John Giannandrea, Google's head of machine learning, recently told a Google I/O panel at the company's developer's conference that we are currently 'kind of in an AI spring.'
Giannandrea evidenced this progress by citing speech recognition and image understanding. Google is investing heavily in machine intelligence - more so than it is any other technology - and has more than 100 projects currently in development internally. Among these are the more well publicized practical applications, such as driverless cars, as well as a number of more novel applications, among which is art and poetry.
The race for artistically creative AI has been going on for a while now, as tech giants move to imbue the technology with more human characteristics. Microsoft recently applied machine learning algorithms to 346 of Dutch master Rembrandt's paintings. Its analysis found that the Dutch master mostly painted men between the ages of 30-40 that looked to the right. It analyzed his use of proportions, lighting, clothes, and multiple other metrics, to extract the key features that defined his style. Using this knowledge, it was able to produce a completely original Rembrandt of its own - or at least some kind of aggregate of his work.
Google has gone one step further, attempting to create entirely unique art. This started with DeepDream, which uses artificial neural networks to learn how to recognize shapes in pictures. It fed images into the neural network and asked it to emphasize features it recognized - in this case, animals - to create a sort-of unique set of images that it unveiled last year.
While both of these could be said to create new images in some sense of the word, Google is now trying to take things one further and have AI create entirely new images without any assistance. It unveiled its ‘Magenta’ last weekend at Moogest in Durham, North Carolina, with the full launch expected on June 1. The project comes from Google's Brain AI group - which is responsible for many uses of AI in Google products - and will use their machine learning open-source library TensorFlow to train computers to create art. Although details are currently hazy, the demonstration seemed to suggest that the software would analyze other pictures, in the same way that Microsoft did with the Rembrandt’s although on a larger scale that creates a more unique piece - or at least as unique a work as a human would produce.
Reception for current AI-produced artwork has been roundly negative. Writing in the Guardian, Jonathan Jones said of Microsoft's Rembrandt painting : 'What a horrible, tasteless, insensitive and soulless travesty of all that is creative in human nature. What a vile product of our strange time when the best brains dedicate themselves to the stupidest challenges, when technology is used for things it should never be used for and everybody feels obliged to applaud the heartless results because we so revere everything digital.' Criticism of Google's efforts has been equally forthcoming, with many commentators comparing it to the kind of psychedelia you find on the door of a frat house toilet.
Looking at the art, it’s hard to disagree with these appraisals, and you would be forgiven for being skeptical about Magenta’s ability to produce anything better. There is also the question of whether or not you could consider it art at all. Of course, this depends on how you define art. Is it that it inspires a reaction in the viewer? They inspired quite a visceral reaction in Jonathan Jones, but if inspiring a reaction is all you’re looking for, you could equally consider stamping on someone’s foot to be art.
Tolstoy defined the activity of art as ‘based on the fact that a man, receiving through his sense of hearing or sight another man's expression of feeling, is capable of experiencing the emotion which moved the man who expressed it. Every work of art causes the receiver to enter into a certain kind of relationship both with him who produced, or is producing, the art, and with all those who, simultaneously, previously, or subsequently, receive the same artistic impression.’ In this sense, it is impossible to see how it could be art. The relationship Tolstoy refers too is a kind of transcendental relationship, one which is impossible to have with a machine - no matter how much films like Short Circuit and Ex-Machina try and convince you otherwise. However, the paintings do raise a number of questions. Such as what it means to be human, and what it is we really want from artificial intelligence. Are we looking to entirely render ourselves redundant as a species? Do we want artificial intelligence, or artificial super humans?
The question of what it means to be human is complex. George Orwell once wrote that the essence of being human is that one does not seek perfection. The primary objective of machines is to be perfect. To try and imbue machines with humanity would therefore seem doomed to failure. Art is the essence of humanity, and the best of it shows our flaws - something machines will never be able to understand by their very nature. Machines will only ever really be able to replicate and mimic. Much of what they produce will likely be nice to look at, but endeavors to have them produce art are doomed to failure.
Perhaps the more pressing question is why exactly we even want to? AI when developed as a tool for humans to live, is one thing, and whatever the ramifications it is still ostensibly providing a benefit. There is a clear benefit to driverless cars, for example. But attempts to use AI to produce art veers too close to AI as a technology that entirely replaces humans. Imagination is the most human of qualities, and art the manifestation of that. Attempts to give machines imagination is to render us almost completely useless, and even if it won’t produce a real work of art, is a step too far down a path that we probably shouldn’t be walking down.