If We Demand Zero Accidents from Self-Driving Cars, They May Never Hit the Road

Another accident involving a self-driving bus was in the news.

In September 2026, a self-driving bus caused an accident involving property damage in Fukaya, Saitama Prefecture, and its test operations were suspended.

Whenever I see news like this, I wonder:

How safe does autonomous driving have to be before society will accept it?

Of course, these were test operations, and a temporary pause to check for defects is only right.

But the news report says operations have been “suspended indefinitely until safety can be confirmed.”

Who decides when safety has been confirmed?

Humans Cause Accidents Almost Every Day

Obviously, cars driven by humans cause accidents too.

Drivers miss traffic lights. They hit the accelerator instead of the brake. They fall asleep. They look at their phones. They make bad judgments.

Yet nobody says, “Human driving is dangerous, so let’s stop every car in the country at once.”

With autonomous driving, though, a single accident changes the mood.

“Is it really safe?” “Why continue the experiment?” “Should something that might cause an accident be allowed on public roads?”

Then come the suspension of operations, the investigation into the cause, measures to prevent a recurrence, and the safety review.

Investigating the cause of an accident is only right. The problem is what comes after that.

The Benchmark Shouldn’t Be Zero Accidents

When we assess the safety of autonomous driving, we shouldn’t compare it with an ideal world.

We should compare it with human driving.

Suppose human drivers cause 100 serious accidents per 100 million kilometers, while autonomous vehicles cause 50.

Those autonomous vehicles would still cause 50 accidents.

Those accidents would make the news. People would be hurt.

If we then said, “A self-driving car had an accident. It’s dangerous, so stop it,” we would be left with human drivers causing 100 accidents.

A technology that doubled safety would be rejected because it failed to eliminate accidents altogether.

At that rate, the technology could never be put to use in society.

Only a Machine’s Accidents Become Unacceptable

What’s interesting is that society responds differently to human error and machine error.

When a human driver causes an accident, it can be dealt with as the driver’s negligence. The police investigate, an insurance company pays compensation, and criminal liability can be pursued if necessary.

But when a machine causes an accident, the issue grows much bigger.

Is the manufacturer responsible? The software company? The operator? The local government? The authorities that approved it?

The question of responsibility suddenly reaches all the way up the chain.

So, for an organization, stopping operations becomes the safest choice.

If it keeps the autonomous vehicles running and another accident happens, it will be blamed for failing to stop after the first one. On the other hand, if stopping them delays development by six months, almost nobody is held responsible.

This imbalance is particularly strong in Japanese organizations.

The Gap with the United States and China Isn’t Just About Technology

The difference between Japan and countries ahead in autonomous driving isn’t just the performance of their AI or sensors.

It is also about assuming accidents will happen: deciding what level of risk to accept, who takes responsibility, how victims are compensated, and how data from each accident feeds into the next improvement.

Bringing a technology into society requires designing these arrangements too.

It’s especially important to move beyond the simple choice between “safe” and “dangerous.”

There is no such thing as zero risk in the real world.

Airplanes, trains, cars, and medical care cannot completely eliminate accidents or deaths. Society uses them anyway because it has weighed their benefits against their risks and built systems for responsibility and compensation when accidents happen.

If we impose a condition on autonomous driving alone that it must never cause an accident, the answer is simple.

Don’t let the vehicles run.

A Future Where Japanese Cars Can’t Drive Themselves in Japan

This leads me to imagine a rather unpleasant future.

Japanese automakers perfect advanced autonomous driving technology.

But in Japan, each accident causes an uproar. Trials are halted, authorities investigate, and confirming safety takes time.

Meanwhile, cars keep driving overseas. Accidents happen there too. But data accumulates, and the software improves.

A few years later, Japanese automakers are selling self-driving cars as a matter of course in overseas markets, while cautious trials are still underway at home.

Once the technology has built up a sufficient track record abroad, it is brought back to Japan as “the latest technology, proven safe overseas.”

Japanese people begin using a technology made by Japanese companies only after checking its record overseas.

If we reach the point of importing our own technology back like that, technical ability will no longer be the issue.

If Japan loses the race in autonomous driving, I think it will be because of a culture that cannot tolerate failure, not because of a lack of technology.