How EVE Online’s Project Discovery wrangled half a million players into helping science

For those outside of its sprawling, virtual community of traders, fighters and Space Popes, EVE Online can seem like an incomprehensible tangle. This is the online game, after all, that enabled a corporate lawyer to reimagine himself as an intergalactic tyrant. This is the game where the lines between real and fictional battles frequently blur. This is a game with invisible empires built in remote server centres, populated by half a million players across the internet.

But this is also the game where real-life scientific progress can be made in the folds of gameplay. One corner of the EVE Online universe is Project Discovery: an initiative that started around three years ago, when a physicist and a software engineer pondered the potential for video games to be a force for social good. What if a player’s actions weren’t used only for in-game advancement, but rippled out into laboratories and peer-review journals?

Geneva, 2014: the university researcher Bernard Revaz and the IT expert Attila Szantner meet for a drink. By the end of the year they’ve formed Massively Multiplayer Online Science (MMOS), a company that plugs a lead between scientists and game developers. In 2016, they launched Project Discovery in collaboration with EVE Online; a special pocket of that vast community where players are given rote tasks, but with actual implications to scientific research.

“Game designers tackle the challenge of how to get players engaged with quite repetitive tasks,” Szantner tells me in a London hotel bar, the same morning he is due to receive a Lovie Award for his work with Project Discovery. “In many games, what you’re doing is repeating the same task over and over again, and you enjoy it. In citizen science, we have the same setup. We have to analyse data and are basically doing the same things hundreds of thousands of times.”

The first target for Project Discovery was to reign in EVE’s half a million or so players to help with the Human Protein Atlas (HPA), a Swedish-based programme to map all of the proteins in human cells. The mini-game involved sifting through around 250,000 microscope images from the HPA database, with players identifying and cataloging different cell structures, from nucleus to mitochondria. When enough players agreed on a specific image, it was passed on to the researchers behind the HPA.

“Humans are known to be good at visual pattern recognition,” says Emma Lundberg, director of the HPA’s Cell Atlas. “We spend a lot of time classifying image pattern, and it was intriguing to see if we could use gamers for the same task. Many researchers love the concept, but will not believe it fully until they have seen that high-quality data can be obtained.”

(Above: Project Discovery’s cell classification mini game. Credit: EVE Online)

Szantner tells me his team wanted to come up with a set of tasks that were simple enough not for the player to get bogged down in specialist knowledge, and chimed with the wider game’s story. EVE Online’s creators, Reykjavík-based studio CCP, added narrative dressing to the Project Discovery tasks, and offered in-game rewards for taking part in the initiative. What the team found, however, was that players were more interested in contributing to scientific research than racking up points.   

“There was a special currency introduced in the game for Project Discovery, and we saw that many people didn’t spend it,” says Szantner. “They were just doing it because they wanted to contribute. By far, the strongest motivation is to help science.”

Planets and borderlands

Since July 2017, Project Discovery has shifted into its second stage. Instead of protein mapping, players are now being asked to help the search for planets outside of our solar system (exoplanets). The initiative collaborated with Professor Michel Mayor who, along with Didier Queloz in 1995, discovered the first exoplanet orbiting a sun-like star. With Mayor’s backing, Project Discovery has been feeding players astronomical data from the CoRoT telescope, encouraging them to categorise potential planets.

This translates into a mini-game in which players are given graphs that represent luminosity from one of the 160,000 stars being monitored by the CoRoT telescope. When a body – a potential exoplanet – passes in front of the star, the light dips. Players are asked to identify these dips within a given period; information that can suggest a planet’s size, distance and orbit. Much like the Human Protein Atlas initiative, when enough players agree on a classification, it is handed to scientists – this time at the University of Geneva.

(Above: The real and in-game Michel Mayor. Credit: ESO/Project Discovery)

Project Discovery’s exoplanet hunt is still happening, but where does Szantner see MMOS’s work with video-game citizen science going next? He tells me that “EVE was a perfect match” for crowdsourcing data analysis, and an “easy win” due to the community’s general enthusiasm about scientific progress, but he’s also testing the water for a very different type of collaboration.  

“There is one corporation I can talk about: Gearbox,” he says. “Borderlands is their biggest title. Randy Pitchford, president and founder of Gearbox, is a big fan of this whole concept. He’s very committed to getting this into one of their titles. I’m really interested in that because it’s a completely different audience and genre of game. They’re very creative and clever guys, so I’m absolutely positive that if they say yes, then they’ll do it in a way to fit their game.”

It’s one thing to fit citizen science into a space-based MMO, but another thing to sew it into the fabric of a first-person shooter. Games such as Glitchers’ Sea Hero Quest have shown there’s scope to match navigation and combat with real-world research – in this case, a study of dementia – but these tend to be small games built around a specific purpose. How would a developer approach a bigger game like Borderlands?  

“We’re topic-agnostic,” says Szantner. “We’re platform- and game-agnostic. I think this could be integrated in many types of games, with many different research projects. Spotting animals in the Serengeti is probably not a good fit for EVE, but you could put it into another game.”

I ask Szantner if there’s also scope for this type of approach to be used outside of scientific research. Could video games help with, say, immediate problems such as disaster relief? “It’s a concept, but you could imagine a game like Call of Duty where you have a mission to bomb certain military bases, then you get real images of buildings destroyed by an earthquake,” he suggests, explaining that the research element could involve identifying objects in those real-life pictures.

“It’s really controversial, but if you save lives… Could even a war game contribute to these kind of things?”

There’s clearly an element of tonal disparity here, and Szantner admits his off-the-cuff idea comes with a host of ethical and moral issues. But he’s still adamant that games have scope to crowdsource help: “If we can show that this can also be tooled to save lives, I think that’s beautiful.”

(Above: Project Discovery’s exoplanet hunting minigame. Credit: EVE Online)

For now, at least, the meeting point between video-game worlds and human progress is rooted in citizen science, and even that is a nascent relationship. The results of Project Discovery and HPA’s Cell Atlas collaboration are still being pored over, with Lundberg anticipating the release of a paper that examines the success of player-based categorisation against AI-based image analysis. The publication of papers like this could lay the groundwork for a new conception of citizen science, but also a new idea of what video games can be.

“In our paper we demonstrate that humans and AI are good at different things,” she tells me. “I believe that with clever game design and direct integration of gamer results into machine learning, future citizen science projects can be expected to make a big contribution to large science projects.”

Correction: The original article stated the first project Discovery game involved 13 million microscopic images. This has been changed to reflect that the game actually involved 250,000 images, chosen from the Human Protein Atlas database of 13 million images.