'Mobile phone in 2035 as powerful as our brains'
Within 20 years, intelligent machines will play a major role in society. ‘Selfdriving cars will be 90% safer than human-driven cars and will change transportation globally,’ says artificial intelligence scientist Bart Selman of Cornell University. He gave the first Ada Lovelace lecture of the Leiden institute of computer science on 12 May 2017.
The developments are moving fast in computer science in general, and in deep learning in particular. Through deep learning, computers can be taught to recognize faces, and objects in images, and decipher human speech, just like young children do. With the use of complex ‘neural networks’, scientist train statistical models of our society.
Ada Lovelace lecture
Much of the current progress in artificial intelligence (AI) is driven by the visual development of machines. ‘Twelve years ago, computer vision was a mess. The first selfdriving car Stanley, from Stanford University, did not even use vision. Now, selfdriving cars can see traffic and other road conditions almost as well as humans.’ Bart Selman, prominent scientist in the field of AI at Cornell University in New York State, is speaking. He is in Leiden to give the inaugural Ada Lovelace lecture, a new popular science lecture series of LIACS, the Leiden Institute of Advanced Computer Science.
Selman is originally Dutch and studied in Delft. After graduation in 1983, he went to Canada and the US. He stayed there, for 34 years already now. ‘Usually, I can speak Dutch again properly after about 3 days.’ In his Ada Lovelace lecture, he gives his audience an overview of the developments of intelligent machines. In his view, the discipline started to accelerate about 20 years ago. Selman: ‘In 1997, IBM’s Deep Blue computer defeated the world chess champion Garry Kasparov. Deep Blue had the computational power of a mouse brain. Unfortunately, mice don’t like chess.’
Especially since 2011, a lot has happened in the world of computer vision. ‘Neural networks with more than 100 thousand parameters have been trained with over 200 thousand images. For that, you need the most powerful computers’, according to Selman. So-called graphic processing units (GPUs), originally developed for video games, have enabled training of complex neural networks on tremendous amounts of images labelled by humans. Selman: ‘Crowdsourcing teams of thousands of people are teaching machines how humans perceive the world.'
AI arms race
With AI moving into society, the new research field of ‘AI safety’ has been established. It is prof. Bart Selman’s expertise. Part of the work on AI safety is funded by private foundations and industry. Selman: ‘The investments in AI systems by Google, Facebook, Baidu, IBM, Microsoft and Tesla run into the billions. But the US Military also takes part, with 19 billion dollars budgeted. A real AI arms race is going on.’
The developments will eventually lead to a general presence of AI in society. Selman’s predictions seem fabulous. ‘In 2035, a mobile phone will give its user access to the computational power compared to that of the human brain. In 5 to 10 years, self driving trucks will be 90% safer than trucks with human drivers. The AI control will react in real time and will sense its environment continuously, without, for example, falling asleep. This will change transportation globally.’
One downside to the innovations may be the expected rise in unemployment caused by intelligent machines. ‘That is already an issue’, says Selman. ‘Technological unemployment is unpreventable, so best is to mitigate the effects. Selman: ‘Europe may in fact be better prepared for technological unemployment than the US. In Europe, including the Netherlands, there are some experiments with a form of universal basic income. The US is likely less open to such proposals.’