Betting and artificial intelligence – now and then
Isn’t artificial intelligence something very new and modern, I hear many of my readers asking, when they see the title of this My Way -blog.
Well, it is obvious, that Sports Betting gains a lot of the almost unlimited potential AI offers. But already for years, some models, which are in a sense "pre-AI"-programs, have influenced sports betting.
Actually, I have never bet on teams I "like". From the first bets I made as a teenager, I sought value based on my probability estimates - which I counted in a brutal and basic "model". The model itself and the input values were updated in a way, which has a lot of similarities with some of the modern AI algorithms. But I do not dare to count, how many hours of my best young days I spent calculating my probabilities. The current AI would have done my weekly workload in seconds.
At the beginning of this millennium, my company European Sports Betting Consultants created a neural network called Neurodesigner Sports Analyzer Designer with a Tampere-based company Cybersoft. Neurodesigner was even sold to the general public. It gave decent results, but I never liked the idea of winning without knowing why. It was psychologically demanding - especially in the losing runs - to place the neuro network bets without knowing why?
There's an old proverb that says fire is a good servant but a bad master. In a way, AI is in the same situation.
Just because AI offers great working tools and help, it is no use, if the ideas behind the algorithms are not sound. One has no use for a map and a compass, if one doesn't know, where one is trying to reach. When so much data and ways to use it are available, it is the task to find the essentials. When AI does not base its starting knowledge on logical and essential facts, no profits are to be expected.
Some attempts at futures trading AIs have ended in catastrophes because the algorithms were trained with wrong ideas and invalid data. It makes no sense to use too old or too small datasets as an input for artificial intelligence. Even the underlying processes are in a constant state of change. Risk management of the AI operations should therefore be very strict even with the "idea risk" and "analysis risk". Some of those financial AI operations didn't understand or manage those risks, and they went bust.
Even advanced programs need the support of the human mind.
The best AI programs used in poker and chess are created by giving the basic strategies of the game to the AI from the good human players before the AI was programmed to start playing against itself and improving its play towards perfection. The human mind understands abstract, deep things better than any mathematical model. Artificial intelligence is evolving.
Even in Sports Betting, AI is a welcomed extra assistant for a good analyst. Already the linear regression and probit models have been given decent results, because in Sports Betting only a handful of key factors are meaningful to the probabilities, but with modern AI, one can, with a realistic workload, fine-tune the analysis process, monitor it and teach the models to improve.
European Sports Betting Consultants have for years utilized an AI-powered model ESBC PowerAI for estimating the probabilities in the betting markets. However, we do not use the AI-generated estimates purely, "as is". Instead, every odds estimate made by artificial intelligence is reviewed by a human expert. Our senior odds compilers Hannu and Karri have worked for ESBC for over twenty years, and My Way as a winning sports bettor started in 1979. There are some unique situations and finesses, that the algorithms simply cannot be taught. Therefore, even in Sports Betting, as in numerous other paths of life, AI and human "silent knowledge" and "intuition" are stronger together than individually alone. Even in the future, we at ESBC will not give complete control over artificial intelligence. I repeat: it is dangerous to win without knowing how.
In horse racing, AI is even more valuable than in sports betting. In racing, there are much more factors affecting the probabilities, and their weights vary a lot from case to case. The correlations between the factors can be very difficult to define and calculate, so that creates another challenge for the traditional models in race betting – even if some quite simple models have beaten Hong Kong Horse racing for millions.
Perhaps the biggest use for AI in betting is as a tool for creating and placing bets when the probabilities have been calculated. In creating positions in difficult combination pools like pick 15 pools betting or Pick6 racing jackpots, artificial intelligence can be very helpful.
We use an algorithm called ESBC Vakio Wizard in the Finnish P13 1-x-2 jackpot.
It uses a genetic algorithm, which ”moves” in the space of millions of possible combinations towards more profitable but not too impossible combinations. When it finds a good neighborhood line with good profitability and a sufficient probability to hit, it stops and creates a combination bet of a couple of lines. Vakio Wizard hit a win of €301600 on Boxing Day 2017, and because the bets it makes are only a couple of euros, the smaller wins keep it profitable for years - before the next big jackpot.
Even forecasting the odds movements accurately can be as profitable as determining the ”correct” probabilities. In the markets forecasting Artificial Intelligence can be very useful. We work already with an AI idea following the odds movements in the Sportmarket. The data from Sportmarket is very useful because it is based on the odds of the World's biggest bookmakers, the odds, which have actually been matched by big bettors and layers.