No longer a pejorative waste of time, video games have led to significant, unbelievable advances in medical science.

For much of their history, video games were seen as light entertainment — fun, engaging, and sometimes blamed for causing distraction or escapism or even mental disorders. They were rarely considered scientifically important. But behind the scenes, video games became a key experimental platform for developing modern AI. Before AI was used in hospitals or labs, it was learning to survive, adapt, plan, and optimize in digital worlds. These lessons now drive breakthroughs in medicine, including advances in protein folding that changed biology and helped pave the way for Nobel Prize–recognized work in chemistry.

This isn’t a story about games directly causing medical advances. It’s about what games gave researchers: controlled complexity. A video game is like a small universe with its own rules, limits, feedback, and clear results. In this environment, you can run millions of experiments without risking lives, wasting materials, or waiting months for answers. When you want machines to learn, not just follow instructions, that makes a big difference.

Why Games Became the Perfect AI Training Ground

Video games provided AI researchers with a rare and powerful asset: complex environments governed by consistent rules, rapid feedback, and virtually unlimited opportunities for trial and error. These features made games ideal training grounds for reinforcement learning, a method in which an artificial “agent” improves its performance by receiving rewards or penalties based on its actions. Unlike traditional programming, reinforcement learning does not rely on explicit instructions for every scenario. Instead, systems learn through experience — discovering strategies, correcting mistakes, and gradually improving.

A breakthrough occurred when researchers demonstrated that a single system could learn to play multiple Atari games solely by observing the screen, without specialized strategies for each game. This showed that neural networks could connect what they see with what they do, and get better through practice. Even though these games are simpler than real life, they still require skills that seem very “cognitive”: following moving objects, guessing what will happen next, and making choices when things are uncertain.

Next came games that needed more planning. Go, for example, made AI balance short-term moves with long-term strategy. DeepMind’s AlphaGo demonstrated that learning systems could develop planning skills beyond a single task. Later, MuZero showed that an agent could perform well in games like Atari, Go, chess, and shogi by learning to plan, even without knowing the rules in advance. This approach — learning what matters, then planning — is a big reason why games have been so useful for researchers working toward more flexible AI.

Games also taught AI researchers a practical lesson: how to train agents at scale. Complex games reveal weaknesses. Real-time strategy games like StarCraft II require multitasking, long-term planning, and making decisions in a changing world with limited information. AlphaStar reached Grandmaster level in StarCraft II by using multi-agent reinforcement learning, showing that these training methods can handle messy situations with many choices. The goal wasn’t to make a better gamer, but to build learning systems that can work in complicated, changing conditions — much like real-world problems.

From Winning Games to Solving Protein Folding

Once you see games as learning laboratories, the bridge to medical science becomes easier to understand. Biology and chemistry are full of “environments” too — systems with rules, feedback, tradeoffs, and enormous complexity. The question becomes: can the methods that helped machines learn in games help them learn in the molecular world?
Protein folding was a perfect test.

Proteins are chains of amino acids that fold into precise three-dimensional shapes, and those shapes largely determine function. Misfolding can contribute to disease; correct structures are essential for understanding how drugs bind to targets and how mutations may alter function.

For decades, predicting a protein’s structure from its sequence alone remained a stubborn bottleneck. Fighting disease meant understanding this process, and little progress had been made within decades of research.

AlphaFold changed this. In a 2021 Nature paper, DeepMind showed that AlphaFold could predict protein structures with unmatched accuracy. Protein folding isn’t a game, but the approach is similar to game-trained AI: learn from big datasets, improve predictions step by step, and aim for better results. The impact was so great that many researchers called it a turning point for structural biology.

Then came the distribution effect. The AlphaFold Protein Structure Database made these predicted structures widely available, including through an open-access website. This is important because scientific progress depends not just on discoveries, but also on how quickly and widely they spread. When more labs can access structural data, they can ask better questions, test more ideas, and look for new treatments.

The Medical Ripple Effects: From Lab Bench to Clinic

Protein folding is the main story, but it’s not the only way game-based AI is helping medicine. Reinforcement learning and similar methods are being tested for medical decision-making, where researchers or systems make a series of choices over time and see results slowly. In healthcare, decisions are rarely simple or final — they change based on each patient’s needs. This is exactly the kind of problem reinforcement learning was designed to handle. The complexity and the uniqueness are fundamental to medicine.

A broad guide to deep learning in healthcare, published in Nature Medicine, describes how methods including reinforcement learning may impact medicine alongside computer vision and natural language processing. The benefits are clear: better pattern recognition in images, improved understanding of clinical notes, and more support for complex treatment planning. But healthcare has extra challenges — like safety, bias, and accountability — that aren’t issues when an AI agent fails in a game.

Digital Medicine outlines how reinforcement learning could be applied to personalized treatment decisions across clinical domains, while also stressing the challenges of real-world deployment. That balance is important. Game environments are clean compared to clinical reality. In medicine, data can be messy, patient populations can differ, and “reward signals” (better outcomes) can be delayed or confounded by outside factors.

Radiation therapy planning is one area where the link between simulation and real care is clear. The goal is to give enough radiation to treat a tumor while keeping healthy tissue safe. This is a real-life optimization problem with serious consequences. Researchers have tested deep reinforcement learning systems for automatic planning in intensity-modulated radiation therapy, showing how these systems could help with complex planning. In this case, the “game” is a simulation in which the system learns to improve its plans under strict rules.

So how does the Nobel Prize in Chemistry fit into this story? The Nobel isn’t given for AI or for games. But modern chemistry now relies heavily on advanced computing. In the Nobel Prize’s official background for 2024, the Academy explains the science and the role of modern computational methods in today’s chemical discoveries. Simply put, today’s breakthroughs often come from combining molecular science with machine-based modeling.

The main point is that video games helped create a new way of working. They gave AI researchers safe, detailed environments to build systems that can generalize, plan, and adapt. Once these skills were developed, scientists could use them in biology, chemistry, and medicine. What started as play turned into important tools, and now those tools are speeding up research in diagnosis, treatment, and drug discovery.

If you want the main idea in one sentence, here it is: games taught machines how to learn, and those learning machines changed what science can achieve. Video games didn’t take attention away from serious research; they quietly helped AI get ready to solve some of the hardest problems in nature — problems that affect how long, how well, and how safely people can live. So, let’s not denigrate gaming because it may yet prove to be more than we ever could expect from a simple activity that appeared to be only for entertainment.