London Data Week: Accountability Across AI Labour

This document outlines initial correspondence by Sumona and Jonas regarding potential contributions.

Executive Summary

This London Data Week session treated accountability as the through-line: how artificial intelligence reshapes questions of power, labor, governance, and inequality. Participants discussed AI not only as a technical system but as a social and economic infrastructure built upon human labor, data extraction, and governance decisions. A recurring theme was the disconnect between those who contribute to AI systems and those who ultimately control and benefit from them—and what accountability must mean across that gap.

The discussion highlighted concerns around hidden labor within AI supply chains, the limited representation of Global South communities in model development and governance, the challenges of effective regulation, and the emergence of what was described as agentic inequality—inequality in access to capable AI systems rather than merely access to the internet.

The session concluded with a collective reflection on what fair AI requires, emphasizing accountability, dignity, care, transparency, and mechanisms for redress.

Participants

The workshop brought together participants from diverse professional backgrounds, including public health, psychology, law, education, technology policy, design, and academic research.

• Andrea — Public Health Worker, Lambeth Local Government
• Myra Viedge — Clinical Psychologist
• Wendy Petersen — Senior Learning Technologist, Red & Yellow
• Dega — Academic Content Editor
• Lufthi Noorfitriyani — Designer and Master's Student
• Nigel Hartell — Independent Consultant and Former Chief Architect
• Alexia Dubreu — Project Lawyer, International Bar Association
• Osei Kagyah — Tech Policy Analyst
• Monica Visani Scozzi — PhD Researcher
• Martyna Durlik — Community and Content Manager
• Sumona Bose — Sociologist and Practitioner
• Tatjana Peric — Researcher
• Others — Workshop Participants

AI as Opportunity and Risk

The opening discussion revealed a shared recognition that AI presents both opportunities and risks. Participants described AI as a powerful tool capable of improving services and expanding access to information, while also reproducing existing inequalities and creating new forms of exclusion.

Several participants highlighted the lack of public understanding regarding how large language models function. Concerns were raised about overreliance on AI systems, embedded biases, and the tendency for these technologies to appear more authoritative than they actually are.

A participant summarized this tension by describing AI as a “double-edged sword,” capable of empowering users while simultaneously reinforcing harmful social structures.

The Hidden AI Supply Chain

A central theme of the workshop was the human infrastructure underlying artificial intelligence. Although AI products are often presented as automated systems, participants emphasized that they depend on extensive networks of human labor. These include data contributors, content moderators, dataset annotators, platform workers, researchers, and infrastructure providers distributed across the world.

The discussion drew comparisons between AI and globalized manufacturing systems such as fast fashion. In both cases, consumers are often disconnected from the workers and communities that make production possible.

Participants argued that understanding this supply chain is essential for evaluating fairness, accountability, and responsibility within AI systems.

Global South Representation and Governance

A recurring concern was the imbalance between contribution and influence. Participants noted that countries in Africa, Asia, and Latin America contribute substantial amounts of data, labor, and lived experience to AI systems while remaining underrepresented in governance structures and decision-making processes.

Indonesia was highlighted as an example of a country generating enormous volumes of social media data yet having limited influence over how models are trained, deployed, and regulated. This dynamic was summarized through a simple observation: the Global South contributes significantly to AI systems but exercises limited control over their development and governance.

Accountability and Data Ownership

The conversation repeatedly returned to the question of ownership. Who owns the data used to train language models? Who is accountable when harms occur? Can contributors meaningfully opt out of data collection and processing?

The discussion drew on research from Fair Work Kenya, which investigates accountability across AI supply chains. Participants noted that data frequently changes hands through complex chains of platforms, brokers, developers, and service providers, making responsibility increasingly difficult to trace. This produces an accountability gap in which contributors often lack both visibility and recourse.

Regulation as a Dual-Use Framework

Participants broadly agreed that regulation is necessary. However, there was also recognition that regulation can itself become a tool of exclusion. The workshop explored how regulatory frameworks may simultaneously protect users while reinforcing geopolitical and economic inequalities. Export controls, access restrictions, and competitive pressures can shape who benefits from AI innovation and who is left behind.

Rather than treating regulation as inherently positive or negative, participants characterized it as a dual-use framework whose outcomes depend on how it is designed and implemented.

Agentic Inequality

One of the workshop's most significant concepts was agentic inequality. Traditional discussions of the digital divide focus on connectivity and internet access. Agentic inequality shifts attention toward access to capable AI systems. This framework identifies three dimensions:

1. Availability of models
2. Quality of models
3. Quantity of available models

Under this view, two individuals may both have internet access while experiencing very different opportunities because they have access to different levels of AI capability. Participants suggested that this form of inequality may become increasingly important as AI systems become embedded in education, work, healthcare, and governance.

Human Vulnerability and AI

The discussion also focused on the relationship between human psychology and AI systems. Participants highlighted concerns regarding overtrust, manipulation, cognitive offloading, and the growing tendency to treat AI outputs as authoritative. Several participants argued that effective governance must account for the fact that people are inherently vulnerable to persuasive technologies.

At the same time, there was disagreement about AI's impact on learning and cognition. Some participants emphasized its potential to support critical thinking and personalized learning, while others warned that excessive reliance may undermine memory formation and deeper engagement with knowledge.

What Fair AI Requires

The workshop concluded with a collective reflection on the values necessary for fair AI. Participants identified several recurring principles:

• Accountability
• Dignity
• Empathy
• Care
• Transparency
• Justice
• Redress
• Human flourishing

A particularly influential theme was the importance of accountability mechanisms that allow harms to be identified, addressed, and remedied after deployment.

Future Research Questions

1. How does AI diffuse through society?
2. Who owns the data that goes into language models?
3. How can Global South representation be built into model training and governance?
4. How can accountability be maintained across complex AI supply chains?
5. How should agentic inequality be measured and addressed?

Collective Governance and Alternative Futures

While much of the discussion focused on risks, participants also explored possibilities for building more equitable AI systems.

A recurring theme was that current debates often assume that AI development is driven primarily by governments and large technology companies. Participants questioned whether alternative models of governance could provide greater representation, accountability, and benefit-sharing, particularly for communities in the Global South.

Several contributors argued that communities generating data should have a stronger role in determining how that data is used, who benefits from its use, and what forms of oversight are required. This perspective challenges the assumption that data is simply a resource to be extracted and instead treats it as a collective asset connected to social, cultural, and economic rights.

The discussion also highlighted the importance of developing governance mechanisms that operate across borders. Because AI systems are trained, deployed, and used globally, decisions made in one country can have significant consequences elsewhere. Participants noted that meaningful governance will require collaboration between researchers, civil society organizations, policymakers, workers, and affected communities.

Rather than asking only how AI can be made safer, participants encouraged a broader question: What kinds of AI systems should societies choose to build, and for whose benefit?

This shifts the conversation from managing risks to actively shaping technological futures. It also emphasizes that fairness is not simply a technical property of a model, but a social outcome that depends on participation, representation, and democratic accountability.

The workshop concluded with a shared recognition that the future of AI remains open. The challenge is not only to govern existing systems more effectively, but also to create institutions capable of ensuring that technological progress contributes to human flourishing across diverse communities and contexts.

Conclusion

The workshop demonstrated that AI is not merely a technical innovation. It is a social system built on human labor, shaped by governance decisions, and distributed through existing structures of power. Understanding AI therefore requires examining not only models and algorithms, but also the people, institutions, and communities that make these systems possible. The challenge ahead is ensuring that the benefits of AI are shared more equitably while creating meaningful mechanisms for accountability, representation, and redress.