AI Traffic Cameras Were Always a Bad Idea

Disclaimer: I am still learning about public policy and base this blog post on my current understanding of the topic. I will continue improving this post as I learn more. Any feedback and corrections are welcome.

Will we have safer roads in Kerala with the new AI traffic cameras?

Maybe.

But is this the best way to allocate resources to ensure safer roads?

Not really.

Keeping the controversies aside, in this article, I will explain why this is the case and how this entire scenario is a textbook example of shortsighted policymaking.

But first, let’s look at the current state of roads, road accidents, and the “Safe Kerala” project, which resulted in these AI traffic cameras.

Kerala’s Roads

Kerala has a dense road network. The state has 3,31,904 kilometers of road, or 5.6% of India’s total ¹, while sharing only 1.18% of the country’s total area ².

But Kerala has narrower highways than the rest of the states, while studies have repeatedly shown that wider roads can reduce accidents ^{3 4}.

It was the only state to request it from the National Highways Authority of India besides Goa, stating difficulty in land acquisition ⁵. Although new projects like the ongoing widening of NH66 in north Kerala look to improve this, they will still lack things like wide medians found commonly in other parts of the country ⁶ due to this difficulty.

A massive 24.2% of households in Kerala own a car compared to 7.5% nationally ⁷. Such a significant figure could maybe indicate a lack of reliable public transport facilities.

Road Accidents

Over the last year, 4,317 people died in road accidents in the state.

Driver error being the cause of 80% of the accidents ⁸ implies that improper infrastructure (or lack thereof) is only one of a myriad of reasons, including improper training during driving tests and negligent driving ⁹.

Pedestrians in Kerala also have a hard time as most cities and towns are not walkable. They account for 26% percent of the total death in road accidents, which is quite concerning ^{8 9}.

The lack of proper infrastructure for pedestrians, like footpaths, crosswalks, and bridges, misuse of existing infrastructure by parking on pathways, stopping on crosswalks, and extending private spaces (like shops) to roads, and people’s/government’s general indifference to pedestrians, are significant contributors to this problem.

AI Traffic Cameras

On 20th April 2023, the Government of Kerala started operations of its new fleet of traffic cameras under the Safe Kerala project. The fleet includes ¹⁰:

• 675 AI traffic violation cameras
• 25 AI parking violation cameras
• and 18 AI red-light violation cameras

The traffic violation cameras can detect the following violations:

• Not wearing helmets on motorcycles
• Not wearing seatbelts on cars
• Using phones while driving
• More than one pillion rider on motorcycles

In the future, these cameras will also detect lane and one-way violations. Or at least, that is what they say.

The cameras are accompanied by control rooms in each of the districts in the state. Additional personnel, including 85 squads, 99 vehicle inspectors, and 225 assistant inspectors of the Motor Vehicle Department, will also be added to the service under the Safe Kerala project ¹¹.

Keltron delivered the entire project for 128 crores (just for the infrastructure) ¹² and had the help of private companies on tenders. So the AI—or the computer software that processes the images from these cameras, detects traffic violations, and issues the challan—was built by a private company.

Keltron expects it will send around 25,00,000 challans every year. The first few days of operation saw 4,50,552; 3,97,487; 2,68,378; 2,90,000; 2,37,000; and 2,39,000; violations per day. If this trend continues, it will be close to 7,30,00,000 violations annually.

Although “if you create more laws, there will be more lawbreakers” is a solid argument on why these cameras were a bad idea, I would like to look at it from different perspectives and create a more concrete statement.

Is Kerala Ready?

Is Kerala ready to be implementing AI traffic cameras?

No.

As I pointed out in the above sections, Kerala’s problems are bad drivers and worse roads. Wider, high-quality, pedestrian-friendly roads with better safety measures are much more needed than the publicity points you get from using the “AI” buzzword.

More rules would mean more violations and more people searching for alternative modes of transport, like public buses and trains. Does Kerala have the required public transport infrastructure to support this new influx of travelers? No.

The pandemic catalyzed the decline of the public transport system in Kerala. The system was already falling under with more people opting for private vehicles over the past decade ¹³. It is challenging to bring this back up organically.

There is also a lack of personnel to aid the functioning of these cameras ¹⁴.

Having the basics walking well before trying to run with technologies like AI is a logical step the government seems to have missed.

Prevention is Better

The goal of the Safe Kerala project is to make roads safer in Kerala. A good indicator to track this goal would be the number of accidents.

