Better than humans?
How artificial intelligence advances autonomous driving
Chat bots, text robots and distribution drones – AI is already an integral part of our everyday life. It also plays a central role in automated and autonomous driving as it helps a car recognize its immediate surroundings, among other things. The new Audi A8 for example, uses the method of “deep learning” for its image processing; a method of Artificial Intelligence that takes its cue from the learning process of a human mind. During the programming phase, the software is trained to recognize relevant characteristics of objects, such as other vehicles.
One of the pioneers of deep learning is Sepp Hochreiter, professor at the Johannes-Kepler University (JKU) Linz. Already in 1991, together with Jürgen Schmidhuber, he developed the so-called Long Short-Term Memory (LSTM), a method of deep learning. The LSTM networks laid the foundation for the language recognition integrated in all Android smartphones today. Since then, Hochreiter and Audi have founded the “Audi.JKU deep learning center” to promote automated and autonomous driving. He believes that the great days of AI have arrived: “Due to the rapidly advancing digitalization, the amount of data is growing consistently while the processing power of computers allow us to keep up with the speed. AI systems are constantly learning and are better at many things than mankind has ever been.”
“AI will shake our lives up completely.”
- Sepp Hochreiter
Artificial Intelligence at Audi
“We are ready to trust a taxi driver – a complete stranger – more than AI, although AI can recognize and respond to any situation better and faster than a person.”
Deep learning and intelligence from data
In order to explain the special nature of autonomous driving at Audi, Hochreiter elaborates on the new AI technique: “Until now, the order of the day was to look at each individual image. However, AI must analyze all images in their context.” Another example: a cyclist signals his intention to turn on the corner, but lowers his hand a few meters before the corner. One image alone would make it difficult to interpret this situation. The car would probably make a break, even if the cyclist has not reached the junction yet.
AI must therefore learn to think ahead and to interpret the collected data within the situation’s context – in this case with the help of additional images of the road ahead, for instance. It has to learn to select – just like the human brain. Only with deep learning and the continuous collection and evaluation of data, AI can continue to develop and improve.
Building confidence in self-driving cars step by step
The biggest challenge of automated and autonomous challenge, besides the technical requirements, may well be the acceptance of the customers. To the AI pioneer Hochreiter, the hesitance in this area is inexplicable: “We are ready to trust a taxi driver – a complete stranger – more than AI, although AI can recognize and respond to any situation better and faster than a person.”
“Autonomous driving offers a tremendous opportunity to make driving safer.”
According to Roland Pfänder, manager of software development at Audi Electronics Venture GmbH, Audi must “gain the customer’s trust one step at a time”. Only in this way people can experience autonomous driving as a new form of mobility and thereby, turn the additional gain of freedom into reality. “Our goal is to raise the car’s degree of automation to a level that our customers see it as a significant relief, while experiencing the highest possible comfort”, says Pfänder.
Furthermore, Pfänder also sees a great responsibility towards society and the customers. “We have about 3.300 traffic fatalities per year in Germany alone. Worldwide, the number rises to a staggering 3.500 vehicles per day. Autonomous driving offers a tremendous opportunity to make driving safer.” Hochreiter adds, “Accidents can never be completely avoided, but with AI they will become more improbable. Sensors react better and faster than humans.”
In order to establish a trust in technology, Audi deals with the ethical, legal and social issues of automated and autonomous driving. To this end, Audi has founded the beyond initiative. In this interdisciplinary network, Audi experts exchange their views with experts from various disciplines. In addition to Sepp Hochreiter, Iyad Rahwan from MIT Media Lab and Luciano Floridi from the University of Oxford also participated in past events, to name a few. In this initiative, software engineers come across philosophers, legal experts meet psychologists. For Hochreiter, one of the key factors for social acceptance is to make people get used to the new technology: "People have to see and feel that it works."
The future is near: no traffic jam through networked driving
On paving its way towards autonomous driving, Audi has set the first milestones: the new Audi A8 is the first series-produced car specifically designed for highly automated driving (so-called Level 3) in specific situations. Meaning that the A8 can take control of the steering wheel and pedals in traffic jam. In order for automated and autonomous driving to actually be taken to the streets, the legal framework has to be adapted and or redefined. Pfänder has observed a positive development: “Legislators all over the world are already working on regulations for automated systems." What does the far future look like? Hochreiter’s vision is clear: “Cars will communicate with each other: where they are, where they are headed to, and with which speed. If such a thing is possible, traffic jams will be a thing of the past.” Even Pfänder sees enormous potential: “We are convinced that we will make the roads more secure through the use of Artifical Intelligence.”
This article was written in the scope of the Audi internal presentation “Responsibility Perspective”.