Artificial Intelligence, Machine Learning, and Deep Learning are just some of the technologies that today are driving the expansion of what Klaus Schwab has dubbed in a recent book as the “Fourth Industrial Revolution” (see previous blog for a review of his book). This lexicon of terms populates the scientific papers people read in an effort to educate themselves about this galaxy of sophisticated concepts. The technology is sophisticated and understanding it can be daunting, but it is possible to become enlightened without burning up too much grey matter.
Firstly, let me explain a few key terms to help you differentiate between Artificial Intelligence, Machine Learning and Deep Learning. This graphic shows that they are inextricably linked, and are often referred to as cousins of each other:
What I would like to try to do in this blog is explain some of these terms, and in particular to try to demystify Machine Learning. In a few words, Machine Learning is the capability of a machine to learn from data patterns to estimate a predictive model of behaviour or wants. In other blogs I will address some of the other concepts and hopefully demystify those.
Let me begin with a few words of solace and encouragement; stripped down to the basics, the fundamentals of this technology can be understood by the average layperson. That’s not to say you can become an overnight expert in any of these fields, but you will be able to understand at least their design and purpose. And as I say in the heading on this piece, Machine Learning is just like pouring a pint of Guinness. Stick with me while I explain what I mean. Hopefully it will make sense.
In the 1980s, while at college, I worked as a barman. I was a pretty decent barman too. I liked the job and I enjoyed keeping my customers happy. Little did I realise that during this time in my efforts to keep the “punters” happy I was applying the basics of Machine Learning. As a term it was coined in 1959 by Arthur Samuel at IBM. Trust me, I’d never heard of it in the 1980s, but I was applying some of the concepts to my work with me, in this case, being the machine.
Every Sunday, without fail, one of my customers would come in after church for a few drinks before going home for Sunday lunch. He arrived same time. Took the same seat. Ordered the same drink. It never wavered. One Sunday I saw him through the window heading towards the bar. I anticipated his order and pulled a pint of Guinness. He had no sooner sat on the same stool as always when I slid the fresh creamy pint in front of him. He looked and smiled a big appreciative grin.
It became the norm from then on that every Sunday I would look out the window and on seeing him approach, put his pint on. That went on for years until I left the job. Throughout that time he was very happy customer. No queueing for him. A pint ready-made and set up. It was for him the acme of customer service. As a customer experience, it could not be bettered (other than getting it for free!).
But wait says you, what to heck has this got to do with Machine Learning? Good question: well, it has everything to do with it, I would argue. In doing what I did every Sunday for that customer I was applying the methodology of Machine Learning. To explain further: I was identifying someone; I was discerning their pattern of behaviour and calculating the probabilities of their taking a certain action; I was learning from that process, and based on all of this data I was processing and predicting what I should do to keep that customer happy. In short, I was building a little algorithm in my brain which kicked in to action every time I saw that gentleman approach the bar on a Sunday.
This fact was underlined to me when one Tuesday afternoon the same punter came in to the bar. My reaction was to stand there waiting for him to order his drink. A Tuesday visit was an anomaly. I had no experience of him visiting on a Tuesday. I could have guessed he wanted a pint of Guinness and set it up. But it would have been just that, a guess based on zero data with no pattern of behaviour to guide me. As it turned out, he ordered a coffee as he was taking a break from work. Of course, had he come in every Tuesday over a reasonable period I could have worked out a pattern of behaviour with certainty, and thus made sure to have his hot coffee waiting.
Just as in Machine Learning, I had a well-formed algorithm that allowed me to act with certainty on a Sunday, but the same algorithm could not be applied on a Tuesday. That is the core of Machine Learning. Machines can be trained to gather and assess data, allowing them to recognise patterns and predict peoples’ behaviours or wants. How do you think Netflix knows what films and box sets you like to watch? Or travel companies know which advertisements for sunny climes they should present to you in the middle of winter? Machines develop algorithms that can be fined-tuned to guess what you want to watch, where you want to go, what you are likely to buy, or what your tastes in music might be.
The foundation of all of this “magic” is the accumulation of personal data that can be sliced and diced, analysed and parsed and from it the machine can be taught (learns) how to keep you happy (Netflix), or how to get you to part with your money (the unplanned winter holiday). Of course, Machine Learning is everywhere now. It is even driving cars, and is a growing part of our world. Why has something that has its origins in the 1950s, and was available and used in a limited way during the 1980s and 1990s become so prevalent and growing?
Why? Big data. Within the last 10 years it has become possible for companies to harvest huge swathes of data, from multiple and varied online sources. The additional ascent of powerful computers with extraordinary computational power has allowed corporations to analyse and exploit this treasure trove of information. Whole industries have been built around the ability to gather, analyse and exploit big data. As a means for building algorithms, and propelling company sales, it is a powerful tool. It is unlikely to go away. One thing that can be said for certain, increasingly ways will be found to expand the use of Machine Learning over the coming years. As a technology, like a pint of Guinness, it is here to stay.
Aidan Collins is a language industry veteran. He works in the marketing department at KantanMT.