Bengaluru is about to conduct perhaps the most ambitious experiment in artificial intelligence since the invention of the chatbot: putting a robot taxi on its roads.

This is a brave idea. Not because autonomous driving is technologically difficult. But because Bengaluru traffic has spent decades perfecting the art of defeating technology.

The robot will arrive equipped with cameras, radar, lidar, GPS, artificial intelligence and presumably enough computing power to run a small space programme.

It will then encounter a Bengaluru motorcyclist.

The motorcyclist will emerge from somewhere the robot had calculated was physically impossible, squeeze through a gap the robot had calculated did not exist, overtake it from the left, cut across its path and disappear into another dimension.

The robot will have learnt its first lesson in Indian traffic:

The road is not a system. It is a negotiation.

For a human driver, this is intuitive. A red light generally means stop. Unless traffic police are waving everyone through. A green light means go. Unless an auto-rickshaw has decided to stop in front of you. A lane marking means stay within the lane. Unless there are three two-wheelers, a bus and a delivery executive occupying it.

Artificial intelligence may be intelligent. Bengaluru traffic is adaptive.

Then there are potholes.

For a machine, a pothole is an irregularity in the road surface. For a Bengaluru driver, it is an inherited database.

He knows which potholes are harmless, which can damage a suspension, which have been repaired, which have been repaired twice and which have merely received a ribbon-cutting ceremony.

A robot may detect a pothole from 50 metres away. A Bengaluru driver can identify one at night, during rain, through a windscreen covered with condensation and tell you which ward is responsible for it.

The robot's real challenge, however, will be two-wheelers.

Autonomous driving assumes that other road users will behave within certain parameters. Bengaluru's two-wheelers will help the machine revise those parameters.

They will appear between vehicles, beside vehicles and occasionally between the vehicle and the concept of a vehicle.

A robot may calculate that maintaining a safe distance is essential.

The rider behind it may calculate that the safe distance is precisely the space between the robot and the car in front.

This is where Bengaluru's celebrated technology ecosystem may finally produce something useful: an app that allows the robot taxi to ask, “What exactly are you doing?”

The answer will invariably be: “Nothing.”

Pedestrians will provide another challenge.

A human driver knows that a pedestrian standing beside a road is not necessarily waiting to cross. He may be waiting for a friend, checking his phone, contemplating life or preparing to cross exactly when the car reaches him.

The robot, having been trained on millions of scenarios, will probably stop.

The pedestrian will then look at the robot, wonder why it has stopped and cross behind it.

The robot will restart.

Another pedestrian will appear.

The robot will stop again.

By the end of the journey, Bengaluru will have solved unemployment by creating a new occupation: professional pedestrian crossing the road in front of autonomous vehicles.

Then comes the monsoon.

Autonomous vehicles are designed to operate in rain. Bengaluru's rain has a more ambitious objective… to alter the topography.

A road that existed at 3PM may become a lake at 4 PM. A pothole may become a pond. A speed breaker may become an archaeological feature.

The robot will ask: “Is this a road?”

The passenger will say: “Normally.”

And then there are road diversions.

A Bengaluru roadwork sign might say: ROAD CLOSED. USE ALTERNATE ROUTE.

The human driver will look around and find three possible routes, none of which appears on the map.

He will follow a BMTC bus.

This is another skill the robot will have to acquire: follow the vehicle that appears to know what it is doing.

The most interesting encounter, however, will be between the robot taxi and Bengaluru's technology workers.

Imagine a software engineer entering an autonomous cab after spending eight hours teaching machines to make decisions.

The robot says:

“Destination?”

“Koramangala.”

“Estimated journey time: 28 minutes.”

The passenger laughs.

“Try again.”

The robot recalculates.

“42 minutes.”

“Still optimistic.”

Eventually the machine will learn the fundamental rule of Bengaluru: journey time is not a number. It is a range bounded at one end by hope and at the other by experience.

There may even be a new premium feature: Predictive Traffic Anxiety.

Before beginning the journey, the robot could tell the passenger:

“Your meeting is at 10. You are 8 km away. Based on historical traffic patterns, you should have left yesterday.”

That would be genuine artificial intelligence.

Yet there is something fitting about Bengaluru becoming a laboratory for autonomous taxis.

This is the city that taught India to believe that software could transform everything. It produced global technology companies, thousands of startups and a generation convinced that every problem was ultimately a problem of data.

Now the city has produced a problem that may not be entirely reducible to data.

Because Bengaluru traffic contains something algorithms struggle with: human improvisation.

A pothole is not merely a pothole. A hand signal from a traffic policeman can override a traffic light. A delivery rider can become a pedestrian. A pedestrian can become a cyclist. A cyclist can suddenly become a motorist. And an apparently empty road can become a traffic jam in the time it takes to say “Outer Ring Road”.

Perhaps the robot taxi will eventually master all this.

Perhaps it will learn that the horn is not an emergency signal but part of the city's conversational grammar.

Perhaps it will learn that “five minutes away” is a metaphysical concept.

Perhaps it will even learn to take a shortcut through an unfamiliar residential lane because Google Maps says so and then discover that the lane ends at a wedding.

That may be the final test.

If the robot can navigate Bengaluru traffic, survive its potholes, interpret its hand signals, negotiate with its two-wheelers, endure its monsoon and still reach the destination without asking to be taken back to the factory, artificial intelligence will have achieved something remarkable.

It will have become Bengaluru intelligent.

And that may be a higher standard than artificial intelligence.