December 2019 could be remembered, in the future, as the real starting point of the new generation of Artificial Intelligence. The one that (perhaps) will give machines a real “life” – even if different from the biological one.
In the world of cinema it is said that the famous computer of “2001 – A Space Odyssey” (HAL 9000) was named in this way by the three letters of the alphabet that precede… IBM! In short, Stanley Kubrick and Arthur C. Clarke would have named “HAL” their supercomputer equipped with IA as a tribute to the famous IBM brand. Urban myth? Meanwhile, it’s certainly no coincidence that IBM Research is behind the research, carried out in collaboration with the Polytechnic of Zurich (ETH) and with the University of Bologna, which proposes itself as the watershed between the old and the new way of understanding the Artificial Intelligence.
We have been talking about AI for a long time, not only in science fiction but also in everyday life. In recent days, however, at the NeurIPS conference in Vancouver, the innovative work of three IBM researchers, Florian Scheidegger, Costas Bekas and the Italian Cristiano Malossi was presented. Why would their work be innovative? How does it stand as a turning point in research on AI? Simple: in Vancouver, the first AI algorithm able to automatically synthesize deep neural networks was unveiled, in order to significantly improve the efficiency of image classification systems in IoT devices, from intelligent home equipment to driverless cars.
The IA of the IoT devices, as we said, is not an absolute novelty. However, traditional IA models are not designed to meet the higher needs of the most innovative IoT devices, for various reasons: limited memory, high energy consumption, lower reaction speed. The most striking example is that of driverless vehicles: the ability to identify obstacles or people crossing the road is already there, but reaction times that are fast enough to prevent accidents are lacking, as demonstrated – unfortunately – by some tests that have been tragically concluded. The new AI algorithm focuses on the main performance requirements of a given use case and simplifies the search for efficient deep neural network models. In the context of driverless cars, IBM models for the classification of synthesized images would be executed in times in the order of milliseconds, thus giving the vehicle the time necessary not only to identify, but also to avoid a pedestrian.
In short, the future of AI is in the architecture of deep neural networks, and this research is the first, tangible proof. Will one day even the machines celebrate Christmas with us? Maybe! Certainly, for this year, we wish you Christmas greetings in a “classic way”, typing on the keyboard, and always with an eye to the future of technology and, with it, of society… We will take a short break, from December 24th to January 6th, to recharge our batteries for the new year’s professional challenges.
Merry Christmas to everyone (also to HAL 9000…)!