top of page

Responsible gambling and AI – a quick-fix solution?

Updated: Feb 4, 2021

By: Kim Mouridsen, Professor at Aarhus University and founder of Mindway AI

Artificial Intelligence (AI) has been a major buzzword for some years now. An ever increasing number of applications that make use of AI see the light: from targeted advertising to precision medicine. This also happens to be the case in the responsible gambling area where AI is touted as the next miracle solution. However, AI is not a cure-all, unless it is applied intelligently and is supported by the knowledge and experience of domain experts.

Unsupervised learning

The most basic application of AI is to simply search for patterns; and while patterns will certainly be identified, it might not be the most useful or predictive patterns that will be found. AI in the shape of unsupervised learning or outlier detection has little to offer.

Unsupervised learning and outlier detection merely identify differences between entries in a data set, as the AI’s algorithms receive no directed input (or “training”) in order to detect those patterns that are actually prognostic. What you get is a fancy way of organizing data and that’s it.

Targeted learning

Ultimately, the value of applying an AI solution depends on having an appropriate target. Without a target or goal, the outcome is of no great value. An example from a different realm is the prediction of stock market prices by applying AI. The stock market price is the target that we aim to predict.

Conversely, if we do not have a target, there is no value in applying the AI’s algorithm as we still will not know where we are headed.

Finding the right target

In the responsible gambling field, it has become popular to use self-exclusion as the target for AI predictions. However, as stated in the PWC Remote Gambling Research report, 80% of those who perceive themselves as problem gamblers have never used a self-exclusion tool. Furthermore, only 31% of gamblers who have self-excluded in the past define themselves as problem gamblers.

Thus, self-exclusion is not a reliable marker as there is no direct correlation between self-exclusion and problem gambling. One could be tempted to think that relying on the self-exclusion target is churn prediction in disguise rather than AI finding actual problem gamblers.

Nevertheless, AI can be successful in identifying (future) problem gamblers. What it takes is expert knowledge on what characterizes problem gamblers, using experience, responsible gambling insights, and neuroscience to define suitable AI targets.

In short, when combined with state-of-the-art AI, algorithms can be trained to accurately identify problem gamblers. However, there is no such thing as a quick-fix. The algorithm is only as intelligent as the brains behind it.

bottom of page