Driverless cars to learn the meaning of fear
TECHNOLOGY Computers with artificial intelligence are taking on ever more tasks for us, and the prospect of autonomous driving, with a computer in control of our car, is fast becoming a reality. Self-learning systems could make computers safer, better motorists than humans if they can adequately sense the car’s surroundings and observe traffic regulations and speed limits. But according to Microsoft researchers, the computers still lack a crucial quality: they do not feel fear.
The researchers aim to change that. In an experiment they had a computer monitor a series of test subjects in a car simulator. The subjects wore pulse meters, using the pulse as a simple indication of the mental state of alarm. By allowing the artificial intelligence to link individual events such as near- crashes with the subjects’ fear, the computer was later able to repeat the same journey and was then more careful and drove very cautiously in the places where alarm had been registered in the test subjects.
The scientists compared the computer with a similar computer which had not been taught by human fear, only by its own driving errors. The results revealed that the computer with a human fear input was much quicker at becoming a safe autonomous motorist, reaching the same level as the other computer with 25% fewer accidents.
SELF-LEARNING SYSTEMS are often used in connection with artificial intelligence, where computers learn from experience.