If you go to church this Sunday perhaps you'll hear the ancient Easter hymn Christ ist Erstanden in a setting by Johann Sebastian Bach. It is one of nearly 400 Bach chorales - harmonisations for four voices on recurring patterns. They can be sublime. But they are not always what they seem. A couple years ago, in a musical version of the Turing artificial intelligence test, composition students were found to be unable to distinguish between real chorales and those composed by a system called, inevitably, DeepBach.  

Michael Cross

Michael Cross

So what, you ask: Bach was the most mathematical of composers and his chorales perhaps the most formulaic of his work. A computer programmed to generate a stream of notes according to the rules of composition - 'no consecutive fifths', and so on - would sooner or later come up with something that sounded right.

Possibly. But that is not how DeepBach works. Neither does its cousin, The Continuator, which can pick up and improvise, Charlie Parker-style, around jazz riffs: try writing a program for that. These are applications of a computing revolution: systems to replicate human skills not by being programmed to follow 'if-then' rules but to learn continuously from examples and experience. All you need, to quote one author, is a learning algorithm that can roam a digital landscape. Most of the technology isn't particularly new; the breakthrough is in the sheer amount of data that these child-like learning systems can graze upon. Humankind now produces every two days the same quantity of data it produced from the dawn of civilisation to the year 2003.

As a web user, you have already contributed to this learning ecosystem. The characters you typed to find this website help train your search engine's algorithms. Those irritating 'I am not a robot' pop-ups which ask you to click on images containing shop fronts or traffic lights are creating a database from which AI systems that filter images - or guide self-driving cars - can learn. (Paradoxically, we are training a robot to pass an 'I am not a robot' test.)

I labour this explanation because I still encounter too many sceptics about artificial intelligence who base their assumptions on previous generations of systems, coupled with an understandable suspicion of silicon snake oil. But even if, to our secret delight, an over-hyped AI system does not at first do exactly what it says on the tin, we should remember that we are now in a different world from the old days of 'computer says no'. Machine learning systems break the mould in two important ways: they continuously improve with experience and, for the first time since the invention of computing, their outputs can amount to more than what is directly programmed in. This is why so much attention is rightly focusing on the design and content of the algorithms that guide their learning. The Law Society's policy commission on the subject is due to report in June. 

In a new book*, mathematics professor Marcus du Sautoy, examines whether machine learning can break the last monopoly of the human mind: creativity. Musical examples, cited above, indicate which way he believes the argument is going. Du Sautoy also examines the possibility of AI systems creating visual works of art, writing novels and rendering his own profession redundant, by beating humans at creating mathematical conjectures and axiom-based proofs. I'll leave the denouement to him, but against these developments isn't it a little arrogant to maintain that other types of professional creativity will always be beyond the reach of robots? It is up to lawyers to decide whether they want to be Bach or Charlie Parker.

*Du Sautoy, Marcus, The Creativity Code, 4th Estate.