The Perils of Shorthand
You’ve probably heard mention of the October 2012 Harvard Business Review article that described the data scientist as ‘The sexiest job of the 21st century’. There is undoubtedly a certain cachet attached to sexy jobs. To announce that you’re a fighter pilot, a tech entrepreneur, or a surgeon, elicits approving nods at cocktail parties as a small, adoring crowd gathers around you to hear what you have to say.
Not so sexy Data scientists don’t quite garner that level of adulation - at least not in the circles I frequent. However there is a veneer of understanding attached to the profession that is as fascinating as it is potentially damaging.
Try to explain Artificial Neural Networks to a guest at a cocktail party and your listener’s eyes will glaze over as she looks over your head to where the fighter pilot, the tech entrepreneur and the surgeon are holding court. In contrast, the entirely unhelpful but succinct declaration that ‘I work with Big Data’, invariably provokes a relieved smile of understanding.
The cost of shorthand There is a real cost to a business of reducing its understanding of data science to an erroneous set of heuristics. Data Science = IT and Data Science = Big Data are labels we’ve often heard. Not only does such superficial shorthand limit a business’ motivation to truly engage with the data science solution, it also makes the business executives receptive to the hype and fantastic claims of modern-day peddlers of snake oil, parading as data scientists. Meaningful collaboration between business and data scientists has never been more important.
To illustrate, consider the work of computer scientist, Dr. Marwa Mahmoud. She and her team at the University of Cambridge use machine learning to estimate a sheep’s pain through facial recognition. While this may seem to be a trivial use case at first glance, Dr Mahmoud’s research could find application in the early detection of debilitating diseases such as foot rot.
To the untrained eye, a contented sheep looks much the same as one in distress. A placid face, large unblinking eyes and an occasional twitch of the nose. Not being a vet herself, Dr Mahmoud had to rely on a vet to classify the images she used to train her model by relating the nuances of a sheep’s facial expression - e.g. ear raised or flat, nostril v or u shaped - to the level of pain experienced.
So it is with business Generally, a business executive will know much more about the intricacies of his or her business than a data scientist. A banking customer’s flight path to dormancy for example, will be part of the organization’s institutional knowledge and will not be immediately apparent to a data scientist. Unless this knowledge is shared, tested and internalized in joint teams, the data science solution risks missing the mark.
In the field of education, machines are being set the task of personalizing education by learning about pupils from the data the teaching process generates. But machines aren’t doing this by themselves. They’re aided by advances in psychology, cognitive science and other related disciplines.
Neural networks do it blindfold In case you thought the perils of shorthand afflicts only business executives, you’ll be pleased to know that data scientists can be equally at fault. Some machine learning techniques - neural networks are a good example - require little or no understanding of the internal workings of a system. Within limits, a neural network is just as happy figuring out how to bake the perfect cake as it is diagnosing cancer. And paradoxically, the robustness of these methods fuels the hype around data science and leads to the ‘I can solve everything’ claims bandied about by inexperienced data science practitioners.
So the next time a data scientist starts a conversation with you at a dinner party - actually it will probably be the other way around - eschew the shorthand and think about how you might collaborate. Your business will be glad you did.
References:The Economist, 22 July 2017. Briefing: Edtech