Will Robots destroy the Armchair Punter?

When the computer programme, Deep Blue, beat world chess champion Garry Kasparov, in New York City in 1997, it seemed symbolically significant: artificial intelligence was now in the ascendancy over its human counterparts.


Machines overpowering humans

Deep Blue ran on an IBM supercomputer that was able to calculate 200 million possible moves per second, a brute-force approach that became the hallmark of computer chess programmes going forward. Humans might be able to be selective in their analysis, using experience to identify the most likely sources of success when choosing possible moves.

However, when faced with a machine that can analyse so much, and so quickly, the intuition of humans is soon overpowered. Today, the best brute-force computer programmes, such as the unofficial world-computer-chess-champion Stockfish, crush any human player. Even when being run off a cheap laptop with limited processing power.

It was headline news, then (in the chess world, at least), when a paper released in 2017 described how Google Deepmind’s Alphazero programme defeated Stockfish convincingly, despite not using this brute-force approach. Instead, AlphaZero utilised a machine-learning protocol. The principle of this method – although undoubtedly more complicated than this explanation allows – involves the programme, which is only given the basic rules of the game as a starting point. Playing chess against itself multiple times to discover the optimum way of competing.

After just nine hours of this self-learning – the paper claims – AlphaZero defeated Stockfish convincingly in a 100-game match.

Analysing the games that have been released as part of that paper has revealed a (machine) player who is highly aggressive and novel in its approach. This has allowed humans an insight into where they can improve their own games. There are some questions around the credibility of these results, as you might expect, particularly around the hardware and game conditions that Stockfish was expected to operate under. Nonetheless, it seems that a machine-learning protocol may now be the way forward to improve the playing strength of computer chess programmes.


Are you ready to take advantage of the BSP and maximise your profits on the Betfair Exchange?


AI outperforming dermatologists

The revelations of what Artificial Intelligence (AI) can achieve, though, are not exclusive to the world of chess. Each day, headlines report how AI is outperforming humans in all manner of activities. In a recent story, for example, a deep learning neural network was reported to be able to comfortably outperform dermatologists when identifying skin cancers from scan images.

The march of artificial intelligence – as with most technological advancements – raises worry and excitement in equal measure. Some commentators fear widespread job loss through automation and even suggest that machines may challenge our dominance as a species. Others, though, look forward to seeing advancements in fields where human ingenuity has reached its limits and to a world where menial work is removed, allowing us more leisure time.

This dichotomy of fear and anticipation about the technological future is seen in all spheres, and betting has certainly not been immune.

The Dick Francis novel Twice Shy, published in 1981, tells the story of a physics teacher who must return a computerised betting system to its rightful owners. It raises the question of how any armchair punter will be able to compete against those using technology to aid their betting.

This literary trope goes strong today, too. A recent bookmaker television advertisement is set in a dystopian world which sees a human competitor competing with technology. It opens with the line:

“They say that machines have won, that algorithms have turned sports into a maths problem…”

The climax, as you might expect, is that the human smashes the machines (with a handily available football)! The voice-over assures us that sports betting remains “beautifully unpredictable”, provided you place your bets with them, apparently.

Cultural references like these presumably play on a real punter fear that in a world where machines are increasingly utilised to help others bet, the little guy will never be able to profit.

And putting unlikely thriller novels and overblown bookie adverts aside for a second, the fear is reasonable. After all, if the likes of Deep Blue, Stockfish and Alphazero can obliterate humans in chess, couldn’t similar machines do the same in betting markets? In the end, aren’t such markets just sophisticated game-playing environments, with a set of rules and a way to win and lose? Unleashed in such an environment, couldn’t a system operating a machine-learning protocol outperform any human?

If betting markets will one day soon be dominated by syndicates operating machine-learning robots running from super-computer platforms, what chance is there for the lone punter with a spreadsheet running on an outdated PC?

These questions are valid, but it’s important to be measured in the conclusions we draw.


