CrowdVAR: Could Kleros Become the Football Referee of the Future?
By Federico Ast and Stuart Jackson
With the World Cup currently taking centre stage over most of daily life, we’ve seen some interesting discussions within the Kleros Telegram about VAR (Video Assisted Referee) and whether Kleros could become a viable alternative.
The Russia World Cup is the first to implement this technology for assisting referees. This is a step towards higher transparency in a FIFA organization often accused of match fixing, corruption and other nefarious deeds. But more improvements could be made. Even if VAR tends to reduce the likelihood of referee mistakes, the decision is still made by just one person.
Kleros proposes the use of collective intelligence as a neutral third party for dispute resolution. Could it also become a “crowd referee” enabling transparency in football?
How it Would Work
Some features of professional football make it, at least at first glance, suitable for a crowdsourced referee approach. Games are usually broadcasted, which means that potential jurors from all over the world could access the same evidence that referees see in VAR. The rules of football are well known and are the same in every country. Any fan could tell if a specific play was or not a penalty.
In order to be drawn as jurors, users would have to deposit their PNK (Kleros’ token) into, say, a Live Football Subcourt. A predefined number of jurors would be randomly drawn from all those who deposited their tokens. After the selection is fine, jurors would be shown the evidence and would have to make the decision (“Was that a penalty? Yes/No”). Those voting coherently with the majority would keep their PNK, while those voting incoherently would lose theirs. The incentive system would work as described in our white paper.
The beauty of a “live court” is two fold:
- Jurors would have a constant stream of cases to arbitrate quickly. This creates an ecosystem of permanent work hours potentially giving users something that resembles “Full Time” employment.
- End users will be guaranteed a quick turn around in “simple” cases where a human decision is all that’s needed to confirm the outcome of the case.
Both of these factors produce a steady liquidity which keeps the protocol turning efficiently.
Preventing National 51% Attacks
Say Brazil (207 million people) is playing against Iceland (334,000 people). Brazilian supporters could try to do a 51% attack on the court. Since they want their team to win and they hold the vast majority of the tokens, they could try to vote for Brazil in every CrowdVAR decision.
But there is a defense against this.
In Kleros, users can only select the subcourt where they want to participate as jurors. But they cannot select the specific game. The Live Football Subcourt could have, say, 1000 simultaneous CrowdVAR decisions and only one of them would be the Brazil vs. Iceland game. If Brazilian fans wanted to flood the court, lots of them would be sent to decisions in Danubio-Nacional from the Uruguay League or Erchim-Bayangol from the Mongolia First Division.
On average, a large country’s fan base would not be able to overwhelm the system unless it had 51% of the total juror pool. If there is an arbitration decision involving the Brazilian national team and, say, 10% of the total juror pool consists of Brazil fans, that may tilt the results in the case of a particularly difficult arbitration decision, but it would likely not be enough to result in blatantly bad calls.
One might think that there is so much money floating around on sports bets, that bettors could pirate such a system by offering bribes that dwarf whatever internal incentives Kleros uses.
In practice, however, it would likely be difficult for groups of bribers to coordinate themselves to make a common attack in real time. So real threats would only be likely to come from individual parties with very large stakes in the outcome of the match. Nonetheless, using jury sizes in early round voting much larger than what Kleros normally uses would be necessary to combat this.
Of Human Intent: Could Kleros Work Where Sensors and AI fall short?
One could think that progress in AI and its combination with sensors could make human referees obsolete. This may be true for more “objective” calls based on facts that were historically made by the referee “eye sensor” combined with human intelligence. For example, “Did the ball cross the line?” and “Was the player offside?”. These are questions that could be solved in the near future by Goal Line Technology.
But a big part of a referee work is about interpretation. It’s not just about judging whether a player touched the ball with the hand, but also about judging the intention: “Was that handball intentional or not?”
In the long run, players could have brain sensors and that data could be processed in real time by some AI algorithm that detects intentionality in the handball. But in the near term, this is where humans still excel. This is where human jurors organized by Kleros’ cryptoeconomic system could offer the higher value.
Where Kleros Falls Short…
A key consideration for the use of CrowdVAR is that it should not ruin the game experience. The system wouldn’t be very appealing if it were to introduce long delays in the decision. And this is where, at least in the current state of technology, Kleros seems to fall short.
First, because blockchain technology itself isn’t ready for real time applications. It’s still too slow. Second, because Kleros’ cryptoeconomic system may also be slow for real time decisions. A key element for keeping jurors honest is the appeal system. Malicious first round jurors know that, in case of an appeal, second round jurors could overturn their ruling and make them lose the staked tokens. However, if we had to wait for appeals to be solved in the middle of the match, the delay could affect the game experience.
There may be ways to solve this. For example, by just doing the first round voting during the game (and enforce that decision) and then doing the appeals after the match, which will affect token redistribution. This is not ideal, because wrong decisions would be enforced sometimes. But it would still seem better than the current situation where no appeal system exists and the decision is made by just one referee.
Even though, as was noted, the larger number of jurors in CrowdVAR make collusion and bribery extremely hard, at some point, there’s a trade-off between fairness and speed. Sometimes a higher risk of a wrong decision could be acceptable if it allows for a faster call.
Bottom line, limitations still exist for the use of Kleros in football. Some come from Ethereum scalability and others from Kleros itself. The goal of this post was only to expand the horizons about the potential use of our technology in the long run.
Perhaps the main challenges are not technical, but cultural and organizational. Even though VAR seems to allow for better referee decisions in Russia 2018, some have decried its use as “taking the randomness” out of football. And FIFA is notoriously slow in adopting new refereeing technology.
It’s worth pointing, however, that just as any other human practice, refereeing has evolved throughout the ages. And the trend was always towards adding more eyes and brains in decision making. In the 19th century, in the early days of football, there was only one referee (imagine how hard it was to rule an offside). Linesmen came later, adding two extra pair of eyes. The fourth referee was introduced in 1991, and the fifth referee debuted in Germany 2006 World Cup.
There’s no reason to stop there. Evolution seems to keep moving in the direction of the crowd referee. Under the right incentives, 10,000 eyes assisted by AI can make better decisions and bring more fairness and transparency into the beautiful game.
Federico Ast. CEO at Kleros.
Stuart Jackson. Community Manager at Kleros.
Join the community chat on Telegram.
Visit our website.
Follow us on Twitter.
Join our Slack for developer conversations.
Contribute on Github.