AI – Hype, Reality or the Future?
Artificial Intelligence has been a buzz word for many years, with a mixture of reviews and opinions, posing the question: is it hype, reality or a serious opportunity for the future?
The basic concept of AI is that a computer programme or machine can become capable of thinking and learning as a human does; the term was first coined by John McCarthy in 1956.
A multitude of possibilities
Last month we attended the AI Tech World event in London to hear some interesting examples of how AI is used today, alongside its limitations. We heard talks from digital leaders across a range of industries including health, hospitality and transport, as well as the public sector.
In this article, we’re going to focus on one sector and explore an exciting example of how AI is currently being used in the online gambling industry.
In recent years the issue of gambling addiction has become increasingly important and prevalent, with Gamble Aware regulations and controversy within the press as online betting increases in popularity. The nature of online betting makes it difficult to spot when an individual is betting more than they can afford or if an individual has opted to self-exclude themselves from betting following an addiction.
In an attempt to combat these issues, some online betting businesses are exploring the use AI to flag at risk users via indicating duplicated accounts (an individual who has previously self-excluded themselves following an addiction) or users who are at risk of betting more than they can afford.
To identify duplicated accounts, an algorithm can be set to look at a variety of data points of new accounts and runs them through a log of historic data to establish any potential correlations. ‘Metaphor matches’ can be used here to identify when a profile is a similar but not exact match i.e. Joe Blogs and Joseph Blogs if other factors (like address) are the same across accounts.
Whether an individual is at risk of gambling addiction or not is heavily subjective, with a vast amount of variability between individuals. This can make it difficult for both humans and computers alike to fully distinguish. Gambling websites can use AI to look at a variety of decisions taken by a customer to identify if their behaviour indicates a potential risk – these could be factors such as using a credit card (opposed to their usual debit card), increased/abnormal activity or gambling large amounts late at night.
If any of the scenarios above indicate a potential risk, then a team of experts can evaluate the data and take appropriate action, allowing for emotion and a “human touch” to be added into the equation. In this case, the process of AI builds the foundation (sorting through a vast amount of data to bring attention to potential risks) at a faster and more accurate rate than manual monitoring, while leaving the final decision in human hands. The outcomes of the results are then fed back into the algorithm so that it can continue learning by identifying correlating patterns of online behaviour.
In spite of the multitude of exciting possibilities of AI, this is a technology that needs to be approached with care and caution. As with all data analytics, the potential of AI is limited to the quantity and quality of the data going into the model. Before AI can be implemented a reliable data source is essential and if this is not currently in place, this should be a business’s primary current focus and not AI - as anything outputted would be equally as unreliable. Furthermore, the introduction of GDPR regulations increases the difficulties of data handling and storage which in turn can increase the complexity of data collection and can reduce the amount of information available.
A further issue with AI is subjectivity and emotions. Often the story behind the data isn’t black and white, it may require interpretation which surpasses the current capabilities of machines. This was recently highlighted in Google’s ‘Bird and Bike’ challenge. This was designed for developers to compete to write an algorithm which can distinguish between photos of birds and bikes as there is no current algorithm which can do this with 100% accuracy. Currently the competition is still on-going – we’re keeping a close eye on the project to see if anyone is successful!
Like all statistical analysis, the way in which an algorithm is set up can pave the way for the eventual outcome (bias). This limitation was seen in Haar-Cascade where a facial recognition algorithm saw a lower detection rate on dark-skinned individuals as a double T pattern (linking the eyebrows, nose and mouth) in greyscale images were more difficult to detect. Tools such as Google’s ‘What-If’ (a tool which lets users analyse machine learning models without having to write any code) can come into play in attempts to flag bias by visually displaying a variety of different possibilities.
Should you gamble on AI?
To conclude, AI is not yet completely ‘intelligent’ and is still subject to issues such as bias, reliability of data inputs and subjectivity. Despite the issues with AI, with human supervision AI can significantly aid and improve some of the grounding functionalities of day to day operations. While the future of AI is uncertain, focusing on quality data inputs today will pave the way for more intelligent AI models in the future.