Project summary

My project addresses one of the most prevalent and urgent problems in AI safety, the absence of practical human oversight frameworks that can function across various institutional settings globally.

This problem I intend to address is specific and neglected. Most governance work on oversight frameworks at institutional capacity does not exist everywhere and where it exists are not designed to be practical, in plain language to make it easily adoptable. AI systems are being deployed at a rapid rate whereas the infrastructure to correct or override them is absent. That gap is why ungoverned AI is very dangerous.

What are this project's goals? How will you achieve them?

My research will examine what minimum viable human oversight standards can be designed to function across institutions with varying levels of regulatory capacity, producing a framework that is practically adoptable globally.

I am currently drafting a Human Review and Escalation Standard Operating Procedure (S.O.P) through my volunteer role on the Legal committee at the AI Ethics and Integrity International Association (AIEI). This document defines precisely when humans must intervene in AI decision making and how to escalate high risk situations before harm occurs. Its intention is to keep humans meaningfully in control of AI outputs at organizational and institutional level.

The MSc in AI Ethics and Society at University of Leeds, UK where I will be finalising this research provides the academic supervision and research environment to take this work from practitioner level contribution to a rigorous, publishable output.

How will this funding be used?

Minimum Funding: $18,000

It covers the researchers living stipend for 12 months in Leeds, UK

Ideal Funding: $28,000

Ideal funding includes the living stipend, research materials, two conferences, and publication/dissemination

Ideal Funding Budget Breakdown

Researcher living stipend, 12 months Leeds = $18,000

Research materials and resources = $4,000

Conference attendance, 2 events = $4,000

Publication and dissemination = $2,000

Total= $28,000

Who is on your team? What's your track record on similar projects?

This is an individual research project, led and conducted by me. I will draw on academic supervision at the University of Leeds and relevant professional networks in AI governanceand ethics for expert input and feedback.

I have developed experience in AI governance and safety through research, policy work and voluntary contributions. I contributed to a policy recommendation on child-centred AI governance in Ghana through AFRI-SPARC, and have published two essays on AI safety and governance on the EA Forum. I have also been admitted to the MSc in AI Ethics and Society at the University of Leeds, where I intend to develop my research on practical AI oversight. I currently serve as an AI Policy Researcher with the Africa Strategic Policy and AI Resource Centre (AFRI-SPARC) and as an Associated Participant with the AI Ethics and Integrity International Association (AIEI), where I am contributing to the ongoing development of a Human Review and Escalation Standard SOP. All of my work to date has been voluntary and unpaid.

linkedin: www.linkedin.com/in/agwu-nneoma-0927a0b4

What are the most likely causes and outcomes if this project fails?

The main risk is that the research may not produce the proposed oversight framework. However, it will produce useful findings on what makes AI oversight framework difficult to implement across organisational and institutional settings.

How much money have you raised in the last 12 months, and from where?

I haven't raised any money in the last 12 months all my contributions so far have been voluntary and unpaid.
Although, this work was recently reviewed by Max Harms, Corrigibility Research Fund Manager at Lightcone Infrastructure, who passed on scope rather than quality and recommended this platform as a better fit. He identified the comparative angle of this research ; what oversight looks like in institutions with thin regulatory capacity, as a real and underexplored gap.