AI in the Public Interest: Local Data, Global Impact, by Factly and Meta
Factly, in collaboration with Meta, organized the “AI in the Public Interest: Local Data, Global Impact” event today, and they were kind enough to invite me to participate. Multiple people from Factly, including Rakesh (CEO), Shashi (CTO), Venu (Head of Open Data Strategy), have spoken at FOSS United Hyderabad events over the past two years. Rakesh was also recently on the Free/Libre AI panel at IndiaFOSS 2025.
The event was closed-door, with participants from the Government of Telangana, Industry, Civil society, and Academia. The event is an official AI Impact Summit pre-event (but unfortunately the website hasn’t been updated yet). The event was organized under Chatham House Rule, because of which I’ll only provide a summary of the event discussion, my contributions, and my overall takeaways from the event. Before going further, it’s worth noting that this is the second AI Impact Summit pre-event that I’ve participated in, and the experience has been overall interesting. See notes about the first AI Impact Summit pre-event I attended.
Event summary
The event was 3+ hours long, from 10 AM - 1+ PM. After introductions and opening remarks from organizers, all of the participants had a few minutes to share their own opening remarks. There were 20+ participants so this ended up being a significant chunk of the overall event duration.
- Most individuals, especially representatives from civil society organizations, shared on-the-ground, first-hand experience with AI and technology. A lot of their grievances revolved around the fact that technology is forced on the poor, including AI
- Another key insight was the fact that the digital divide is being exacerbated by AI, giving rise to an AI-literate subclass within the digital-literate class
- A lot of the time, “public interest” is conflated with “government service delivery”, instead of looking at it with a broader lens
- Organizations that work on improving transparency and accountability of the government asked whether AI would help or hurt their efforts
- Industry representatives primarily highlighted issues surrounding data quality, frequency of data updates, data governance frameworks, data maturity, data silos within and in-between government departments
- An industry representative highlighted the important need to identify meaningful workflows and build “Vertical AIs”, in contrast with the current approach of expecting people to adapt their workflows to existing solutions
- A representative from academia highlighted the need to bring geopolitics into the conversation and the risk of critical dependency on service providers headquartered outside India
- Understandably, the recent movie Humans in the Loop came up during conversation, and it’s worth watching if you haven’t already. It came up in the context of creating space for local knowledge systems in the current AI paradigm and the “right to be heard”
- An important insight was the fact that people don’t know what the systems knows about them - what information/data has already been collected by the government about them. Efforts like the Jansoochna portal in Rajasthan address this issue, but such initiatives don’t seem to be widespread
- Related to this note, another participant highlighted the value of creating a “whole of government data glossary compendium” which would describe what information is collected by/available with what government department
- I liked the phrase “data-rich but information-poor” which seems to describe most data projects in the world
- This is unrelated, but one of the participants talked about the fact that some of their events don’t charge an entry fee but require proof of a digital commons contribution, e.g., code contribution
- One of the participants brought up the “our phone is always listening to us” conspiracy, so I’d like to point out the excellent recent article No, Your iPhone Isn't Listening to You. But the Truth Is Even Worse - CNET
- One of the participants highlighted the need to look beyond ethical and moral frameworks for AI to build robust resolution systems
- TIL that Sikkim was the first Indian state to launch an open government data portal
After the brief opening statements from the participants, we split into breakout sessions around three themes, and gathered after 30-ish minutes with specific recommendations for the Government of India for a 6-9 month timeline. The recommendations and notes from the meeting will be collated into a public report, to be shared with the AI Impact Summit, in my understanding.
My contributions
I started my opening statements with a provocation about the real cost of AI, highlighting the significant subsidies received by the entire AI supply chain industry, from critical minerals to data centers. I then focused on three points
- AI, and software in general, is irrefutably driven by FOSS, and the people driving the FOSS-powering AI also get to define the culture and the mental models for the individuals participating in the ecosystems. Conversations about leading the Global South on AI or indiginizing AI are practically meaningless when we aren’t participating, let alone creating, the FOSS tools central to the AI industry. The mental models and culture surrounding the tool builders have an impact on what people choose to build with the tools. It’s worth noting that most of the other participants highlighted issues concerning building solutions with the tools or how the solutions were governed, not the tools themselves
- The lack of participation or adoption from “decision makers” regarding the AI-related open standards conversations happening around the world, like CC Signals by the Creative Commons Foundation, Model Openness Framework by the Linux Foundation AI & Data initiative, Open Source AI Definition by the Open Source Initiative. None of these standards/frameworks might fit what we want as a nation/ecosystem, but how do we “lead the Global South for AI” if we don’t engage in global dialogue
- Open-washing is increasing with the proliferation of AI, creating significant noise in conversations with “decision makers”
During the breakout session, I pushed for “Public money, public code” as an official recommendation from the participants of the event. It ended up being worded as “Public money, Public Good” but was accepted as an official recommendation from the event to the AI Impact Summit. I also tried to push for the adoption of the aforementioned frameworks/standards as global best practices to be integrated into our approach/thinking, but, unfortunately, we ran out of time, and I wasn’t able to build consensus on this recommendation.
Overall experience
There’s a lot of value in structuring events such that they invite participation from attendees, and nudge them to come together to agree on actionable short-term/long-term goals. The recommendations might not get adopted, but the simple act of bringing a diverse group of people together, asking people to share their perspectives, and getting them to agree on shared objectives seems to be a great way to help people develop agency. There’s a lot of outrage (and understandably so), and helping people move beyond the outrage towards action seems to be the need of the hour.
I don’t know why it took me this long to see it because this is basically what the GCPP program by Takshashila forces participants to experience. Participants are split up into squads and are expected to come together to work on policy recommendations. It isn’t easy to convince a group of people to move in a certain direction, and it requires a certain amount of willpower, charisma, grounded experience, and clarity. and stubbornness.
I don’t know if the “Public money, Public good” recommendation will be taken seriously by the AI Impact Summit people, but I’m happy that I shared the “Public money, Public code” spirit with a few more people. I also walked away with 10-ish people to start further conversations about FOSS United, including two from the Government of Telangana, so I’m counting this as a win. I’ll share a link to the event report if and when it’s made public.