What’s all the fuss about AI?
By now, you have no-doubt been bombarded with articles talking about artificial intelligence (AI), machine learning, data science or any number of other terms and buzzwords. To put it mildly, there is a lot of talk about the field and how it will change (and has already changed) our lives and the way business is done. But is the hype justified? I use Siri and Alexa, but is there more? Is this all possibly dangerous or unethical? To answer these questions, it is worth talking about what people mean when they refer to these terms.
When people work with data, generally they are trying to detect, and act upon, meaningful patterns. Sometimes this is done by humans themselves, when they have an idea what patterns might exist in the data. For example, a company might believe that their sales vary by age and gender, and so they might prepare a summary of their customers split in this way. However, sometimes humans might not have a specific idea about what useful patterns exist in the data, or may believe that there are valuable patterns in the data beyond what they can see. This is where computer intelligence starts to make its mark.
Although there are many definitions for AI and machine learning, at their heart they all seek to have computers recognise meaningful and valuable patterns in databases with as little intervention as possible. Even though humans are great pattern spotters, there are a few reasons using a machine to recognise patterns is useful:
Humans are good at certain types of pattern recognition, such as recognising people, items, sounds and smells. We may also develop good pattern recognition in our professions. For example, an experienced doctor might be particularly skillful at diagnosing medical conditions. However, there are many types of pattern recognition for which humans are especially bad. One type is pulling useful patterns out of modern databases, made up of millions of lines and thousands of columns, spread over hundreds of tables. This is what much of the data collected by organisations looks like, and so machines are a wonderful asset to harness.
Machines are much better than humans at recognising very complex patterns – whereas human theories typically involve a small handful of dimensions (such as segmenting sales by age, gender and geographical area), machines are easily able to recognise patterns involving hundreds or thousands of variables altogether. In addition to recognising patterns involving so many variables, certain machine-learning algorithms allow for the interaction of all the variables to be considered too.
Machines do not suffer from many of the prejudices and cognitive shortcomings of humans – of which there are known to be a troublingly-large number. We are particularly biased towards things that we think we “know”, even though such knowledge might not really be well-founded or supported by real data. Machine-learning algorithms can be set up to avoid many of these pitfalls, producing results that often exceed the best performance of humans.
Machine-learning algorithms are less sensitive to bad data – unlike summary statistics and business-intelligence dashboards, which rely on the completeness and accuracy of data for their outputs to be valuable, machine-learning algorithms are not attempting to re-present the data in a more usable form. They are merely attempting to identify valuable patterns in the data. As such, they are also able to learn useful patterns from “bad” data, provided the data remains consistently bad. This is particularly useful since much data, even in commercial settings, is far from pristine.
Machine learning can be used to generate specialist expertise in a business extremely rapidly. A few of the most useful applications of machine learning at the moment in commercial settings include:
Sales and marketing – instead of spending fortunes on “spray-and-pray” sales and marketing initiatives, using machine learning to identify prospects more accurately, allowing a more targeted and effective acquisition of new customers.
Customer retention – many companies have ineffective customer-retention activities because their approach is reactive; at the point that customer chooses to cancel his or her relationship, someone tries to get them to stay. However, by then it is usually too late to retain the customer relationship. Machine learning can be used to identify which customers are likely to terminate their relationship with the organisation, often several months ahead of the actual decision. In this case, by acting pre-emptively, far better results are achievable.
Risk assessment – machine-learning algorithms have proven themselves especially adept at being able to assess a variety of risks on a granular basis. Such risk assessment could include medical risks, insurance risk or credit or other financial risks. In many cases, machine-learning methods have outperformed established industry practice in these areas.
There are many other applications of machine learning, some of which are already being used extensively and some which are still being adopted. One thing is clear though; we have barely scratched the surface of the machine learning applications that will radically change the landscape of business and social interaction.