Almost four years ago to this day, 16 teams advanced into the knockout stage of the 2014 FIFA World Cup in Rio De Janiero. A few short weeks later and there were only four teams remaining – Brazil, Netherlands, Argentina and Germany – and, (four year old) spoiler alert, Germany won!

But what’s (arguably) more spectacular is that Microsoft managed to correctly predict the results of the entire knockout stage of the tournament, including Germany’s 1-0 win over Argentina.

“It’s been a fun computer science experience here and we think a lot about where we can take it,” said Microsoft’s Director of Consumer Communications, Craig Beilinson.

And that happened four years ago. In the last few years, artificial intelligence (AI) has come a long way. In an era of self-driving cars, smart suitcases and…well…robots…surely we have the right technology to predict the winners of a simple football tournament, right? Or are we better off leaving football predictions to chance – or to the likes of Paul the Octopus.

Yes and no. (A complex question warrants a complex answer.)

Do we have the right technology?

AI is currently achieving some amazing breakthroughs. Earlier this year the government unveiled its work with artificial intelligence agency ASI Data Science, which created a software that was configured to accurately detect 94% of video uploads from the terrorist group Islamic State and stop it from being viewed.

The software has essentially been “trained” to automatically pick up on extremist content – so, using the same (or similar) technology, is it possible to observe behavioural patterns from previous games, pinpoint key patterns and predict accurate results for sports games?

The UK sports betting company Stratagem have created a software that makes calculations while watching a live broadcast feed of a match - identifying what it believes to be goal-scoring chances (the moments where players have an opportunity to shoot and score).

“Football is such a low-scoring game that you need to focus on these sorts of metrics to make predictions,” said Stratagem’s Founder, Andreas Koukorinis, in an interview with The Verge. 

“If there’s a shot on target from 30 yards with 11 people in front of the striker and that ends in a goal, yes, it looks spectacular on TV, but it’s not exciting for us. Because if you repeat it 100 times the outcomes won’t be the same. But if you have Lionel Messi running down the pitch and he’s one-on-one with the goalie, the conversion rate on that is 80%. 

“We look at what created that situation. We try to take the randomness out, and look at how good the teams are at what they’re trying to do, which is generate goal-scoring opportunities.”

Despite this technology, Stratagem’s software is only accurate 50% of the time - generating most of its data about goal-scoring opportunities the old-fashioned way – with humans. 

A recent study by Rory P.Bunker and Fadi Thabtah looked at the framework of machine learning for sports results predictions, and found:

“ML (machine learning) seems an appropriate methodology for sport prediction since it generates predictive models that can predict match results using predefined features in a historical dataset.”

But, while machine learning models are becoming increasingly popular for sport predictions, the study concluded that more accurate models are needed. 

So, who will win the World Cup?

Back in 2010, you may remember a particularly impressive octopus named Paul who made headlines for accurately predicting the outcomes of World Cup matches. 

And while Paul may not be around to predict this year’s results (may he rest in peace), FIFA have taken a shot at predicting 2018’s winners: Germany will apparently win for the second time in a row, followed by Brazil, Belgium and Portugal. 

But the US-based data company, Gracenote, have a different theory – and their top pick isn’t Germany. 

By analysing team ratings – which includes match results, location, and the importance of the match (such as friendlies) – Gracenote used a predictive algorithm that runs over a million times, producing the chance each team has at advancing in the tournament.

Gracenote’s algorithm found Brazil to be the most likely team to win the World Cup in Russia – assigning them a 21% chance of taking home the coveted trophy. Second, third and fourth went out to Spain, Germany and Argentina.

The algorithm also revealed a 31% chance that either Germany or Brazil will finish as group runner-up, which would spark a second-round knockout matchup between the two football giants.

Gracenote’s predictions also revealed Peru to be a potential dark horse. Having qualified for the first time in more than 35 years, Peru have remained unbeaten in the past 14 matches, and were given an impressive 68% chance of progressing in its group. (Their group also includes France, Denmark and Australia.)

The South Americans were also given a 39% chance of reaching the quarterfinals, and a 22% chance of making their way to the semi-finals. 

And sure, Gracenote’s predictions could bomb - but their approach could also prove to revolutionise sporting predictions going forth. Last year the company correctly predicted Real Madrid would win the Champions League tournament using the same algorithm as they’ve used for the World Cup – so, looking at their own track-record, statistically, they could have a solid chance of being right.