How Machines Learn

Head of Analysis

Michael Baines

Head of Analysis

Posted: 20 July 2017

If you work in marketing, I can guarantee you’ve heard the phrases ‘Artificial Intelligence’ or ‘Machine Learning’ at some point over the last few months.

The recent hype around data science, ‘Big Data’ and cloud computing has sparked a huge resurgence of interest in this field, but Machine Learning as a discipline is nothing new. It has decades of commercial applications but has been lost in all the noise that the introduction of data science has made.

Making the distinction

To some, even the mere mention of the letters ‘AI’ conjures fear and uncertainty. Just what exactly would it mean for the workforce as we know it today if machines start to ‘take over’! For others, like myself, it feels like we are at a pivotal and exciting point in history. With computers now being able to operate at scale and process vast amounts of data that many organisations now generate, making use of Machine Learning has never been more important.

“Quite often, the terms are used interchangeably and incorrectly.”

The first thing I want to do is uncouple machine learning from AI as, quite often, the terms are used interchangeably and incorrectly. So, what’s the difference?

Artificial intelligence is an area of research that attempts to recreate cognitive functions possessed by humans. The end goal of AI was (and to some extent still is) the creation of machine consciousness, i.e. the machine ‘thinking’ and making rational decisions.

Machine Learning:
Very much a subset of AI. In contrast with AI’s stated goal of overall cognitive intelligence, machine learning algorithms seek to recreate answers to specific tasks by learning from their environment. Really nothing too flashy, just beefed up statistics.

How machines learn

So, how do machines learn? It’s surprisingly simple. There are 3 main types of learning patterns a machine can adopt:

Learning patterns 1&2 have been used in marketing for decades. CHAID, K-Means, Discriminant Analysis, Logistic Regression… all terms we, as marketers, should be familiar with, and all examples of machines learning from the inputs we feed them.

Reinforecment learning has gained prominence over the last few years with the emergence of self-driving cars. In fact, it’s even now being used in medical science to determine the best combination of drugs to administer to critically ill patients.

Why you need to know

Operational marketing processes today generally involve a series of rules which, when triggered, initiate interactions with the customer. But who decides these rules? Generally, a marketing exec or account manager who is essentially making educated guesses on what will be most effective in a given situation. Machine Learning, however, can run through billions of data points and establish the optimal outcome in seconds, freeing up huge amounts of operational resource and generating substantial returns on investment.

Domain knowledge and expertise can only take us so far in the decision-making process, and there comes a point where machines can begin optimising our marketing decisions far better than we can.

Michael Baines
Head of Analytics

For more information on how Fuel can help introduce machine learning into your business get in touch.