Can artificial intelligence become creative?

The book. Marcus du Sautoy is worried about the progress of science. This English mathematician, also known for works popularizing his discipline, admits to having been very disturbed by the victory in the game of go of a machine against a human champion in 2016. It is less the victory – than we wasn’t expecting anytime soon all the same – only the way, original, which stunned him. The artificial intelligence, AlphaGo, developed by DeepMind, a subsidiary of Google, had imagined a totally unpredictable blow and outside the canons developed for centuries by humans. This shock reminded him of the one he had experienced some time before, when another DeepMind software had invented the winning strategy in the game of brick breakers, the same one he had discovered as a child. Have machines become creative? Could they make art, but also, more worryingly for him, demonstrate conjectures in mathematics?

To answer this, Marcus du Sautoy proposes to come back to the definitions of creativity, to describe the paths that programs take to approach it and to analyze the reasons why mathematicians could be “threatened” by artificial intelligence.

Creation by exploration or combination

The essay begins fairly quickly with the story of this epic victory at go, then come popular and demystifying passages on algorithms in general and artificial intelligence in particular. Very rich chapters follow on the various and abundant attempts of artist-programs in painting, music, singing and literature (with even a paragraph written automatically, hidden in the reading). By dissecting its secrets, we finally understand how algorithms can succeed in at least two types of creation: that known as “by exploration” (by repeating, varying rules and pushing the limits) and that known as “by combination” (mixing of style, cross-fertilization, etc.). The tireless machines can easily thrive in these fields. Provided you find the rules, which sometimes requires programmers tips, but also a lot of data.

This allows the author to finally focus on math, the activity of which he dissects, to the point of stripping it, simplifying it and coldly “mechanizing” it. And therefore to make it ready to be imitated by an artificial intelligence. Except it’s a bit more complicated than that and for preserve the suspense, we will keep the conclusion silent. We will content ourselves with observing that his final pirouette is an original way of responding to his initial concern, while opening up a rich field of reflections on the relationship between humans and machines.

You have 14.69% of this article left to read. The following is for subscribers only.

Source link