Safe Squares

Kind of following on from yesterday's post: today I started to read a book I bought last week: "The Mechanical Turk" by Tom Standage. It's subject being the [in]famous automaton constructed by Wolfgang von Kempelen that became the closest thing possible, in the eighteenth century, to a global phenomenon: the eponymous 'chess-playing' machine of the book's title. What prompted me to remark on this was a reference made in the book's Preface to an era [the book's year of publication: 2002] '...when it takes a supercomputer to defeat the world chess champion...' These days - just over twenty years later - you can play [virtual] opponents at the game that simply cannot be beaten by humans of any calibre, even on your smartphone. Back in Kempelen's day, the ability to play and excel at chess was seen as evidence of sentience and creative thought. Indeed, that mindset obtained well into the late 1990s, when the aforesaid supercomputers - programmed specifically for this one task - started slowly to push back at the boundaries of human ability in the great game.

Not only do today's strongest chess 'engines' far outstrip the abilities of the very finest human exponents of the game, continuing to learn and improve as they play; but they can be, and are, programmed to play at all levels of the game, in all styles, and make the same manner of mistakes that humans of like ability do, making them a useful learning tool at all levels of the game, including, most particularly, at the top echelon Grandmaster level: we have reached the point where humans are indeed learning from machines and improving in turn. The qualifier here, to my mind, though, is that nobody [very often, at least!] gets killed playing chess, and so it remains a safe sandbox for machine learning development, albeit within a highly constrained, vertical use-case: the jury is still out on many other spheres of AI development in that regard...