Intelligent machines. It’s been a topic of popular fascination for some time. From the celebrity of the Mechanical Turk in the 18th century, to the success of Issac Asimov’s ‘I, Robot’ series in the 1940s. Computers that are capable of calculating, reasoning or even thinking have often captured our collective imagination.
As computing power continues to grow, so too do the number of articles that discuss Artificial Intelligence. Some contain hyperbole, many are guilty of some form of exaggerated speculation and a few conjure up dystopian images of a near-future in which humans are subjugated by machines.
But what is really happening? What if we strip away the hype of Artificial Intelligence and instead focus on the very real and very exciting science of Machine Learning? How is this practical and well established field of study being applied to solve significant problems? What is the potential of Machine Learning when applied to the challenges of People Analytics?
First of all, let’s start with a definition. Larry Wasserman, a statistician, succinctly stated that ‘Machine Learning is the science of learning from data’. Put differently, machine learning as a field is the scientific study of ways to make machines learn as they accumulate data.
But what does learning mean in this context? In the abstract, something can be said to learn if it improves its performance at some task as it accumulates experience. For example, when someone learns to play a musical instrument, we recognise that they are learning since the more they practise (experience), the nicer it sounds (performance). Analogously, Internet search engines get better at completing search queries (performance) as they are exposed to more and more people typing in particular terms (experience) – in an abstract sense they are learning to complete search queries.
But machine learning can also be a lot more familiar than some might expect. To complete the quotation of Wasserman “Statistics is the science of learning from data. Machine Learning (ML) is the science of learning from data. These fields are identical in intent although they differ in their history, conventions, emphasis and culture.”
To see how statistics can be thought of as learning, consider drawing a trend line through a scatter plot. If one were to then use this trend line to make predictions, one would hope that the accuracy of the predictions (performance) would be better than if one had not used the data (experience) to calculate the trend line. This doesn’t really feel like learning because the calculation of the trend line happens (near) instantaneously, but that is essentially the only real difference, the amount of time, computation or data taken to learn something.
With this in mind, if you have an understanding of what statistics is and when it is useful, you already understand what machine learning is all about – it just takes more time/data/computation (usually). If not, think of it as “machines using data to be good at doing things” and you’ll be fine.
Back in 1965, Moore put forward his now famous observation that computing power (or more precisely the number of transistors in integrated circuits) doubles every two years. His initial estimate was that this ‘law’ would continue for another decade. More than fifty years later, computing power continues to grow exponentially, becoming ever cheaper and more available.
Similarly, since the inception of the internet, the amount of data that is both collected and made accessible has rapidly grown. Estimates vary about the current rate of data creation but the popular sound bite that ‘90% of the world’s data has been created in the last two years’ is close to the truth.
So, if we think of Machine Learning as statistics that takes more data and or computation. And we know that both data and computing power are growing, exponentially. It is therefore an excellent environment for Machine Learning methods to become increasingly popular in a time of ‘big data’.
It isn’t all about big data though. The increase in computing power has also made it increasingly possible to automate aspects of data analysis empowering people to analyse their data more thoroughly. In other words, Machine Learning can make advanced and complex statistical techniques more widely available and accessible. We view this as being easily as important for people analytics.
You may not realise it, but you encounter the results of machine learning on a regular basis whenever you go online. One of the most ubiquitous tasks that machine learning systems have been turned to is the problem of making recommendations.
For example, when an advert appears on the side of a webpage, it is not chosen at random. It is a recommendation based on your browsing history or choices you made online. Machine learning has helped suggest the next movie you might enjoy, items you might want to add to your shopping basket, articles or books you might want to read and what web pages might answer your search.
But machine learning isn’t just used to make recommendations. Its applications are growing in number and becoming increasingly sophisticated. Examples of machine learning being used in the real world, include:
A recent high profile example of machine learning is Google Deep Mind’s AlphaGo, a computer that learnt to play the ancient Chinese game of Go. In March 2016, AlphaGo beat a human professional player called Lee Sedol, over a five game series. This was significant, for two reasons.
Firstly, Go is a game with trillions of possible moves, making it hard for players to follow simple strategies and therefore challenging for a computer to master. Humans are able to deal with this complexity by employing their intuitive sense of the game, built up from years of play.
Secondly, AlphaGo beat Lee Sedol, the current world champion and a target who contributors to AlphaGo’s programming had initially thought was beyond them for at least a decade. That said, AlphaGo was learning quickly, playing itself over 1m times every day to improve its decision making.
This example, together with the others listed above, show that the opportunities to apply Machine Learning are growing, as are the sophistication and effectiveness of these techniques. Wherever there is data, there is a chance that machine learning might be usefully applied.
Qlearsite also uses Machine Learning. We apply these same techniques to grapple with the ‘big data’ generated by organisations. As workplaces, HR processes and even employee interactions (e.g., feedback, appraisals, etc.) are increasingly digitally recorded the volume and complexity of data is growing. Within that complex data, lies hidden opportunities to optimise organisations.
Let’s consider the data created every day in an organisation. Training events are logged, assessments are recorded, absences are noted and performance reviews are stored for later referral. In every modern business, many systems are carefully cataloguing daily information about employees. This complex but interconnected data can be used to help answer pressing business questions.
Why do our best people leave? Commonly the answer to this question is compiled using anecdotal evidence and hypotheses. Machine learning techniques can use hard data and facts to provide a fuller, more accurate explanation in two important ways.
Firstly, we can process the ‘big data’ generated by organisations to find the shared characteristics of leavers. By rapidly sifting through the variables to look for statistically significant patterns, we can identify who is most likely to leave in the future and provide supporting evidence of our predictions.
Secondly, we can access better and more unusual sources of people data that are particularly revealing. Employee engagement surveys, exit interviews and other descriptive free text data is recorded in systems but very hard to incorporate in traditional statistical analyses. Natural language processing techniques, enabled by machine learning, allows us to read this text giving us insight into people’s stated motivations, aspirations, doubts and concerns. This soft data, empirically measured, can provide deeply revealing insights about employee behaviours, including the real reasons why some people choose to leave.
Explaining who might leave and why, is just one useful application of Machine Learning for organisation data. The same techniques can be used to analyse and optimise productivity, absence, engagement and all the other people metrics that organisations are seeking to improve upon.
We started Qlearsite because of a simple principle and a simple observation, that all decisions should be fact based and that all too often employee decisions are not. We want to help every organisation use scientific discipline and statistics to analyse big, complex people data and make better decisions.
Our work is to firstly collect and connect data from every type of system in an organisation, both quantitative and qualitative. We then use advanced statistical techniques to help us understand the hidden truths buried within that rich, complex big data. Finally, we want everyone to be able to explore data, setup analyses and understand conclusions. This is a significant challenge.
To overcome this challenge, we use Machine Learning techniques. We know that learning from data, in an automated way, is how we will achieve our ambitions. That is why we place the ‘science of learning from data’ at the centre of our work to understand organisations and their people.