In this book [Artificial Intelligence, Human Intelligence: The Double Enigma], Daniel Andler sets out his stall right from the outset. Human intelligence is one enigma, artificial intelligence (AI) another, but neither of them is a mystery, and that is what he endeavors to prove. The reader will be the judge, but the merit of his approach lies above all in the fact that it clarifies exactly what is under discussion, identifies its roots, and deciphers its ramifications and their implications. AI is emerging as one of the most radical technological revolutions in human history.
The first of the book’s two parts provides an analytical overview of the genesis of AI, right up to its latest developments. First of all, a word about terminology. The author proposes to use the expression “artificial intelligence” — an oxymoron based on woolly concepts — as a new term for a tangible reality, just as we use the word “crowbar” without questioning the relationship between the “crow” and the “bar.” He then uses this term to mean, in turn, AI in the strictest sense, limited to the academic and professional sphere; an abbreviated form of AI that covers applications and algorithms based on deep learning; and finally, AI in the broadest sense, which encompasses all the contributions made by digital technologies to the performance of any given task. AI thus appears to be the driving force behind the digital ecosystem (the “digisphere”), which is developing to the point of creating an additional stratum in the human world.
Now let us consider the author’s approach. Thinking on the subject of human intelligence has developed over centuries without reference to AI, even if some have dreamed of intelligent mechanical avatars. AI is a recent phenomenon. Its appearance owes a great deal to conceptual advances in calculation, information, and testing, and to advances in cognitive science. At the same time, its development is inextricably linked to advances in computers, from which it has retained a fundamental duality: its uses are both theoretical and empirical. This duality is perpetuated in the projects of its creators: one is like a Promethean dream of a copy of human intelligence, while the other resembles an engineer’s dream, putting expertise to work in an empirical approach.
The former gave rise to classic or symbolic AI, which had some successes, but the applications of which nevertheless appeared limited. It has been supplanted by connectionist AI, which, to summarize, gets its abilities through deep learning. This is the form of AI that crunches high volumes of data and can uncover what the data conceal by exploring their regularities. It has no pretensions other than its statistical efficiency, but its undeniable performance gives it a solid foundation and realistic objectives.
