Skip to main content

Having been to several conferences, both in person and virtually, in the last couple of years and attended sessions which have included AI in the title, we have been disappointed by what we have seen and heard. All too often the use of the term has been misleading and suggesting an attempt to grab attention. We have started to wonder at what points management information (MI) or business intelligence (BI) becomes predictive modelling (PM) and then becomes machine learning (ML) and, to be honest, we have been struggling to make sense of it all. To be blunt, there seems to be an awful lot of buzz but also a lot of bull.

So, at the risk of adding to the hype we thought would share some of thoughts via a series of short articles. In this initial one we look at some of the basics.

At the most basic level AI is, of course, artificial intelligence. A simple definition is the replication of human behaviour by machines to perform tasks. Actually, AI has been around since the 1950s.

In machine learning, analytic techniques are applied to develop a model algorithm. This algorithm can then help make better and more efficient decisions.

In many cases, what is referred to as AI is actually ML. Computers need to be programmed to achieve this and do not yet ‘learn’ by themselves in the true sense of the phrase.

‘Deep learning’ is a technique within machine learning and often uses multiple neural network layers which are applied in complex fields such as speech recognition.

The buzz about machine learning has intensified in the last ten years, fuelled by the availability of ever-increasing amounts of data, both structured and unstructured. In the world of life and disability underwriting this includes wearables, credit scores, laboratory information, claims data, pharmacy records and electronic health records. Plus there is a huge amount of unstructured data such as facial and speech analytics which may have potential to improve risk prediction, for example the creation of smoking propensity models. Everyone is becoming more and more interested in acquiring data. A so-called data ‘arms race’ is well under way, one manifestation being Google’s acquisition of Fitbit.

The growth in the use of AI has been fuelled not only by the availability of data but also by increased computing power and improvements in the understanding of algorithms.

So, this explosion in the availability of data creates the potential for innovative rating methods. Of course, there needs to be care in how this data is used to ensure that these models do not stray over (or perhaps even get close to) the line in terms of privacy and discrimination. Significant benefits also come with the potential for unlocking data that insurers currently have in their existing documents, as much of this data is unstructured.

So where does MI or BI stop, turn into PM and then become ML (creating abbreviation overload)? We have heard many presentations that confuse these three, and of course there is a certain degree of overlap. Let us demystify.

Both machine learning and predictive analytics are used to make predictions from a set of data about what might happen in the future. Predictive analytics is driven by predictive modelling which often includes a machine learning algorithm. Predictive modelling has its roots in mathematics whereas ML has its roots in computer science. In predictive modelling, algorithms are developed and validated, whereas in ML algorithms and trained and tested.

There are two types of AI, namely narrow and general. In narrow AI the machine completes a task or series of tasks well. Narrow AI powers predictive recommendations such as ‘Customers like you bought…’. It includes optical character recognition (OCR), natural language processing (NLP) and chatbots – ‘robot’ programmes that enable a conversation; more on these in a later article. The common perception of AI, which is where a machine exhibits human intelligence, remains the stuff of science fiction.

Business intelligence is defined as the visualisation of data and insights to aid business decision-making. So BI is a window into the black box. BI and dashboards have been around for a while so BI might be the way in which the data from an OCR or NLP process are visualised.

In this article we have covered some of the basic terminology. Our series of articles will look at some applications of the emerging technology. In the next one we will look at OCR,  NLP  and text mining, and how these techniques could be applied in the world of life underwriting.

Will AI entirely replace the human underwriter? We don’t think, so especially in the short or medium term. But AI can save the human underwriter time by giving them a better look at the big picture.