AI, public information, and the hope of a new civic consciousness

I came across this "tweet by Shantanu Goel" (https://x.com/shantanugoel/status/2116384793588105702) about a really nifty little utility he built. He's taken Bengaluru's electoral rolls from the Special Intensive Revision (SIR), extracted the information from scanned PDFs, and built a "searchable website" (https://sir-bangalore.shantanugoel.com/) where people can check whether their names appear on the rolls. And it reminded me of something I'd written a few months ago called "AI Makes the World More Legible" (https://bhuvan.substack.com/p/ai-makes-the-world-more-legible). Looking back, that post feels a little messy. I was trying to get at an idea that was still taking shape in my head. But seeing projects like Shantanu's makes me think there's something to it. First, to be clear, I don't know if Shantanu used AI coding tools to build this site or did it the old-school, artisanal, gourmet way. But I've been seeing an explosion of similar projects lately, many built using tools like Claude Code and Codex. And I think they're a wonderful illustration of what I was trying to say. The central argument in my earlier post was simple: AI makes the world more legible. What do I mean by that? One of the things AI tools are exceptionally good at is grunt work. Mountains and mountains of grunt work. Think about all the information that's technically available to the public but practically inaccessible. Government documents buried in obscure corners of websites. Scanned PDFs that aren't searchable. Data spread across hundreds of files in different formats. Historical books gathering digital dust in university archives. Complicated legal and regulatory documents written in language that appears designed to defeat human comprehension. The information exists. But good luck doing anything useful with it. AI tools are increasingly good at extracting, cleaning, formatting, structuring, and analysing this information. They're also good at taking complicated, abstruse language and explaining it in terms that ordinary people can understand. In other words, they can take information that is otherwise illegible and make it legible. And then there's the sheer amount of work they can do. AI coding agents are essentially indefatigable. They don't get bored halfway through extracting data from the 847th PDF. They don't complain about having to clean up another thousand rows of badly formatted information. The only time they stop is when you tell them to, something breaks, or you run out of tokens. Of course, they make mistakes, sometimes spectacular ones. And for anything consequential, having a human in the loop to verify the output is essential. But the broader point remains. Work that might previously have required a team of people and weeks or months of effort can, in many cases, now be done by one person with the right tools in a fraction of the time. And this is why I increasingly think AI might be one of the best things to happen to people working in public policy, civic technology, and the broader public interest. Consider the amount of public information generated in India. Governments, regulators, municipalities, courts, statistical agencies, and public institutions produce staggering quantities of information. Unfortunately, an extraordinary amount of it is buried in scanned PDFs, badly formatted tables, fragmented databases, and websites that seem to have been designed by people who harbor a deep and personal hatred of their users. Information that is technically public is often functionally inaccessible. I've seen people use AI tools to extract and organise groundwater data, for example. There are similar opportunities with municipal budgets, land records, government spending, environmental data, legislative proceedings, and countless other public datasets. These are things that should be easy for citizens, journalists, researchers, and civil society organisations to access and understand. But they often aren't. I've been experimenting with this myself, albeit on a much smaller scale. I have two digitisation projects, "Akshara" (https://akshara.ink/) and "Paper Lanterns" (https://paperlanterns.ink/), where I use vision-language models and other tools to rescue old public-domain books, historical letters, speeches, and essays from the digital wilderness. A lot of this material is already available on places like Archive.org and university archives. But often, it's available only as a scan of a 100-year-old book, complete with faded pages, broken typography, and formatting that makes reading it an exercise in ocular self-harm. So I use AI to extract the text, clean it up, restore the structure, and make it readable again. I'm not doing anything particularly revolutionary here. But the fact that someone like me, with no formal programming background, can work on these projects largely by myself is remarkable. And it points to something much bigger. Until recently, building a useful public-facing website or application required at least some programming ability, or the resources to hire someone who had it. Today, AI coding tools are becoming capable enough to help ordinary people build fairly sophisticated applications. Not just toy websites or another bloody to-do app. Genuinely useful tools that combine data extraction, databases, search, maps, and public interfaces. One small example from my own life is "letmefuckingwalk.com" (https://letmefuckingwalk.com/). I love walking. I walk when I'm happy, when I'm sad, and pretty much whenever I need to think. Unfortunately, Bengaluru's footpaths seem to have been designed with two objectives: either make walking impossible, or ensure that anyone foolish enough to attempt it ends up breaking one or both legs. That's assuming a footpath exists in the first place. Out of sheer frustration, I asked Claude to help me build a website where I could upload photographs of broken, missing, dangerous, or encroached footpaths. I also wanted other people to be able to contribute photographs of the footpaths in their neighbourhoods. The idea was to create a public record of just how terrible the city's walking infrastructure is. Again, it's a relatively uncomplicated project. But I could build it myself, without assembling a team of developers or spending months learning to code. And that, to me, is the exciting part. These tools are making it easier for ordinary people to turn their frustrations into something useful. You don't need to be a programmer to build a tool that makes public information accessible. You don't need to be a data scientist to start examining government datasets. You don't necessarily need a large organisation or substantial funding to make a small contribution to a public problem. Of course, none of this means that the underlying data is automatically correct. AI can misread documents, invent details, or produce analyses that look convincing but are completely wrong. Making information searchable can also create privacy risks, especially when personal records are involved. Human verification, responsible data handling, and good judgment remain essential. But the barrier to getting started has fallen dramatically. And that brings me to a broader thought. I wonder if these tools might enable a new era of civic consciousness. Hopefully. Think about how difficult it is for an ordinary citizen to understand what their government is actually doing. Where is the money going? What happened to that promised road? How much was allocated to a particular scheme, and how much was actually spent? What has your elected representative been doing? What happened to the commitments made during the last election? Why does one neighbourhood have decent infrastructure while another looks like it survived a meteor strike? The answers are sometimes out there, scattered across government reports, budgets, parliamentary records, court documents, municipal websites, and other sources. But finding them, connecting them, and making sense of them is often an enormous undertaking. What happens when that work becomes dramatically easier? What happens when someone can take thousands of pages of government documents and turn them into a searchable, understandable public resource? Or build a tool that tracks what elected representatives promised against what they actually delivered? Or make it possible for citizens to understand how their local government spends money, without needing a degree in public finance and the patience of a Buddhist monk? I think there's something potentially transformative here. Not because AI will magically make governments more transparent or politicians more accountable. Technology alone won't do any of that. But perhaps it can make the information needed for accountability easier to access, understand, scrutinise, and share. And perhaps it can empower people who previously didn't have the time, money, technical skills, or institutional backing to work on these problems. Of course, there's a difference between making something visible and getting someone to act on it. A beautifully designed dashboard showing broken footpaths doesn't automatically fix a single footpath. A searchable database of government spending doesn't automatically stop corruption. But it's difficult to hold anyone accountable for something you can't even see. And that's why I keep coming back to this idea of legibility. AI can help shine a light on things that were previously obscured, buried, fragmented, or inaccessible. Not necessarily because anyone was deliberately hiding them, but often simply because of bad design, bad formats, inadequate resources, or bureaucratic indifference. Whether that increased legibility translates into greater civic participation and accountability is a different question altogether. I don't know the answer. But I genuinely hope it does. And seeing little projects like Shantanu's makes me a bit more optimistic that it might. B.