, but a practical lever for better equilibria
Takeaway
A healthy information economy incentivizes the creation and curation of information and distributes power more evenly.
Collective bargaining for information is one step toward that goal.
A coalition approach is critical here. Need to get AI capabilities, AI safety, HCI, policy, and data buyers (AI companies) on the same page; incentives are more aligned than we might think!
Reach out if you want to help build it.
nvincent@sfu.ca • matt@radicalxchange.org • lihanlin@utexas.edu
Related Work
These papers provide technical and conceptual foundations for CBI. Browse all references at Shared References.
- Data Leverage
vincent2021dataleverage - Data Strikes
vincent2019datastrikes - Data and Labor
vincent2023datalabor - Collective Bargaining
hardt2023collective - Data Shapley
ghorbani2019datashapley