Playtesting with AI – a new game changer
Much like the heroes in a game, game developers are the heroes behind the scenes. Building entire worlds and making them come alive is no easy feat; from creating multiple complexity levels, crafting new characters, or putting new missions to the test, there’s always something exciting going on.
Nowadays, as the IT industry continues to dip their toes into the world of gaming coupled with artificial intelligence, there’s one interesting area in particular that dramatically benefits from employing one sub-area of AI, which is Deep Reinforcement Learning. The area we’re referring to is playtesting.
For years, playtesting has been human-driven. Essentially, humans perform exhaustive playtests to find gaps, bugs, or verify a balance. When the opportunities to improve are discovered and communicated to game developers, they implement fixes to improve the quality of the game.
Now, as you may have guessed, human playtesting can be inaccurate, prone to biased opinions, and in all honesty, humans grow tired a lot easier than algorithms do — and on the contrary, can’t play below their skill after mastering a game.
With Deep Reinforcement Learning, playtesting is a different ballgame and there’s a lot to be said of the advantages it infuses into what was typically a costly, lengthy, and complex task.
Deep Reinforcement Learning, in which the technology learns how to playtest efficiently and accurately, can take sequential actions over time and provides the optimal framework to achieve high levels of accuracy. Algorithms are trained to devise their own solutions in different playtesting scenarios and are already showing highly promising results.
For one, Deep Reinforcement Learning can reduce playtesting from hours to minutes with automated tests.
Game developers, in the spirit of meeting the increasing demand for new content, need the quickest means possible to achieve meticulous playtesting and run it at scale. With automated Deep Reinforcement Learning, game developers receive feedback faster and wait times are reduced, all leading to improved game balance, superior user experience and opportunity to focus more on creativity and quality.
On a per-level basis, automated playtesting with AI agents embedded with Deep Reinforcement Learning allows for more iterations, enabling game developers to fine-tune the game quickly and effectively.
Even so, many industry experts still consider that Deep Reinforcement Learning has not reached its full potential yet. One reason as to why this intelligent technology has not reached all it can is due to the lack of expertise in the field. Deep Reinforcement Learning techniques require design prowess and with that in mind, game developers need to overcome many challenges. For example:
- Controlling AI agents on target so they learn how to behave in unseen scenarios. The virtual agents need an ongoing stream of data to learn.
- Managing unpredictability as it has proved to be efficient in synthetic scenarios like playtesting but still has a long road ahead when it comes to real-life scenarios.
- Overcoming the time complexity as it is hard to predict how much time an AI agent needs in a real data environment.
- Balancing different design constraints, specifically AI personas behaving in a specific way.
The road for game developers in Deep Reinforcement is still under construction. Nevertheless, the array of benefits it brings are more than worth incorporating in playtesting with a developer-centered approach, infusing promising solutions to an industry that requires more and more high-quality content.