Machine learning is used to draw up predictive models from datasets. An email client can for example file certain messages directly into the spam folder: having gone through thousands of messages, the software has learned how to identify junk mail.
One of the conferences at this year’s Microsoft TechDays in Paris looked at how applying this offshoot of artificial intelligence can allow businesses to improve their performance.
It may sound like something out of a science fiction film, but we already use machine learning applications in our everyday lives: every time we make a search on Google or play on a game console with gesture recognition, for example. The concept in itself isn’t actually that recent either: Microsoft has been working on the possibility of evolving a machine’s behaviour by analysing huge volumes of data for the past fifteen years or so.
With machine learning, machines can perceive their environment and recognise objects, whether natural language or handwritten elements, faces, shapes, etc., something which is impossible to achieve with traditional algorithms. Provided that the data set is large enough, it is possible now to create predictive mechanisms by closely reading and observing the data, whatever its nature.
In practical terms, machine learning can be used to create tools that can detect credit card fraud, determine how successful a product launch will be, conduct precise stock market analyses, classify DNA sequences, monitor whales by recording their calls or detect psychopathy by analysing tweets.
BUSINESS applications ALREADY AVAILABLE ON THE MARKET
We’ve come a long way since Garry Kasparov could beat Deep Blue, the chess-playing computer developed by IBM in the 90s. The latest genius of American enterprises is Watson, an artificial intelligence programme that understands natural language; Watson Analytics is a cognitive computing platform with powerful predictive analysis tools for business use.
So machine learning isn’t just for analysts and data scientists: marketing, sales, finance and HR professionals without a scientific background can now understand and interpret company information by making simple natural language queries.
Product managers, marketing professionals, Account Managers and HR Directors can thus ask questions such as: what are the main sales drivers for a product? What sort of perks are likely to keep an employee in a company? Which prospects are the most likely to become customers?
To come up with an answer, Watson Analytics looks at the data supplied by the user, gives a score on the cleanliness of the data and says how much potentially interesting analysis could be derived from the set. It then looks for any interesting correlations that may be occurring and displays them as graphs. Based on the type of results the user chooses to visualise, Watson Analytics gets a better idea of their areas of interest. It will then come up with tailored analyses to optimise a marketing campaign, offer better services to clients or more generally, to make decisions based on various predictions.
Watson Analytics’ rival over at Microsoft is Azure Machine Learning, a predictive analysis application which, like Watson, enables users to explore data and make forecasts, irrespective of their level of data scientist skill: it uses an intuitive interface, drag-and-drop gestures and simple data flow graphs.
The algorithms are the same ones used for the Xbox game console or Bing’s search engine, but it offers the same potential as IBM. The only difference is Azure Machine Learning is designed to be incorporated in to other Microsoft tools such as Azure Storage, its cloud-based storage solution, or more technical services such as HDInsight which can process huge volumes of data.
PredICTING LIFT BREAKDOWNS OR READING CLINICAL rEports
“Machine learning has the ability to totally transform business models.”
For Bernard Ourghanlian, Director of IT Security at Microsoft France and master of ceremonies at TechDays, machine learning will soon be vital for businesses.
A few cases in point: machine learning means that the US Post office, USPS, can now automatically process 98% of mail with a handwritten address (compared with just 10% before). ThyssenKrupp Elevator, meanwhile, can repair lifts before they break down, thanks to the technology. And IBM’s Watson was enlisted by a cancer centre in New York a few months ago to read and analyse millions of pages of medical journals and clinical reports.
Of course, not all uses of the technology are quite so spectacular: the most common applications of machine learning are far more prosaic, typically in the e-commerce sector: predictive analyses of shopping baskets, cross-selling and recommendations (“You may also like”), estimating shipping costs and lead-times, and fraud detection and prevention.
During this year’s TechDays, French startup Alkemics claimed to have reinvented the client experience thanks to machine learning. Alkemics’ collaborative platform enables e-tailers to enter detailed information about their products and add photos etc. and subsequently coordinate and target their merchandising more effectively. Distributors, meanwhile, have added-value product information and users enjoy an enhanced purchasing experience.
DATA AND EthiCS
To describe this technology, which is ubiquitous and yet virtually invisible, Bernard Ourghanlian talks of ambient intelligence and quotes Mark Weiser, the ‘godfather’ of ubiquitous computing:
“The most profound technologies are those that disappear. They weave themselves into the fabric of everyday life until they are indistinguishable from it.”
It does naturally raise a number of ethical questions: will these smart machines invade our private lives? How is our personal data protected? The French government debated this very issue last September and published a report on digital technology and fundamental rights.
Photo credit: r2hox – data.path Ryoji.Ikeda – 4 (Flickr.com, licence CC BY-SA 2.0)