Reinventing the human brain: How A.I. will revolutionize driverless cars
If a driverless car is to truly master any traffic scenario, it must become as smart as a human. Or smarter. Artificial intelligence expert Dr. Nicolaj Stache reveals how carmakers are planning to pull off this audacious task.
Picture yourself in the future, in your brand new self-driving car. Your ride is easily able to accelerate, break and steer itself through the sharpest bends with ease. You lean back and enjoy the drive…
Now imagine that your car is approaching a crossing where a cop is regulating traffic. He signals your car to stop. But your vehicle does not realize he’s a policeman. It can’t tell him apart from a random pedestrian and chooses to ignore him. You might just be in real trouble!
This is only one of many scenarios that require driverless cars to not only become all-seeing but, in fact, truly intelligent. We talked to a man who knows how to teach them: Nicolaj Stache is heading the Artificial Intelligence Center at Continental.
2025AD: Mr Stache, what is your vision for artificial intelligence (A.I.) in automated driving?
Nicolaj Stache: The vision that drives us is to replicate the human driver – only without replicating human mistakes. In other words, we are aiming to substitute the human brain through artificial intelligence. That’s still a long way away, but we are working on it. (laughs)
2025AD: Artificial intelligence is on everyone’s lips, yet the term has remained rather intangible…
Stache: I think a widely accepted definition would be that a machine is intelligent once its behavior cannot be told apart from human patterns. The “Turing Test” can help determine that. In this experiment, a test person communicates with a second person and a machine via computer interface. If the test person cannot tell whether his counterpart is machine or human, the machine passes the test and we have an authentic case of A.I.
2025AD: Artificial intelligence emerged initially as a fast-rising field of research, but was all but abandoned in the late 1980s. How do you explain the incredible comeback during the past 15 years?
Stache: In the late 80s, A.I. research hit a brick wall. Scientists envisioned forms of A.I. that simply could not be achieved with the technology at that time. To advance A.I., you need tremendous computing power – thanks to high-end processors and cheaper technology, the tide has turned.
2025AD: How can automated and driverless cars benefit from A.I.?
Stache: Wherever cars will be required to understand complex traffic scenarios, we should consider A.I. Especially in urban environments like city streets and busy intersections, we have many moving objects that automated and driverless cars must recognize and react to in an appropriate way. A.I. will help future cars to anticipate the intentions of the many other road users and foresee their next moves. Fully comprehending these demanding traffic contexts will indeed be the supreme challenge for A.I., but it will make traffic safer and more efficient.
2025AD: Can you reveal what you are currently researching at the Artificial Intelligence Center?
Stache: Our current focus is on the perception and especially the interpretation of said traffic scenes, in particular for automated driving: how do cars detect and properly identify other road users and obstacles even under challenging conditions? How can a car for instance distinguish between a police officer and a regular pedestrian? Our work touches on many different areas in this respect.
2025AD: How exactly do you go about that?
Stache: For instance, we feed a car’s neural network with training data – like pictures of traffic participants. Then the computer is put to the test in real-life scenarios where it must recognize and classify each participant by applying the trained network. It must then take the right action, for example brake if necessary.
2025AD: What do you mean by neural networks exactly?
Stache: Imagine neural networks serving as a car’s brain. To put it simply, they are structures of countless units of artificial neurons connected to each other. Each unit has parameters to influence the relationship between its inputs and its output. The optimal set of parameters is task-specific and can be determined by numerical algorithms – this process is called “training”. Once the network is trained, it can be used to classify new input, i.e. previously unknown information. Such input could be for example raw data coming from the car’s sensors, which is classified into several categories such as cars, trucks, road boundaries, people and others.
2025AD: Will cars be able to learn from their own decisions – or even from their mistakes?
Stache: This is possible – at least for virtual cars in a simulated environment. For real cars, there is still research to be done. First, for safety reasons, it must be ensured that the change of the parameters made by learning does not have negative side effects. Another prerequisite is to enable the vehicle to recognize mistakes of the system all by itself. Only if this is the case can it learn from an error. For example, in simulation or in computer games this already works: Here the system can recognize good and bad behavior simply by the score of the game.
2025AD: That seems like it could create tons of opportunities. Will future cars also be able to learn about their drivers?
Stache: Absolutely! A car may recognize its driver through biometrics, address him by name and even sense his mood and state of mind. Imagine you are getting into a car wearing a heavy coat and feeling hot; or you are under pressure and feel stressed out. Your car may intelligently adapt parameters like seating, temperature and even speed and route to make the driving experience as pleasant as possible.
2025AD: Will the car also adjust its driving style intelligently?
Stache: That is something we are working on. The car learns the unique driving style of its human owner – and imitates it when in self-drive mode. We can assume that this will create the most pleasant automated driving experience for the human.
2025AD: So, if I tend to rev the engine and make the tires screech for a racing start, will my car then do the same?
Stache: (laughs) …well, in due consideration of all safety aspects! Your car will obviously adhere to given law and never do things that threaten you or other traffic participants.
2025AD: So besides taking care of me as a driver, will A.I. also compensate for my mistakes?
Stache: Definitely. For example, if you accidentally leave the lane, it will help you back on track properly. Current driver assistance systems can already do that. A.I. will be all about safety and comfort – imagine automated cars to be your guardian angel and an intelligent extension of your own self at the same time.
2025AD: Science fiction especially has been thriving on A.I. from day one. How does A.I. in reality differ from how it is depicted in movies?
Stache: This is also a cultural phenomenon: in the Western world, A.I. is often depicted as an evil force to threaten humanity. In Asian countries, robots are seen as our friend and helpmate. Depictions of A.I. are a projection of human fantasy and imagination. It is our job to show how in automated driving it will benefit society as a whole.
2025AD: Can A.I. become too smart – to a point where it might become scary?
Stache: At the end of the day, A.I. will be all about making traffic safer: by helping minimize road deaths and maximize safety. I do not think A.I. can be too smart if it is restricted to a particular task only. If our cars were starting to make evidently wrong decisions, they would simply not be intelligent enough.
2025AD: How do you gain acceptance for the technology?
Stache: A.I. faces extreme safety requirements, which is why we are approaching it with the utmost caution. It will be important to have security frameworks for A.I. so it never threatens or hurts people. This is the right way to win over people’s hearts.
2025AD: Your vision for A.I. is to match and even surpass the human driver: will this mission ever be accomplished?
Stache: I am sure that through the better perception performance of A.I.-based systems, i.e. a better “understanding” of the reality, we will be able to reduce the number of traffic fatalities significantly. That is the mission that counts to me. But I don’t think we’ll be able to relax and say “we’re done” any time soon. Each saved life counts!
What’s your view on automated driving and artificial intelligence? Would you trust it? If not, what would it take to make you feel more comfortable with it? Tell us in your comments below!