How a World of Warcraft bug could provide insights into the world’s latest epidemic.

On the 13th of September, 2005, there broke out one of the worst plagues ever seen in the virtual world.

The plague happened in World of Warcraft, a popular mass multiplayer online role play game or MMORPG. This affected the gameplay of about 4 million players worldwide.

The disease started after players downloaded a software update, which included a new challenge for high-level characters. A new area, with a new powerful creature to fight.

In Zul’Gurub, the capital city of the jungle troll tribes of the virtual world, players could enter a temple inhabited by a powerful snake-like troll named Hakkar the Soulflayer. Any players starting a fight with the beast fell ill under its spell. And, a player under Hakkar’s spell saw his life points decrease every few seconds.

Since the only players able to reach the Soulflayer were of a high-level, the spell generally had a benign effect on them. Upon winning the fight against Hakkar, players healed themselves and went back to their normal lives. All well and good.

But.

On their way back, while celebrating their victory, players forgot to heal their little companions.

Players often adopt pets for bonuses. However, since the penalty of losing these non-player companions is very high, players protect them from fights and keep them in stasis — a suspended state, where they’re effectively frozen and don’t do anything. In stasis, affected pets don’t lose any life points, which means the player has plenty of time to find ways to heal them.

But when these pets were finally released from stasis, far out of Hakkar’s temple and in the more accessible reaches of the world, they released the Soulflayer’s deadly disease from their bodies. This caused outbreaks of the disease wherever they were taken out of stasis.

Thus began Corrupted Blood, the largest epidemic ever seen in a virtual world.

Once the disease was transmitted from the pets to players, it became transmissible between players and highly contagious. The disease spread quickly and easily, thanks to the high mobility of the characters in the game. In densely populated areas like cities, the disease contaminated a huge number of players.

High-level players survived, and spread the disease somewhere else before finding a cure. Low-level players, who weren’t supposed to encounter Hakkar and his disease, simply dropped dead within a few seconds.

The outbreak could have stopped with the death of the contaminated low-level players and the curation of the high-level players, but there was also a third entity at play: the non-player character.

Non-player characters are characters who are not controlled by a player, but by the computer. Such characters often play a key role in the gameplay, for example, the guard of the city gives quest for players to hunt animals or kill monsters, and the shopkeeper sells outfits to increase players’ power. They can’t be killed — because then who’d be there to guard the city or sell stuff to future players?

During the Corrupted Blood outbreak, computer-controlled characters carried the disease but were too strong to receive any damage from it: after all, which non-player character is programmed to die or vanish away? These people were what epidemiologists would call “healthy carriers”. Since they could neither be healed nor die, they kept spreading the disease to players interacting with them, thus maintaining the epidemic.

The disease stayed active for a month, and it was  only when the creators of the game realised the bug, and reset the servers after fixing it, that the disease finally disappeared.

The virtual epidemic which happened in World of Warcraft was very similar to real epidemics.

First, diseases often originate from human encounter with infected animals. Think of the Ebola virus, thought to have originated from bats and later “jumped the species barrier” by adapting to infect humans instead. Or swine flu, where much the same thing happened with (you guessed it) pigs.

Second, diseases can spread easily thanks to the long-distance movement of infected people. In today’s hyperconnected world, where an overnight flight can get you halfway round the globe, this is even more of a problem than in the past. There’s that much less chance of a disease being isolated in a remote corner of the world and never spreading because nobody moves out till it’s all over.

Third, diseases can be maintained through healthy carriers, who don’t show any symptoms on their own but contribute to infecting others anyway. This happens if people are resistant enough to a virus that they don’t fall ill, but not resistant enough to destroy it completely.

And, last but not the least, the way players behaved during Corrupted Blood was very close to the way people behave during a real epidemic.

Most models predicting the spread of disease are based on “computerised rational algorithms”. In such models, decisions are taken based on maths and rational “costs-benefit analysis”. Option A is more likely to save people than Option B, the algorithms say, so Option A is what everyone’s going to choose.

There is one major limit to these models. Machines are not dumb enough.

We know thanks to Richard Thaler, Nobel laureate in Economics, that people do not spend their money rationally. And when facing an epidemic, people are not rational neither. Human are not machines.

The thing about humans is, they’re very influenced by how options are presented to them. Consider the following scenario, where 600 people are sick from a deadly disease, and you have to choose between two policies:

• Policy A will save 200 people for sure.
• Policy B offers a one-third chance to save everyone, but a two-thirds chance that all 600 patients will die.

In this situation, most people tend to take the first option, Policy A, which is less risky. But imagine another scenario, where again 600 people are sick from a disease, and you have to choose between the following policies:

• If you go with Policy C, 400 people will die for sure.
• If you go with Policy D, there’s a one-third chance of killing no-one and a two-thirds chance of killing everyone.

If you’re like most people, you’d tend to choose the more risky option, Policy D, since there’s at least some chance that everyone will get saved.

So far so good. But if you read the two scenarios again, you’ll see that, mathematically speaking, policy A is the same as C, and policy B is the same as D. So it’s not logical for people to choose A over B but D over C (it’s like saying “A is better than B but B is better than A”). No rational, mathematical creature would do it — but real humans have no such qualms.

And that’s why video games are interesting. By watching real human reactions, these games can help us better understand human behaviour under an epidemic, and improve our models accordingly.

During the Corrupted Blood epidemic in World of Warcraft, the behaviour of the players was very similar to real people facing an epidemic. Players helped each other. They ran away from affected areas and went to the countryside. Some tried to spread the disease to as many players as possible.

Some players voluntarily went to contaminated areas out of  curiosity; they wanted to know what was going on and thereby contributed to the spread of the disease. In real life, this could be associated to journalists or reporters going to infected areas and coming back to sane regions to write their reports — or perhaps just by curious ordinary people. (Hint: don’t be one of them).

After few days of outbreak, players organised themselves. They started quarantines in the affected cities, and organised shifts of healing groups to cure the affected players and reduce the spread of the disease. Some players considered the disease a “fun experience”, and tried to keep the epidemic going as long as possible. They avoided being healed, thus spreading the disease in cities and to non-player characters.

Video games offer a cheap and safe experimental setting to observe human irrational behaviours that computers cannot imitate. Researchers can estimate the occurrence of such behaviours and add them to current models, to better anticipate future epidemic outbreaks like the current Coronavirus.

Unfortunately, in the case of the Corrupted Blood crisis, nothing was put in place to measure human behaviours. Researchers who heard of the virtual epidemic, realised the potential of video games to get better data for their mathematical models, but they didn’t manage to actually collect that data.

What they did manage to do is get the conversation going. Researchers and video game designers are now working together to improve the way they record and monitor epidemic outbreaks.

So when the next epidemic breaks out, real or virtual, we’ll be ready.