There’s a scene near the start of Slumdog Millionaire in which a little boy jumps into a pit of human waste just to get a film star’s autograph. He climbs out, covered in it, and he’s smiling. When I first watched it, I felt sick. What bothered me more, though, was that many people in India didn’t seem that bothered. It was just another scene. And I get why. If you live in almost any big Indian city, you’ve seen overflowing bins, burning rubbish heaps and drains full of plastic so often that you don’t really see them anymore. You walk around the pile and keep going.
Also read: Claude Sonnet 5.5 vs Opus 5.5 vs Fable 5.1: Which model you should use
I don’t want to be someone who just walks around the pile. I think many people my age feel the same way. We’re used to calling things out when they’re wrong, and we’re also growing up with tools that can actually do something about it. When I got into computer science and machine learning, one question kept coming back to me. If AI can recognise faces and help doctors read X-rays, why can’t it help us notice the rubbish we’ve all learned to ignore?
This is a bit personal for me. One of my parents worked as a director in a major municipal body in Delhi, and garbage came up at home far more than you’d think. The main thing I learnt from those conversations was that waste piles up faster than any city can clear it. Sanitation workers work really hard, often in pretty awful conditions, but there are never enough of them. Municipalities just don’t have the money to have someone on every street all day. So most of the time, a city only finds out about a problem when someone complains, a photo goes viral, or an inspector happens to walk by. By then the pile is bigger, the drain is blocked, and the stray dogs have already found it.
Basically, the system mostly reacts after things go wrong. I want to argue that computer vision could help cities catch problems earlier and direct their limited staff to the places that need them most. But I’ve also realised something I didn’t expect when I started: the technology by itself won’t clean anything. It only helps if it fits into how cities actually work and is fair to the people who deal with our waste every day.
First, how big is the problem? According to the Central Pollution Control Board’s 2020-21 report, Indian cities produce around 1.6 lakh tonnes of solid waste every day. Most of it is now collected, which is real progress. But only about half is actually treated. Much of the rest ends up in landfills and open dumps, some so large you can see them from the highway.
The government is trying. Swachh Bharat Mission (Urban) 2.0 started in October 2021 with about ₹1.41 lakh crore, and its goal is to make every city “garbage free.” That means door-to-door collection everywhere, people separating their waste at home, processing it properly, and cleaning up the old dumpsites. I think these are good goals. The problem is that, to meet them, cities need information they often lack. Where is the garbage? What kind is it? Has anyone actually picked it up? That’s where I think AI can help.
Here’s the basic idea, without too much jargon. Computer vision is the part of AI that helps computers understand images. You show a model thousands of labelled images, such as “this is a plastic bottle”, “this is food waste”, or “this is a clean road”, and eventually it learns to recognise those things on its own. Some models (classifiers) simply tell you what’s in a photo. Others, like the YOLO models (“You Only Look Once”, which I think is a great name), find every object in the picture and draw a box around it, fast enough to work on live video.
The catch is that a model only knows what it has been shown. There are some good free datasets out there, like TACO, which has photos of litter in real-world settings, and TrashNet, which has photos of sorted recyclables. When I started building my own small prototype, I quickly realised these weren’t enough. A model that learned mostly from clean European parks and nicely photographed bottles gets really confused by an Indian street, where you might have a torn plastic bag, vegetable peels, broken bricks and a sleeping dog all in one shot, with harsh sunlight on top of that. Honestly, that was the most frustrating and the most useful thing I learned. If this is going to work here, it has to be trained on pictures from here.
There are a few ways cities could use it. They could run it on CCTV cameras they already have. They could put cameras on garbage trucks or give supervisors a phone app. Or they could place small, low-cost devices near bins and common dumping spots that run the model on-site, so video doesn’t have to be sent anywhere.
So what could it actually do? It could show which areas keep getting littered and catch illegal dumping as it happens, not weeks later. It could check how full bins are, so trucks go where they’re needed instead of following a fixed route to half-empty bins while others overflow. At recycling plants, it could help sort waste into dry, wet, plastic and electronic waste, which is exactly what SBM-U 2.0 wants. And it could make complaint apps smarter. If you take a photo of an overflowing bin, the app could identify it, tag the location and send it straight to the right ward officer.
Also read: 3 AAA games, 1 under Rs 1.5 lakh laptop: Alienware 15 gaming experience
The first problem is accuracy. Real streets are messy. The light changes, garbage comes in every possible shape, and a model that looks amazing in testing can fail badly in a busy market. If it gives too many false alarms, workers waste their time. If it misses too much, nobody will trust it.
The second problem is privacy. A camera looking for garbage is also looking at people. Without clear rules about what gets recorded, how long it’s stored, and who gets to see it, a cleanliness project could slowly turn into a surveillance project. I think these systems should be built to spot waste, not to identify people, and India’s data protection law should fully apply to them.
The third problem is the one I didn’t think about enough at first: what happens after the camera spots something. If an alert goes out and nobody acts, the whole thing is pointless. Someone has to be responsible for each alert, there needs to be a deadline, and there has to be a record of whether it was fixed. Otherwise, AI just makes a fancier list of complaints for everyone to ignore.
And then there are the waste pickers. India has millions of informal waste pickers, by most estimates, and they recover a huge amount of recyclable material without getting much credit. Automated sorting could push them out. I don’t think it has to, though. The same technology could help them find valuable material, bring them into the official system and make their work safer. I honestly don’t know which way it will go. I think it depends on what cities decide now.
Some cities have already started trying this. The Greater Chennai Corporation has proposed installing AI cameras on 55 enforcement vehicles to catch illegal dumping and other waste-rule violations, with all footage monitored from its Integrated Command and Control Centre. What I liked about this is that those vehicles already use GPS and fixed routes, and any change in route sends an alert to a dashboard, so the AI is being added to something that already works. In Kerala, Thiruvananthapuram launched a project that uses AI to capture people dumping garbage in public places and water bodies, as well as the vehicles bringing it. Instead of watching everything all the time, the AI flags suspicious footage for someone to check, which seems like a smart way to handle privacy. Dubai Municipality started a pilot in January 2026 with AI cameras on garbage trucks, where images are analysed right away and violations appear on dashboards so teams can respond quickly. In all three cases, the camera isn’t the whole plan. There’s always a control room and a team that actually goes and does something.
That’s probably the biggest thing I’ve learned, both from experimenting with my own model and from years of hearing how municipalities really work. AI can help a lot, but it can’t do the hard part for us. A camera can spot a pile of rubbish. It can’t hire more sanitation workers, make people follow segregation rules, or ensure waste pickers are treated fairly. People have to decide to do those things.
What AI can do is remove our excuse for not seeing. I think rubbish has stayed in our cities partly because it was never anyone’s problem at any specific moment. If a computer can point to a pile and say “this one, here, right now, and it’s your ward’s job,” that changes things. It gives stretched-thin municipalities better information and gives ordinary people a louder voice.
For a long time, many of us have been running past the same pits. I want to stop, look, and help build tools that make a difference, with privacy, fairness and accountability built in from the start, not added later as an afterthought. That’s a big part of what I want to work on in college, and hopefully after.
Hanut Lal is an International Baccalaureate (IB) Diploma Programme student at Heritage Xperiential Learning School, Delhi, and an intern at the Global Responsible Tech Lab (GRTL) based at the University of Mannheim. He is passionate about leveraging artificial intelligence to address real-world challenges, particularly in education, healthcare, and responsible technology. His work focuses on developing human-centred AI solutions that create meaningful social impact, with a particular emphasis on AI and interactive learning: designing tools and experiences that make learning more engaging, accessible, and responsive to individual students’ needs.