09/08/2018

Jordan: Beware of the artificial intelligence hype.

Michael Jordan , a researcher at Berkeley University, warned in his plenary session this Thursday (9) about the abuse of the term "Artificial Intelligence".

With the amount of data available in the world growing faster than processing power, there is indeed a major challenge in integrating computing, statistics, and what is called "data engineering." This challenge is addressed in popular debate through buzzwords such as "artificial intelligence" and "machine learning."

The problem, according to him, is that these words don't capture the nature of this challenge. "Many aspects of the problem have little to do with imitating human intelligence or understanding the brain," he says. "Look at current and future networked systems, on a planetary scale, for medicine, commerce, transportation, etc." They automate important tasks, but they are not intelligent in a way comparable to that of a human being.

According to him, the notion of intelligent machines that can come to life and become a Skynet like the one in the movie "The Terminator" confuses more than it clarifies. In a recent article he published on Medium , Jordan stated that this fascinating and frightening idea "unfortunately distracts us."

What actually happens, according to him, is "algorithmic decision-making under conditions of uncertainty, in large-scale networks and markets." However, these decisions fundamentally depend on the parameters provided, and can have unintended consequences.

These consequences can be seen as an effect of the wisdom embedded in the system and undermine the credibility of these algorithms, however useful they may be for scaling processes important for progress in all areas. "I believe that placing too much trust in these brute-force algorithms is misplaced faith," he wrote.

To meet the challenge, Jordan pointed out that concepts created three centuries ago are being used, but implemented through the most advanced technologies of today. According to him, it is necessary to overcome the obstacles of implementing these decision-making algorithms in the real world, where errors exist, so that they can be controlled using the best tools of mathematics.