“By 2020, there will be 10 million self-driving cars on the roads,” Business Insider predicted back in 2016. A year earlier, the project manager for the development of self-driving cars at BMW said that after 2020, the company will introduce a feature that will allow the driver to check email, talk on the phone, or read the news instead of driving.
At the suggestion of the media and visionaries like Elon Musk, the self-driving car project has become the most publicized techno experiment of the 21st century. But so far, no manufacturer has gone beyond conservative benchmarks.
I closely follow the development of the auto industry, which includes one of the projects of our holding. I will share my vision of why autopilot is entering our everyday life so slowly.
AN IMPERFECT ARTIFICIAL INTELLIGENCE
Almost all developers are sure that machines with a completely autonomous control system can be created, but they can’t say when exactly it can be done.
“One reason for these repeating cycles is our limited understanding of the nature and complexity of intelligence itself…”, writes the Santa Fe Institute’s Melanie Mitchell.
In the 1960s, the AI sector was inundated with investment. But after 20 years, when it became clear that it was not possible to create stable algorithms, investors switched to other projects, explains Mitchell. This could happen again now: Mitchell cited recent surveys of AI researchers who predicted “a 50% chance that human-level AI would be created by 2040–2060.”
Another problem is data for machine learning. Algorithms are trained on data from DVRs about different traffic situations, but collecting this data is quite expensive.
AI systems also generate unpredictable errors in situations where the real picture differs from the training data. For example, a car learns to bypass pedestrians but does not know how to react when a deer jumps out onto the road or a trash can pops out of nowhere. Since these incidents happen less frequently, there is less data available to train the neural network.
Waymo is a self-driving taxi service in Phoenix’s East Valley. The company chose this place for its good internet coverage, excellent weather conditions, and wide, deserted streets. And they have operators constantly monitoring situations in case of emergency.
Imagine how this service would work in, for example, Kyiv—a city with significant elevation differences, heavy traffic, narrow streets full of pedestrians, and imperfect roads.
Based on this, the most logical idea is to build a separate infrastructure for self-driving vehicles with good lighting, markings, and sensors. It sounds like a trillion-dollar science fiction movie rather than real life, especially in crowded metropolitan areas.
The minimum program is to allocate lanes for self-driving transport on current roads to limit areas and speed. This is what test cities are doing. AI is still too delicate to freely infuse into the traffic controlled by unpredictable people.
Developed countries are working to legalize the use of autonomous vehicles, but progress has been slower than technology. For example, the U.K. promised to allow hands-free driving with automatic lane-keeping and speeds up to 60 km/h by the end of 2021. In its new project, the Japanese government has set a goal to use tier four self-driving cars in more than 40 locations by 2025.
In the spring of 2020, the United Nations Economic Commission for Europe passed the first international autonomous vehicle regulation, giving the go-ahead to use a lane-keeping system on highways with separating bumpers.
In the U.S. at the federal level, only guidelines from the National Highway Traffic Safety Administration are in effect for automated vehicles so far. Local rules of states are different.
The question of responsibility remains a cornerstone in all jurisdictions: Who is to blame if the self-driving car crashes into another vehicle or knocks over a pedestrian? The owner, the person at the driver’s or passenger’s seat, the car manufacturer, the software manufacturer, etc.?
No clear rules are defined. Insurance companies also need to adapt their approach. Another blocker is that people don’t believe a robot is safer than humans. Still, KPMG predicts that the use of self-driving cars could reduce the number of accidents by 90% by 2050—by reducing human errors.
DESPITE ALL THE PROBLEMS…
Today, the self-driving vehicle market is estimated at $10 billion. According to forecasts, it will grow by 41% per year, reaching $326 billion by the end of 2030.
Self-driving cars are in demand in taxis, delivery, and cargo transportation. For example, the freight market will grow on average by 10% per year.
However, market players themselves cannot name the exact date autonomous vehicles will become mainstream. Chris Urmson, CEO of Aurora, which bought Uber’s self-driving unit, believes the technology will take 30 to 50 years to deploy.
In 2018, Waymo wanted to increase its fleet to 82,000 vehicles. But a year ago, the company had only 600 cars at its disposal. It is not reasonable to expand the fleet while the technology is still developing.
Price is yet another barrier. For example, a Jaguar I-Pace equipped with Waymo sensors costs $130,000, while an original new car costs around $75,000.
Let’s not forget that previously, riding in the backseat of your limousine was the norm for rich people. And now wealthy car owners take the driver’s seat. Luxury brands release flagships with a focus on the driving experience.
No one knows for certain when a real autonomous vehicle, which will freely take us on any road, will appear. For the next 10-15 years, most of us will definitely have to drive our cars ourselves. But isn’t this why we bought them in the first place?
Yura Lazebnikov, Managing partner of TECHIIA Holding / Investor in Tech innovative projects