The complexity of betting markets

First, betting markets are infinitely more complex than the lab-based or game-playing environments where artificial intelligence has been unleashed. In games like chess and go, there are clearly-explainable rules, with a finite number of moves that can be made at any point, and the objectives are very clear: to try and win.

With punting, it is harder to define what data should be analysed, and what its accuracy and value is. The objectives are cloudier, too. Programming a computer to recognise that a loss might still represent probabilistic value is not straightforward.

So even if machine-driven algorithms are to take over the betting world, as the bookie advert suggests, it’s unlikely to be anytime soon. The task is complex and, currently, the technology is not widely available. AlphaZero was powered by a supercomputer owned by Google and its team of developers trace their lineage all the way back to the IBM team involved with Deep Blue. AlphaZero’s defeat of Stockfish represented nearly 40 years of development.

To take on the more complex task of betting markets, then, will take time and significant investment.

The hope for armchair punters is that the liquidity and available returns in betting markets is not significant enough to warrant that level of investment. It is noted that hedge funds set up to try and bring the quantitative analysis approaches of the financial markets to punting have not yet met with widespread success.

Centaur Corporate’s Galileo fund ceased trading in 2012 having lost $2.5 million and it’s unclear whether London-based Strategem – recently set up with the advertised aim using machine-learning robots to bet on sports – is being successful in raising the £25 million it wanted for a betting fund. Investors might well be nervous that the task is difficult, the development costs significant, and success far from certain. The armchair punter can perhaps rest easy; their fate is not yet sealed.



Second, it’s important to set the development of machine-learning protocols into context. They represent an evolution of technology, not some paradigm shift to a new existence. Punters have long been impacted by technological developments. When the early bookmakers worked out the mathematics of an over-round and recorded bets into a ledger (both technological advancements), this set into motion a few centuries where the individual punter was largely powerless to defeat the institutions that had access to those technologies.

More recent technological developments have been better for punters, though. Global communications have meant that information once only available to a few is now freely shared.

The launch of Betfair has driven down betting margins to a point where casual punters can reasonably expect to profit.

And cheap home computing has meant that anyone can quickly assess data which previously required significant human resource. Technology has meant that it’s never been a better time to be a lone, casual punter, and there’s no reason machine-learning should necessarily change that.

As an example, an initiative known as Leela Chess Zero has sought to replicate the success of AlphaZero. However it uses a distributed network of computers to run from which consists of those people willing to download its software onto their home PCs. This suggests that, even if machine-learning protocols do drive improvements in betting analysis, those improvements may well be available for all to benefit from. Not hidden from view by a small number of large betting syndicates.


Becoming a profitable punter

Lastly, the theme of many of these articles on The Hub has been that anyone can become a profitable punter. So long as they can take steps to eradicate the unhelpful cognitive biases (that we all possess) that lead to poor decision-making. What technology like machine-learning protocols offers us is the promise of highlighting those biases to us, so that we can overcome them.

This is what is happening in the world of chess and there is no reason to think that the same thing won’t happen for us as punters. YouTube is now chock-full of videos analysing the latest brilliancies of our new chess engines. Some of which have fundamentally shattered long-held biases about what represents good game play.

I don’t think sports betting is “beautifully unpredictable” as the bookie advert claims. I would claim the opposite. But I don’t adhere to their dystopian vision of machines and algorithms either. In recent years technology has allowed the little guy to compete with the big fish, and as long as we take the game seriously, it can continue to do so.


ABOUT THE AUTHOR – JACK HOUGHTON 

As a passionate sports’ fan and punter, Jack has written about sports and betting for over a decade, winning the Martin Wills Award for racing journalism in 2002 and writing Winning on Betfair for Dummies, first published in 2006 and now in its second edition, having sold over 35,000 copies in two languages.


Related Articles

Psychology of Financial Crashes

Most of the articles in this series on the psychology of betting have been concerned with our in-built cognitive ...

Framing Effect Bias – Part 2

Last time we looked at the Framing Effect Bias, and how the language used in data presented to us ...

Loss Aversion

When it comes to the psychology of betting, a lot of focus is given to how we stop ourselves ...