Designing and Governing AI Agents as Economic Actors

The rise of autonomous AI agents which produce, transact and coordinate with little human oversight, is expected to transform the economy in the near future. Institutions, enterprises and workers are already beginning to embrace this technology, but are ill-prepared for dealing with the seismic effects of this paradigm shift. This project will build upon our ongoing research: AI Agents Economies: A Design Space [https://docs.google.com/document/d/1aokOxQA-7C8oqA_VsPkMP0nTpvpOBU6kELSU66_p9Go/edit?tab=t.0], which provides a conceptual toolkit for designing governable AI agents by unbundling the economic agent into three taxonomised dimensions: i) Capabilities, ii) Autonomy, iii) Agency, and operationalising the 'permeability' of the border between AI and human economies as a set of governable Gates and Objects. The project will further operationalise some aspects of this paper, or focus on a specific governance intervention in this design space.

See more details in the project proposal: https://docs.google.com/document/d/1lk5xL7NG8UdgX-9soFwpH7_UWFNUk_jr0j34Tp4Ufy0/edit?usp=sharing

Here is the theory of change for our overarching goal of providing the blueprint for designing AI agents and the interface with our economy:

• The Problem: The default trajectory of AI agent deployment is the "accidental emergence" of highly permeable agent economies, driven by commercial pressure, racing dynamics which result in a techno-determinist paradigm. This blind approach creates severe systemic risks, specifically gradual disempowerment (where humans lose the ability to steer the economy) and agentic inequality (where power concentrates with those controlling superior agents). 
• The Opportunity: We have the ability to create an economy where AI agents amplify human judgment and autonomy — and don’t replace it, erode it, or centralise it in the hands of a few. A well-designed economy infused with AI Agents could help move us towards a world of where major challenges are solved and resources are abundant.
• The Urgency: We are at a critical juncture where agent architectures and economic interfaces are being introduced. If these structures calcify without oversight, we risk locking in a future where human agency is structurally eroded. In contrast to usual reactive regulation playing catch-up to frontier technology, it is more effective to build governance mechanisms into the foundation of a new system than to regulate it after it has become entrenched infrastructure.
• The Intervention: Governance gaps identified will be addressed with useful knowledge frameworks that underpin actionable solutions. 
• The Outcome: Our work will equip researchers, policymakers and AI companies with the blueprints needed to implement responsible design.

This SPAR project aims to supervise mentees who would be interested in working on a complementary (more focused) proposal that ties to our framework and aligns with its goals.

The team will be led by Alex, and Elsa will meet with the entire team once a week and provide asynchronous feedback on their work. Mentees should collaborate actively, particularly in the initial exploratory phase of the project. We also aim to support mentees in their career growth in the AI safety field. Mentor involvement may vary around project milestones.

We are looking for a mix of skills, and open to applicants from various backgrounds. 

• Technical AI: Applicants with a technical background who can anticipate what future AI Agents could look like architecturally.
• Non-Technical: Strong background areas such as Economics, Law, Policy, Social Science etc are highly valued. We need people who can reason about incentives, market dynamics, regulatory structures, and how humans share activities with AI agents.
• General: Ability to synthesize complex information, reason abstractly about novel systems, and communicate clearly. We value proactivity and willingness to challenge defaults and assumptions.

Min 8h, preferred 12h