Nudging people to wear helmets and seatbelts will reduce deaths caused by accidents but not the number of accidents. But if we could prevent accidents in the first place, it would automatically reduce deaths caused by accidents.

The project is trying to optimize for the wrong indicator without considering how it could better impact the broader goal of making the roads safer.

Arbitrary Implementation

The 700 cameras do not even come close to covering Kerala’s 3,00,000+ km of roads.

Any attempt to place these cameras will be arbitrary, meaning some people will be regularly monitored while some others won’t be monitored at all.

So, in practice, the traffic rules will only apply to a subset of the drivers in Kerala.

Better Alternatives

If the goal is to make our roads safer, there are far better ways to allocate resources than AI traffic cameras.

For instance, the Motor Vehicle Department could use these resources to improve the driving tests in the state and ensure new drivers are fully equipped to drive safely. This should nip the problem in the bud and produce better drivers and, as a result, safer roads.

You could argue that the results could take a while to show with this solution. But, even then, the government could have spent these resources to improve existing roads.

The public transport system of the state also seems like a better recipient of these resources. Increasing the frequency of buses on existing routes and starting service on new routes would encourage more people to use public transport.

Other states in India have also leveraged AI tools to improve road safety.

The iRASTE project in Telangana uses AI to build driver assistants that can predict accidents and alert the driver. The project was initially rolled out in 14 buses and will be extended to 200 buses. These AI systems can also continuously monitor the road infrastructure and provide feedback to improve and maintain roads to avoid future accidents ¹⁵.

The Karnataka State Road Transport Corporation also has a similar project that uses AI to detect collisions, driver drowsiness, and lane departures ¹⁶.

Both projects are far more direct in preventing accidents than the Safe Kerala project.

It is also interesting to see that NITI Ayog published a strategy guide on how to adopt AI in India, detailing how government can use AI to improve areas like healthcare, agriculture, education, infrastructure, and transportation ¹⁷ without ever mentioning using cameras this extensively to detect traffic violations.

There could be even more viable alternatives to spend money on than AI cameras. Did the government really think this was the best idea?

Unsound Technology

AI, although having existed since the last century, grew and became what we know now in the past decade.

The technology is so novel that we kind of know how it works, but not really.

Governments worldwide are trying to get ahead of the AI revolution and control its exponential growth. India’s IT minister Ashwini Vaishnaw recently announced that the government has no plans to regulate AI ¹⁸, which is a good decision for now.

But this also means that the AI used in these camera systems is built and operated at the discretion of the maker, a private company. We have no clue about the data used to train the AI or how accurate the AI is.

It does not stop at accuracy. AI models are also prone to inherent biases due to imperfect training data and training of the model. Meaning the system can be discriminatory to a particular group of people, and there is no way to prove otherwise reliably.

The government doesn’t provide or enforce any quality benchmarks to evaluate this black box of technology before using it directly on the public. NITI Ayog released a set of guidelines in 2021 ¹⁹ for building responsible AI, but since these are just guidelines, we cannot know if the AI camera system follows them.

The proposed Digital Personal Data Protection Bill ²⁰ could also impact the future of software, especially AIs, as they use a large amount of data. But the central government can exempt state governments from the provisions mentioned in the bill on specific grounds, which could be the case for the Safe Kerala project.

Unintended Consequences

Shortsighted policies often result in unintended consequences.

Discouraging more than two passengers on a motorcycle is a good idea considering the safety of the passengers and fellow drivers on the road.

But there are a lot of families consisting of two parents and children that rely on a single two-wheeler for transport. Strict enforcement would force such families to use the public transport system, which wouldn’t be a problem if the system is sound.

Yes, people shouldn’t travel with three other people on a 50cc scooter, but what else can they do?

These consequences will disproportionately affect people with lower incomes.

Privacy Violation

The current laws ²¹ for using monitoring cameras in India allow a lot of room for abuse without accountability.

The government still needs to address such potential privacy concerns as these traffic cameras increase in the state.

The state and the central government should also invest resources in building legislation to regulate the usage of AI on the public. Jumping the gun and implementing these technologies without fully understanding or being able to control their implications can quickly turn catastrophic.

Traffic Rules Aren’t Bad

The purpose of this article was not to say that enforcing traffic rules for safer roads was a bad idea but to argue that AI traffic cameras were not the right solution to the problem for multiple reasons.

These reasons suggest shortsighted decision-making by the government for publicity points which they could have easily avoided if someone with clout had asked, “Do we need to do this?”