Operationalising Transparency in State AI Deployment

As part of the Tech+Society fellowship housed in the Centre for Digitalisation, AI and Society at Ashoka University, I am building two-part scholarship operationalising transparency for state deployment of AI—first, mapping the extent of AI use in the public sector led by the state or implemented through public-private partnerships, and second, designing a transparency and informational accountability checklist for policymakers and the FOSS community to help build and implement better, more transparent and citizen-friendly systems.

My project at Ashoka seeks to answer four key questions:

1. There is rapidly increasing integration of AI systems into public function, government services, and e-governance. Is this process transparent, and is information around those systems available to the common man, who is also the subject of such AI-based governance?

2. What does transparency in the deployment of AI by the State mean? To what extent can it be ensured with the present legal and technical tools we have?

3. If those are not enough, can we build a checklist to assess how transparent a ‘smart’ public-facing system is, and present it as a toolkit for policymakers to adhere to while building future systems?

4. What are the limitations to such an exercise—what sectors/usecases can this not apply to, and what do we do about systems that are inherently non-transparent?

Some months into this project, I’ve been exploring the principles of governance and informational transparency in AI contexts. This includes investigating how making system information like model cards or source codes open and public will help make a system just, fair, and accountable.

I wish to utilise the BoF format to achieve three main outcomes for this project:

1. To stress-test the findings and thesis of the project and the AI transparency checklist, especially with regards to how practical they are to implement by the FOSS community;

2. To understand the bottlenecks and pressure points around public sector AI deployment, and how we can tackle inherent non-transparency of blackbox systems; and

3. To gather experiences, lessons, and shared learnings from the FOSS community on civic tech governance—what works, what doesn’t, and how to get there.

Participants at this BoF space can expect to:

1. Go through and provide feedback on years of policy research on state deployment of AI, which maps 100+ use cases of public AI systems within India;

2. Co-create robust and transparent public AI systems by contributing to the transparency checklist accompanying the report; and

3. Participate in the first public showcase of the Indian Public AI Register—a tracker being developed alongside this project—which lists, furnishes information about, and attributes a transparency score to 100+ active public AI systems in India. This register is currently at the data collection stage and will go live around the second week of September.