Making robots more autonomous and obtaining Artificial Intelligences with a certificate of 'reliability', are 2 of the technological frontiers being worked on in the Leonardo Lab, the series of applied research laboratories developed by Leonardo as accelerators of digitization for industry and presented in Genoa.
"Supercomputing, thanks in particular to Davinci-1, is the backbone of all our research units", explained Alessandro Massa, Head of the Leonardo Labs on the occasion of the presentation of the 3 laboratories present in Genoa. The 10 research laboratories scattered throughout Italy, one of which in the USA, on the various technological frontiers are fueled above all by the potential offered by supercomputing in all its forms, both in standard programming and in 'intelligent' form.
Among the many sectors there is that of robotics, as in the joint lab developed with the Italian Institute of Technology (Iit): "one of the objectives - explained Massa - is to increase the autonomy of robots, make them more and more independent from operator and get to obtain machines with reasoning skills. Today we are able to have machines with very high performances but only in very specific areas, the so-called 'narrow' AI ". The goal now is that of a generational leap with "AI more similar to humans, capable of reasoning inductively and associating concepts".
Another frontier is to make AI 'certifiable': "to apply them in high performance or risky contexts. If, for example, I want to entrust them with the control of a plane, I must first of all be able to 'validate' the software. The problem is that for them same definition, AIs do not have explicit behaviors because they learn from data ". The question to be answered is: can I really rely on software that is not predetermined but learns and adapts to circumstances? In other words, making what happens inside the algorithm 'transparent' today is often incomprehensible, the so-called black box problem. "Being able to give this answer is very difficult - added Massa - is something similar to being able to demonstrate how a person will behave in front of an event for which he has been trained. Not easy but the direction in which we must go is necessarily this ".