How Big Are the Problems of Intelligent Driving?

"When we checked the backend data, we found that the test was run a total of 5 times, and the one everyone saw was the most extreme and unsuccessful attempt."

When the commune mentioned the recent "Dongchedi intelligent driving test, which crushed all domestic intelligent driving," the head of intelligent driving at Company A said that nowadays, all are intelligent connected vehicles, and every action and situation in the test will be recorded.

Strangely, however, the previously always high-profile legal departments have collectively fallen silent, even though they hold complete evidence of Dongchedi's "selective" testing and lack of honesty and integrity.

Fortunately, it's Dongchedi—not only does it understand cars, but it also understands Chinese relationships and the Chinese people.

If it were some other small independent media outlet doing this—throwing cold water, or rather, throwing mud at domestic autonomous driving technology—the test video would probably have been taken down long ago, and the "haters" would already be in trouble. In the current public opinion environment, it's just too easy to label someone.

The car manufacturers that were "heavily criticized" are certainly aware of the backing behind this test. With CCTV's endorsement and simultaneous statements from the Ministry of Public Security, along with high-profile production and exclusive use of highways, it can be said that no other media or organization has this capability.

So in the end, collectively they dared to be angry but did not speak out. Except for a few car companies that couldn't hold back and responded with meaningless words, the vast majority of car companies swallowed their grievances.

Moreover, with Musk flaunting his top achievements and stoking the flames, the public opinion quickly escalated from domestic to international.

Trump supporters and patriots have always been at odds with each other, so their fans are fervently engaged in mutual criticism. They analyze frame by frame, look for flaws, and search for underlying connections, with the best performers certainly being the sponsors. Those who haven't passed any criteria yet should consider contributing more.

The public opinion was in turmoil, and different groups of people surely had different views. Some saw rankings, some saw conspiracies, some saw gaps, and some saw safety.

01Public opinion and traffic

“If you haven’t controlled the variables, what’s the point of doing the test?”

I believe many people have seen such evaluations. Indeed, during actual testing, each vehicle's speed, the following distance to the car in front, and the timing of the lead vehicle's lane change all vary.

In reality, each car manufacturer has its own differences in intelligent driving assistance control logic and parameter settings. This is inherently a difficult task to control variables, not to mention the infinite variable factors during actual driving. Even if you control the variables of your own vehicle, the variables of other road participants are always changing.

The inability to control testing standards and variables actually to some extent indicates that there is currently no unified safety standard for intelligent assisted driving in China. Everything relies on the automakers’ and intelligent driving suppliers’ own understanding of driving and road conditions. For a technology that concerns the safety of life and property, this in itself is problematic.

Leaving aside the testing of intelligent driving assistance systems, even when we look at the safety crash test standards established over more than 130 years of automobile history, a vehicle that achieves a five-star rating in laboratory crash tests is already impressive. However, it still cannot guarantee 100% absolute safety for occupants in the event of a real collision.

Even if crash tests have absolutely controlled consistent variables, in actual road traffic participation, no car accident can occur exactly in the manner envisioned by car manufacturers or technical standards. As a saying in the automotive safety industry goes: No two traffic accidents are ever exactly the same.

The issue of whether variables change or not is not the core of this autonomous driving test; influence and traffic are.

As Professor Zhu Xichan from the School of Automotive Studies at Tongji University believes, in today's world where traffic is paramount, if there's no collision, no one will watch, so collisions are necessary; creating conflict is what makes it interesting. Especially when nearly 40 cars are ranked together, this contradiction and point of conflict become even more significant.

In addition, giving Tesla the best results, mixed with patriotism, actual user experience, consumer cognitive biases, and media guidance, inevitably leads to public opinion.

From this perspective, the educational test conducted by Dongchedi holds certain educational value and serves as a wake-up call. This is not to exonerate Dongchedi; at least many viewers and netizens in the comments section have expressed similar opinions and views, which serves as an awakening to the understanding of intelligent driving safety.

However, the author believes that even if Dongchedi's test was intended as a popular science effort to warn everyone that driver assistance is not absolutely safe, the pursuit of sensationalism and negative public opinion can easily lead the outside world, including consumers, to reject intelligent driving and the technological progress behind it.

