Artificial Intelligence is becoming less science fiction and more science fact. For all the Hollywood hype, it is already very much with us. There are a number of developers – Google, for example – making significant strides in the areas of machine learning and Artificial Intelligence (although the two terms are often used interchangeably). Elsewhere, IBM have just announced a self-determining $5 million competition to develop AI. There is even an academic Journal dedicated to the subject – The Journal of Artificial Intelligence Research.
The IBM brief is just to demonstrate usable AI by 2020. Meanwhile, although not strictly an exercise in AI, bankers HSBC have just announced the trailing of voice identification software for their telephone banking service. They use what they call identification of 100 different ‘signifiers’ to recognise an individual speaker’s vocal fingerprint.
Language is everything
The use of the word ‘signifiers’ is – if you’ll excuse the repetition – significant. It’s a term coined by the father of linguistics Ferdinand de Saussure and it goes right to the heart of what until now has been the great divide separating human communication and computer code. In a nutshell, we humans use our signifiers (words, sentences, texts) pretty loosely. We use words casually, imprecisely and creatively. And to make matters worse, we change their meanings (technically ‘signifieds’) over time. What is more, according to a recent study by Professor John Sutherland of University College London, the rate of that change has never been faster.
As Google’s senior research scientist, Greg Corrado, told reporters in September, if you’re in the business of trying to write a system capable of routinely communicating with humans, that sort of imprecision can be difficult to deal with. Unless, of course, you’re Google. Assessing the significance of Corrado’s statement, industry expert Daniel Lee has pointed out that Google’s RankBrain is making precisely that leap across the communicative divide.
By combining a machine learning logic with the vast amounts of data (in this case written language) that Google is able to garner, its technicians are claiming to have developed a system that is capable of grasping the semantic as well as the systematic connections between words.
AI in action
A research paper delivered by Google’s Tomas Mikolov and colleagues uses the example of Rankbrain’s ability to pair capital cities with their appropriate countries (Paris – France, London – England etc.) on the basis of its own internal learning. No input has made those pairings or their relative hierarchical relation. Those relationships have been established by RankBrain’s ongoing machine learning as part of its assessment of the traffic passing through it.
In the short term, that sort of association building is designed to improve search engine usability and SEO. Lee, again, describes Google’s RankBrain in terms of searching for a discrete term, such as the Premier League, and having related concepts that will also be called into view e.g. Manchester United or Arsenal.
In that way, Google searching is starting to move one step closer to a conversation than a simple search function. But whilst we recognise that semantic leap as a facet of what we see returned in response to our searches, what is happening at a deeper level is far more significant. Google scientists are telling us that RankBrain is systematically mapping the associations, connotations and uses of different words and ideas and developing what is in effect its own understanding of the data that it scans. If you Google Deep Learning you’ll see just what they mean.