How Artificial Intelligence Is Revolutionizing Game Testing and Game Play

Lionbridge data scientist presents cutting-edge, game-related AI research at premier conference

Queen to e4. Checkmate! I lost again… to the game! Artificial Intelligence (AI) has left mere mortals in the dust when it comes to computer games. As frustrating as it is, our minds simply cannot process the millions of possibilities a computer can handle within nanoseconds. 

However, for us nerd gamers, AI also offers a significant opportunity. Can we use AI to break computer games to make them better? And can we train AI to play even more like humans for enhanced results?

These are the questions Lionbridge’s Emmanuelle Rodrigues Nunes ponders every day. A Brazilian native now based in the U.K., Nunes is a Senior Data Scientist at Lionbridge who is assigned to work with Microsoft. She uses complicated AI strategies to make games more stable, more reliable, and more fun to play—all at scale.

This fall, she presented two abstracts on AI’s revolutionary impact on game play during the Royal Statistical Society (RSS) International Conference in Manchester, England.  Nunes is both a member and fellow at RSS.

What AI Developments Promote Better Game Playing? 

Evolutionary Strategies algorithms

In her first abstract, Nunes explored Evolutionary Strategies (ES) algorithms as an alternative to the prevalent Reinforcement Learning techniques. How is that different? Put plainly, ES algorithms are inspired by the same evolutionary mutations we see in nature. Like Darwin’s theory of evolution, these algorithms pick the fittest option, which then becomes the starting point for the next generation. ES algorithms are capable of self-adaptation. They continuously improve, with scalability and simplicity among their key differentiators. 

Why does this improvement matter to a game developer or the average gamer? That’s a great question. The Quality Assurance (QA) process aims to break the game over and over. It then asks, “How do we fix it?” However, in typical game play, that is not normal behavior. Algorithms that can emulate regular human play give us astonishingly predictable barometers to determine the likelihood of an application encountering a fatal error and unexpectedly crashing/terminating and how stable the application is, all within the context of typical user behavior. 

The usage of ES algorithms provides a more accurate and representative measure of product health and how issues may impact users. Also, unlike its human counterparts, the algorithms can play 24/7/365. The technology does not need breaks or any of the other necessities that testers need. So, while we sleep, eat, and work, the game plays and plays and plays.

Measuring game stability through survival analysis

Nunes’s second project—although different from ES algorithms—is also related to a game’s health metrics. It’s a new way to understand how stable the application is. This is important because stability is paramount in the overall end-user experience. A failure in stability can also greatly detract from a game’s in-app objective(s) from being achieved and damage trust in the application’s reliability. While 100% stability is always the target, this is rarely achievable. As such, it is important to measure this key quality factor in relation to acceptable tolerances. Leveraging Survival Analysis concepts gives us a fuller understanding of the health of our gaming products over the lifecycle. They also give us insights into stability. 

All this complexity solves some very basic problems. It enables developers to know when a game is ready for launch. These types of automations also quickly provide developers with confidence and certainty that was never previously available. And they work at scale.

How Can Automations and People Work Together to Enhance Game Testing? 

Even just a few years ago, most QA was based on tester data only. While cutting-edge technological developments are in and of themselves game changing, Nunes says people will still have an important role. Nunes advocates for automation and human testers to be used in a complementary way.

As “human” as the algorithms are, they cannot emulate the mundane errors a person will make. Most importantly, they cannot comment on game elements that are bothersome, distracting, or frustrating. Nor can they revel in the utter magic of a beautifully developed game. Those attributes remain firmly part of the human condition and an invaluable part of the process. 

With automation on the horizon for quality assurance, we can expect faster time to market, more efficiency, and an overall better experience for the end-user. This way of using automation is new, always changing, and evolving quickly. Nunes looks forward to continuing to tackle these challenges. She hopes to make the use of these technological advances a standard. This way, we can elevate game play across the industry. 

linkedin sharing button
  • #game-testing
  • #player-experience
Davida Wexler with Janette Mandell