AI and Gambling

October 24, 2019 6:45 AM
  • Andrew Tottenham — Managing Director, Tottenham & Co
October 24, 2019 6:45 AM
  • Andrew Tottenham — Managing Director, Tottenham & Co

intelligence (n.):

Story continues below

    the ability to acquire and apply knowledge and skills

artificial intelligence (n.):

    the theory and development of computer systems able to perform tasks that normally require human intelligence, such as visual perception, speech recognition, decision-making, and translation between languages

Much has been made of developments in Artificial Intelligence (AI) and how it can be used to solve real world problems. Within the world of gambling there are several general areas where AI could be useful, but before going further it is helpful to understand what AI is and how it works.

AI is, like any other computer program, a set of instructions, an algorithm. The difference from non-AI programs is that AI analyses extremely large data sets to look for patterns in or connections between variables in those sets. Once it identifies what appears to be a pattern it makes an assumption, for example that W, X, Y and Z are connected in some way, and assesses the validity of these assumptions through repetition, continually refining the assumptions, trying to get the best fit, and “learning” as it goes. (This is a simplistic description, but you get the drift.) Once it has completed its work it is able to make a prediction with a probability attached, such as that there is a 68% probability that Chelsea will beat Ajax in the Champions League game. After the match, the data from that game and others can be analysed and further predictions made.

Humans are able to see patterns in visual form. For example, our brains have evolved to recognise individual faces in a large crowd, even when moving. For a computer this is a surprisingly difficult task. However, humans are not able to comb through millions of data points to hunt for patterns, as AI can.

Gambling, whether online or land-based, produces vast amounts of data, both about player behaviour and game performance. For land-based table games, it can be a struggle to capture accurate game and player data, but this is improving rapidly. In this data-rich environment, AI can be used to optimise game design to better appeal to customers; create marketing campaigns that are more likely to appeal to a target group; customise the user interface so that customers only see games that are likely to appeal to them; and optimise revenue streams from slots or casino games. A casino floor can be treated holistically, predicting how changing a few machines might impact player behaviour across the entire casino floor.

One discussion I had whilst at G2E revolved around whether integrated resorts are too focussed on maximising casino revenue at the expense of the profitability of the entire integrated resort. Could AI be used to maximise the total profitability of all of the revenue streams? The AI specialists in the rooms thought this was an exciting question.

At G2E I was on a panel looking at technology in gambling and how it can be used to minimise the risk of harm to customers. Manu Gambhir, CEO of 24/7, the largest India-facing online rummy site, has been trying to use AI to identify players who are or likely to become problem gamblers. One of his challenges is that whilst he has a great deal of player data, he doesn’t have a set that is solely problem gamblers or those that are likely to become problem gamblers. And so the AI software isn’t able to identify anyone with any degree of accuracy, because it has nothing to learn from.

To overcome this challenge, Gambhir interviewed psychiatrists, asking them what traits they thought might indicate problematic behaviour. Using that information, the AI program was able to predict who might be a problem gambler. Interviews with a sample set of these “problem gamblers” ensued; some did say that they had a gambling problem. Armed with this information, the AI program is now able to predict with about 60% certainty who will become a problem gambler. As more data is received, probably from further interviews, Manu hopes his prediction rate will improve to over 90%.

We know that betting on the outcome of sporting and racing events has been going for over 2,000 years, because we have records of bets placed in Ancient Greece. The Romans codified the practice, even allowing betting on gladiatorial combat. Today betting is nearly ubiquitous, and events now generate reams of data, not just who scored individual goals and which team won, that is pawed over by “professional” bettors. AI has made big strides in this area, to the point where machine learning AI programs find that successfully predicting the outcome of a sports event is fairly simple.

The US-based company Unanimous AI uses “swarm intelligence”, a combination of groups (swarms) of people and AI, to make more accurate forecasts and predictions to help them make better decisions. Using this approach, Unanimous AI placed an online ad for volunteers to form a swarm and twenty minutes after starting was able to accurately predict a “superfecta” at the 2016 Kentucky Derby – which horses would come first in four races. The bookmakers had offered 540 to 1 on that particular bet. And to prove it was not a fluke, Unanimous AI predicted not only the winner of the 2017 Super Bowl but also the final score, 34-28.

Unanimous AI is not the only company in this field. Some use armies of “game analysers” who track matches and enter details of what is happening on the field. This is then overlaid with betting prices to determine a betting strategy. It will not be long before game analysers become redundant, with advances in visual AI allowing games to be automatically analysed and processed, creating better, faster and more accurate sources of data, and ultimately more accurate predictions.

You might think that this would lead to the end of bookmaking, but remember that if customers consistently win over a period of time, bookmakers will either limit the amount they can bet or close the account. This type of AI thus gives a very short-lived advantage.

AI is ideal for developing winning game-playing strategies, especially games where skill is a factor in determining the outcome. In 1997, Deep Blue, IBM’s chess-playing AI engine, became the first computer to beat a chess Grandmaster, in this case Gary Kasparov. IBM’s question and answering AI, Watson, beat Jeopardy champions to win $1 million in the final in 2011. It wasn’t long before Google entered the game; its Go-playing AI, AlphaGo, has chalked up impressive wins since it was introduced in 2016. In 2017 it won 60 games and lost none against some of the world’s best Go players.

Libratus, Carnegie Mellon’s entry into the AI game-playing field, won $1.76 million from professional poker players Jason Les, Dong Kyu Kim, Daniel McAulay and Jimmy Chou in 2018. Over twenty days they played almost 120,000 hands of no-limit Texas Hold ’em; Libratus not only learnt how to play winning hands successfully but also to bluff effectively – and opportunely – when it had a poor hand.

It is clear that machines are becoming better and much faster than humans at tasks that require large amounts of background knowledge and skill, and at tasks where large amounts of data have to be analysed. And although we call it artificial intelligence it isn’t really intelligence, it is just an algorithmic process. The machine “knows” nothing. The AI is only as good as the quality of the algorithm and the data it is processing.

 

So what is intelligence? I will leave you with a quote from R. L. Gregory, British psychologist and Professor of Neuropsychology.

Innumerable tests are available for measuring intelligence, yet no one is quite certain of what intelligence is, or even just what it is that the available tests are measuring.”