Opportunities and threats within the AI supply chain

The opacity of the global AI supply chain allows leading AI companies to act with little accountability or oversight

Addressing the opacity and lack of accountability that define the global AI supply chain, and turning these into opportunities for change, is a priority for The Collaborative. This problem definition outlines our understanding of the problem and how we see it evolving. We expect this statement to evolve as we learn and in response to changes in the field.

The AI supply chain begins with physical materials — the minerals and machines needed to make the microchips used to train and deploy algorithms. It includes the energy inputs and other infrastructure used to build and run the data centres housing those microchips; the end of the lifecycle of the built environment of AI; the software used to train AI models and run inferences on them; the data used as inputs to that software; and human work involved in annotation, training, and evaluation.

The breakdown in accountability

AI companies extract and consume minerals, water, and energy to build and run infrastructure. They also extract behavioural and other data, often treated as free inputs to the models. The opacity of the AI supply chain enables these practices. Networks of subcontractors and corporate relationships blur the lines of responsibility and accountability.

Few accountability measures exist. National regulators lack jurisdiction beyond their borders. The activities of AI companies often straddle broadcasting regulators, consumer protection watchdogs, data protection agencies, and other watchdogs, none of which was designed to oversee AI as a technology. Safeguards that exist for other industries — satellite imaging, nuclear energy, bioweapons – are missing for AI.

This lack of accountability hinders new public interest entrants from driving the AI field to deliver outcomes that matter to people and communities.

Accountability is broken at each of its four stages: standard setting; investigation; answerability; and sanction. For example, there is no agreement about what to do with state-of-the-art LLMs that are built on data extracted without due regard for privacy, confidentiality, and other safeguards. To the limited extent that sanctions exist, they tend to be ineffective and sporadic.

A more open field for the public interest

This lack of accountability hinders new public interest entrants from driving the AI field to deliver outcomes that matter to people and communities. To bend the arc of AI toward the public interest, it is crucial to lower barriers to entry for public interest entrants and to open paths for new actors to pursue collaborative, open, and sovereign alternatives. A transparent, inclusive, and accountable global AI supply chain would open the field to these entrants and enable interest groups in and around the supply chain, such as communities close to data centre buildout and workers, to strengthen their positions and leverage to catalyse change.

Learn more about the problems we are focused on: