When Games Become Laboratories: How Virtual Worlds Are Shaping Real Science

April 2025  ·  Adrian Cowell  ·  Innovation, AI & Simulation

In September 2005, a software bug in a dungeon raid spawned what epidemiologists would later describe as one of the most instructive disease outbreak simulations ever observed. No ethics committee approved it. No grant funded it. Eight million players simply logged into World of Warcraft and, over the following weeks, unwittingly demonstrated the mechanics of a pandemic with a fidelity that mathematical models had never quite achieved.

That incident; the Corrupted Blood plague; is now well known in epidemiological circles. But it sits at the beginning of a much longer story about what games and virtual environments can teach us when the lines between play and research dissolve. That story is accelerating rapidly, and for those working in medical education and healthcare simulation, it has direct implications for what comes next.

The Accidental Pandemic of Azeroth

The Corrupted Blood incident began with a programming oversight. A boss encounter in World of Warcraft’s Zul’Gurub dungeon was given a debuff spell intended to affect only players inside the dungeon. A flaw allowed players’ virtual pets to carry the debuff out into the wider game world, where it spread through cities, killing low-level characters instantly and ravaging higher-level ones continuously. Major population centres became ghost towns. Some players fled. Some tried to help. Some deliberately spread the infection.

Dr Eric Lofgren and Dr Nina Fefferman, writing in The Lancet Infectious Diseases in 2007, identified what made this significant: it was not the mechanics of transmission that mirrored a real outbreak, but the human behaviour [1]. Epidemiological models typically abstract human behaviour into fixed assumptions; compliance rates, movement patterns, quarantine responses. Corrupted Blood showed none of those assumptions held cleanly. Players with high-level characters ran into infected areas to help others. Curious onlookers gathered near outbreaks. Virtual healthcare workers contracted and spread the disease. People who were told to quarantine simply did not.

When COVID-19 arrived in 2020, researchers returned to Corrupted Blood immediately. The parallels were striking: a highly contagious, rapidly spreading disease; inconsistent compliance with public health guidance; superspreader behaviours; and the emergence of informal volunteer networks alongside deliberate sabotage. The Doherty Institute published analysis in 2022 examining the extent to which the virtual outbreak genuinely predicted real-world pandemic behaviour patterns, concluding that the data, while imperfect, offered insights that traditional models had missed [2].

The limitation was honest: the data had not been collected for research, and player behaviour in a game is not identical to behaviour under genuine mortal threat. But the incident established something important. Virtual worlds, because they involve real people making real decisions with emotional stakes, capture a kind of behavioural truth that pure modelling cannot.

EVE Online: The Economy as Laboratory

While epidemiologists were finding lessons in Azeroth, economists were discovering a rather different laboratory in the universe of EVE Online. CCP Games, the Icelandic developer behind the space-based MMO, built an economy with minimal developer intervention; a player-driven market with thousands of goods, supply chains, speculative trading, and price formation driven entirely by supply and demand. It was, in effect, a near-perfect economic sandbox.

In 2007, CCP hired Dr Eyjólfur Guðmundsson, an economist with a PhD from the University of Iceland, as the game’s Chief Economist; making CCP the first game company to employ a professional economist in an analytical role. Quarterly economic reports began to be published, analysing inflation, market volumes, regional price variations, and the economic effects of in-game events. Researchers at the University of Ghent later found significant correlations between real-world economic indicators; unemployment rates, exchange rates; and player trading behaviour in the game, suggesting that virtual economies reflect real-world conditions in ways that could be useful for economic modelling [3].

EVE’s contribution to science did not stop at economics. The game launched Project Discovery, a citizen science programme embedded directly into gameplay, in which players analysed real scientific data as a game mechanic. Phases of the project tasked players with mapping protein structures from the Human Protein Atlas, classifying exoplanet transits, and; during COVID-19; mapping flow cytometry data to help scientists understand immune responses. By crowdsourcing that final task within the game’s active player base, the project accumulated 1.37 million data points and was estimated to have saved scientists 330 years of equivalent analysis time [4].

Foldit: Players Solving Science’s Hardest Problems

If EVE represents citizen science at scale, Foldit represents something more targeted: deliberately designing a game to solve a specific scientific problem that computers alone could not crack.

Developed at the University of Washington in 2008, Foldit presents players with protein folding puzzles. The challenge of predicting how a protein will fold from its amino acid sequence had defeated computational approaches for decades; the search space of possible configurations is astronomically large, and human spatial intuition turned out to be surprisingly effective at navigating it. In 2011, Foldit players solved the crystal structure of a retroviral protease from the Mason-Pfizer monkey virus, a problem that had resisted computational approaches for fifteen years. Players produced an accurate 3D model in ten days [5].

When COVID-19 emerged, Foldit was repurposed rapidly. In February 2020, players were challenged to design protein binders targeting the coronavirus spike protein. Thousands of designs were submitted across three rounds. Researchers selected 99 promising candidates for real-world synthesis and testing as potential antiviral agents [6]. A game had generated a shortlist of anti-COVID drug candidates within weeks of the pandemic’s onset.

Plague Inc. and the CDC

Not all game-science crossovers arise by accident. James Vaughan built Plague Inc.; a strategy game in which the player’s goal is to evolve a pathogen and exterminate humanity; with genuine epidemiological mechanics at its core. Transmission routes, incubation periods, symptom severity, drug resistance: the model is simplified, but it is grounded in how infectious disease actually works.

In 2013, Vaughan was invited to speak at the United States Centers for Disease Control and Prevention, where the game was praised by Ali S. Khan, then Assistant Surgeon General, for raising public awareness of epidemiology through a “non-traditional route” [7]. When COVID-19 arrived in January 2020, Plague Inc. became the top-selling app in China almost overnight and the top paid app on the iOS App Store globally by February. It had been downloaded over 160 million times by 2021.

Ndemic Creations responded by adding real CDC outbreak bulletins to the game and, in collaboration with the WHO, CEPI, and GOARN, releasing a dedicated “Cure Mode” in which the player’s objective was reversed: stop the pathogen rather than spread it. Published research in Games for Health Journal found that engagement with Plague Inc. during the early pandemic period was associated with measurably improved understanding of disease transmission mechanics and public health intervention strategies among players [8].

Serious Games in Healthcare: The Shift from Entertainment to Infrastructure

The examples above all describe games that found scientific utility either by accident or through creative repurposing. A parallel track has been developing more deliberately in healthcare simulation: the design of purpose-built serious games for clinical training.

Brighton and Sussex University Hospitals NHS Trust uses a simulation game called The Floor routinely in Major Incident training for emergency department staff, and it forms part of the formal teaching programme for junior doctors across grades [9]. Bournemouth University, in collaboration with University Hospitals Dorset, developed Admission, an immersive simulation game designed to prepare junior doctors for the competing pressures of ward life before they encounter them in practice [10]. The military has invested heavily in gamified trauma training, with research published in JMIR Serious Games demonstrating improved preparation for austere medical conditions through game-based simulation [11].

A scoping review published in Frontiers in Public Health in 2022, covering nearly two decades of literature, found consistent evidence that serious games improve knowledge acquisition, clinical decision-making, and procedural skill compared to traditional instructional approaches, particularly for scenarios involving high-stakes, low-frequency events [12]. The argument for simulation-based training is not merely pedagogical efficiency. It is patient safety. Errors made in a simulation carry no consequence. Errors made in a ward do.

The New Frontier: AI Agents Playing All the Roles

The most significant recent development in this space is a departure from human players entirely. The progression from Corrupted Blood (humans behaving emergently in a game) through Foldit (humans solving structured problems in a game) to the current frontier of AI agent simulation represents a qualitative shift in what is possible.

Stanford and Google’s AI Town project in 2023 demonstrated that LLM-powered agents could inhabit a persistent simulated environment, develop social relationships, spread information, form plans, and even organise elections; without human direction. The agents behaved with a coherence and social complexity that surprised the research community. Behaviour that had previously required thousands of human participants to observe could now be instantiated with twenty-five agents and a few API calls.

Tsinghua University’s Agent Hospital project followed in 2024 [13], applying the same principle to a healthcare context. A fully simulated hospital environment populated by LLM-powered doctor, nurse, and patient agents processed over 10,000 simulated cases and achieved 93.06% accuracy on the MedQA benchmark; a dataset derived from medical licensing examination questions. Crucially, the agents could evolve: physician agents improved their diagnostic accuracy through simulated experience, just as a junior doctor improves through clinical exposure. No real patient was placed at risk in the process.

Research published in Nature Communications Medicine in 2025 validated the use of LLM-powered simulated patient systems for medical education, finding that AI-driven patient simulations offered potential to transform how clinical communication, diagnosis, and management skills are taught and assessed [14].

What This Means for UK Healthcare Simulation

The trajectory described here; from accidental virtual plague to deliberately designed AI hospital; points toward a near-term future in which the NHS and UK medical education institutions can run sophisticated, contextually accurate simulations of entire clinical environments, populated by AI agents with realistic behavioural profiles, at a fraction of the cost of physical simulation facilities.

The value proposition is not simply efficiency. Simulation at scale enables scenarios that no physical facility can reproduce: a hospital under simultaneous pressure from a major incident, a flu surge, and a staffing crisis; a ward where communication failures compound over a twelve-hour shift; a clinical team making sequential decisions with incomplete information under time pressure. These are precisely the conditions in which most serious clinical errors occur, and they are conditions that cannot ethically be created in real environments for training purposes.

Research currently underway is exploring the adaptation of AI Town’s open-source architecture for a UK hospital simulation context, designed specifically to meet the training requirements of the NHS and to align with GMC standards. The aim is not to replace the clinical simulation suites and standardised patient programmes that medical schools already operate, but to provide a complementary layer: high-volume, accessible, endlessly configurable simulation that students and trainees can engage with independently, at any time, from any location. The kind of infrastructure that allows a foundation doctor to encounter a deteriorating septic patient at 2am not for the first time in a real ward, but for the fifteenth time in a simulation they have run themselves.

Games have always been a way of rehearsing reality at a safe distance. What has changed is the fidelity of the rehearsal, and the scale at which it can be conducted. From a WoW dungeon bug in 2005 to an AI-populated hospital simulation in 2024, the distance has been shorter than anyone expected; and the direction of travel has not changed.

---

References

1. Lofgren ET, Fefferman NH. The untapped potential of virtual game worlds to shed light on real world epidemics. The Lancet Infectious Diseases. 2007;7(9):625-629. thelancet.com
2. The Doherty Institute. Did an accidental blood plague in World of Warcraft help scientists model COVID better? The Conversation. 2022. theconversation.com
3. Data Science Milan. How are games used for scientific discovery. datasciencemilan.org
4. PCWorld. How EVE Online players are solving real-world science problems: Meet Project Discovery. pcworld.com
5. Khatib F et al. Crystal structure of a monomeric retroviral protease solved by protein folding game players. Nature Structural & Molecular Biology. 2011. scientificamerican.com
6. GeekWire. A protein puzzle game called Foldit turns up 99 promising ways to confound coronavirus. 2020. geekwire.com
7. Wikipedia. Plague Inc. wikipedia.org
8. Sun Y et al. Health Communication in Games at the Early Stage of COVID-19 Epidemic: A Grounded Theory Study Based on Plague Inc. Games for Health Journal. 2021. pubmed.ncbi.nlm.nih.gov
9. The Floor. Clinical simulation game, BSUH NHS Trust. thefloorgame.com
10. Bournemouth University. Medical simulation game developed by BU and UHD to train junior doctors. 2023. bournemouth.ac.uk
11. JMIR Serious Games. Gamification in the Design of Virtual Patients for Military Medics to Support Trauma Training. 2024. games.jmir.org
12. Frontiers in Public Health. Application of Serious Games in Health Care: Scoping Review and Bibliometric Analysis. 2022. frontiersin.org
13. Li J et al. Agent Hospital: A Simulacrum of Hospital with Evolvable Medical Agents. arXiv. 2024. arxiv.org
14. Nature Communications Medicine. Simulated patient systems powered by large language model-based AI agents offer potential for transforming medical education. 2025. nature.com