How far away is the dream of truly self-driving cars?

Tram Ho

You must have heard that fully autonomous vehicles will dominate all roads around the world in 2018. That didn’t happen, and you’re told that they will appear by the end of 2020. And when that again failed to materialize, it was only said that things were very close.

It seems that now, experts have started to think more realistically. According to a recent article in the Wall Street Journal, many computer science experts believe that it will be at least another decade before completely driverless cars become a reality, and others believe that they will probably never come.

Big problem

Driverless vehicle technology is a very difficult challenge to solve.

Tesla CEO Elon Musk once said:

We have to solve a big problem of AI working in the real world to create a fully autonomous, completely unattended system, that is, the entire road system today is designed for. for biological neural networks with optical image receptors

Despite that, Musk continues to stick to the idea that Tesla’s Full Self Driving feature will soon turn the cars it produces into fully autonomous vehicles.

Giấc mơ xe hơi thực sự biết tự lái còn xa lắm? - Ảnh 1.

FSD is an incredibly powerful driver-assistance feature, but no matter how close we get to achieving that goal, there’s no sign Tesla has come any closer. with fully self-driving technology than any other electric car maker.

Why were the previous experts so wrong?

In 2014, deep learning technology boomed thanks to the work of many computer scientists, including Ian Goodfellow. His work on developing and forming an adversarial generative network (GAN, a neural network that helps AI produce results by acting as both constructivist and schizophrenic) makes one think that it is almost as any level of autonomy can be achieved through the application of algorithms and neural networks.

GAN is the technology that enables modern AI to achieve results as impressive as humans, including DeepFake, the This Person Does Not Exist website, and many others.

The rise of deep learning is like a tidal wave pushing the field of AI research into the air, turning Google, Amazon, and Microsoft into AI companies almost overnight.

And according to experts, the AI ​​field will reach nearly 1 trillion USD in value by 2028.

But those results have yet to materialize into the transcendental technologies that experts like Ray Kurzweil have predicted. The AI ​​that works in today’s modern world is mostly very specialized, meaning it’s designed just to do a very specific job and nothing else.

When you visualize everything a human driver needs to do – from paying attention to your surroundings, to finding your way, to operating a vehicle – the problem becomes clearer: dozens of designs are required. (or hundreds) of specialized AI systems that work together, or find a way to create a universal AI (“one AI to rules them all!”)

What’s the future?

Inventing and developing a universal AI is likely to be an important catalyst for solving the driverless car problem, but doing so is as difficult as asking for three wishes from a genie.

That’s because “universal AI” is just another way of saying “an AI capable of doing any related task that a human can do”. And so far, the idea of ​​universal AI remains a distant technology.

Giấc mơ xe hơi thực sự biết tự lái còn xa lắm? - Ảnh 2.

In the near future, things will not change much. Driver assistance technology will continue to be developed at a moderate pace, and tomorrow’s cars will certainly be safer and more advanced than today’s. But it’s hard to find a reason to believe that they will be able to drive themselves on busy streets in the near future.

We will most likely see special areas distributed in major cities around the world, and also highways designed specifically for driverless vehicle technology in the next few years. But that doesn’t mean the age of driverless cars has arrived.

No car maker (besides the gamble that Tesla so eloquently announced) has shown it is moving towards developing a fully autonomous consumer car within the next five years. .

And it’s a sign that it’s going to be at least a decade or more before we see a purchasable, consumer vehicle that doesn’t have a steering wheel or any manual controls. .

Driverless cars are probably not impossible. But developing them requires more than just clever algorithms, trial-and-error computing techniques, and computer vision technology.

According to experts, we will need a different type of AI, a completely different approach, a major change in infrastructure, or all three, to bring the dream of truly self-driving cars closer. to reality.

Reference: TheNextWeb

Share the news now

Source : Genk