Whether it’s Huawei or “NIO, Xpeng, and Li Auto,” including companies like Geely and BYD, they all have ample data to demonstrate how many potential collisions have been avoided after implementing intelligent driving systems. This number is at least in the millions, and the contribution to safety is, of course, enormous.

Therefore, the danger does not lie in the intelligent driving system itself, but in the incorrect perception and usage, including the gradual trust and reliance process, the overestimation of the intelligent driving system's capabilities, and the neglect of the 0.001% risk.

Ultimately, no matter how harsh the lesson, it is a price paid for one's own cognition and abilities. This applies to driving by oneself as well as driving with intelligent driving systems.

02"I trusted it until the crash."

Many users are expressing that the self-driving system and capabilities of their cars are really strong, reducing the fatigue of driving, and they can even often trust it to drive by itself.

What truly alerted him was undoubtedly the extremely low probability event that happened to him personally. This kind of incident, where owners have accidents while using the intelligent driving system, has been encountered by almost every car company that boasts about how advanced their intelligent driving systems are.

Once an accident occurs, a common thought flashes through their minds: I trusted it until the moment of the crash.

One of my colleagues, who entered the automotive industry and encountered the rapidly developing new energy vehicles, was also very appreciative and interested in intelligence. Until one day, he got into an accident while using the intelligent driving system of a new startup. The instant the airbag exploded, it completely jolted him awake. Fortunately, he wasn't seriously hurt.

In retrospect, when he reviewed the accident, the core reason was that he relied too much on the so-called intelligent driving system. Initially, he used it with his eyes and feet, but as it became more convenient over time, he stopped paying attention, and then the problem arose.

Fortunately, this was an accident in the urban area at a speed of 60-70 km/h. If it had been on a highway at a speed of 120 km/h, the consequences would have been unimaginable.

The author still vividly remembers the "killing" remarks about autonomous driving made by Su Qing, a former technical expert of Huawei's ADS team and now Vice President and Chief Architect at Horizon Robotics: When machines enter human society and coexist with humans, they will inevitably cause accidents. Our aim is to minimize the accident rate, but in terms of probability, it is something that could happen.

Many aggressive engineers in the IT field believe that on the path forward, there's no need to be constrained by things with extremely low probability. They think it's better to proceed boldly and address any issues and bugs as they arise. However, engineers in the automotive field need to maintain a greater respect for life safety, as a single bug could potentially cost several lives.

The probability in tests is merely a statistic, but when it actually happens to a consumer, it is 100%, and their family cannot afford such an extremely low-probability event. It's no wonder that Su Qing, who works in autonomous driving, once almost fell into depression. On one hand, the industry's momentum pushes the technology forward, while on the other hand, the unresolved potential accident probability brings torment to her conscience and heart.

Especially when automobile manufacturers need marketing and need to sell cars to achieve commercial value, it is often very easy to lose the boundaries of intelligent assisted driving.

On one hand, they say intelligent driving is still in the assistance phase, yet on the other hand, they claim it is very safe, almost to the point of falling asleep. They launch intelligent driving insurance to encourage hesitant consumers to boldly use it, but when an accident occurs, the responsibility is shifted away from intelligent driving, ultimately making it the consumer's responsibility.

Ultimately, after intelligent driving is adopted, it remains a situation of "a thousand people, a thousand faces." Different road conditions, different scenarios, and different drivers will react differently to various extreme situations. In the current context where L2 driving assistance still holds the driver as the primary responsible party, the challenge that intelligent driving needs to solve is how to make cars safer.

Regarding safety, many statements at the recent World Artificial Intelligence Conference have been discussing this topic.

Stuart Russell, an American computer scientist and AI expert, calls for caution against the arms race mentality in AI development and emphasizes that safety should be prioritized to truly serve the long-term well-being of humanity.

Geoffrey Hinton, known as the "Godfather of AI," has also expressed concerns about AI safety issues. Although the development is rapid, the risks cannot be ignored, such as the probability of destroying human civilization. Additionally, some large tech companies lobby to relax regulations in pursuit of commercial interests, which is an extremely dangerous trend.

This series of viewpoints also seems to point to the current field of intelligent driving.