GUEST:
It’s a total understatement to say that the growing presence of machine learning in games is big news. The technology promises to turn the industry topsy-turvy by changing our perception of what’s possible in a gaming experience.
Game studios are ground zero.
They’re applying machine learning to first address long-standing challenges in game development and design, like accelerating the process of building out art and levels and balancing the backend infrastructure.
“You can actually tailor game design and tailor levels to an individual’s experience,” George Dolbier, the CTO of Interactive Media at IBM, said at a talk at the Intel Buzz Workshop in Seattle last June.
More interesting for us laymen, it’s being used to create progressional curves that tailor themselves directly to each player according to their individual behavioral data. A simple example would be in a Tetris-like casual game where a player would first go through a few “seed” rounds of gameplay so that the machine learning element can first learn about how they play before it creates a customized roadmap that will ensure maximum engagement.
But this smart tech goes well beyond remedying industry challenges, streamlining otherwise manual processes, or even making games increasingly intuitive. It’s going to literally instil life into computer-generated characters and unprecedented depth into storylines and game mechanics in a way that will make you, if you’re old enough, reminiscent of sci-fi films like Cronenberg’s 1999 eXistenZ based on a Philip K. Dick novel.
Pinocchio is officially on his way to becoming a real boy.
Imagine gaming worlds populated by nonplayer characters (NPCs) that are interactive to such nimble and nuanced depths that they establish genuine emotional connections with players and can hold their weight in lengthy, windy, and clever conversations. Machine learning will fill the open spaces with dynamic content and narrative while still abiding to the game’s mechanics and script.
Game worlds are about to become super-intelligent, and its influence will be particularly glaring in the world of multiplayer games and competitive gaming.
Esports is already on a steady course to passing the billion dollar revenue mark by 2018, according to SuperData Research, and its positive momentum is owed considerably to it being positioned at the center of multiple industry intersections. Machine learning is yet another to add to the party and while it’s still the early days, there’s already plenty of traction.
“ESL and its associated brands, such as ESEA, ESL Play, and Badlion utilize machine learning to better connect with customers and fans by learning behavior trends and increasing our technical capabilities. At ESEA, for example, Machine Learning helps to increase the efficacy of the ESEA Client’s anti-cheat capabilities to detect cheaters and people abusing others in the community,” Ralf Reichert, CEO of ESL at Turtle Entertainment, told me.
Broadcasters like Twitch are actively using the technology at specific pain points like managing real-time moderation. Their AutoMod technology allows for more effective filtering of inappropriate or hateful content, affording broadcasters more control on chat activity. While tools like this have been around for some time now, the machine learning aspect allows for these platforms to get better at their job over time without the constant manual intervention of human-driven changes.
“For the first time ever, we’re empowering all of our creators to establish a reliable baseline for acceptable language and around the clock chat moderation,” Twitch moderation lead Ryan Kennedy told VentureBeat.
Gambling in esports is rising in parallel to the industry at large, which makes it unsurprising to see quite a bit of research going into developing predictive analysis techniques that allow for forecasting match outcomes.
One recent paper was published by Lund University, entitled Esports Analytics Through Encounter Detection, and presented last March at the MIT Sloan Sports Analytics Conference. It discussed how segmentation techniques that apply machine learning can be brought into play in tracking player and team behaviour in order to identify the patterns required to predict which team is most likely to win in a particular setting.
The authors describe the process as “using esports related data, primarily behavioural telemetry but also other sources, to find meaningful patterns and trends in said data, and the communication of these patterns using visualization techniques to assist with decision-making processes.”
Esports is awash with big data sets and initially the industry will look to leverage machine learning in order to digest the mined intelligence found in player historical game records. It captures the fodder, digests it, and ultimately transmutes it all into pattern recognitions that can drive intelligence in the form of constructive insights.
Berlin-based startup Dojo Madness leverages big data for exactly that purpose. They’ve rolled out a series of real-time coaching companion tools that help aspiring players master major esports titles like League of Legends, Dota 2, and Overwatch. They’ve also released an analytics platform, coined Shadow, that’s tailored to the needs of professional teams.
“Our Sumo Apps are serving the broad middle bracket of the player pyramid, about 80% of the playerbase. In Q1 of this year we are exceeding 1MM DAU across our SUMO Apps driven by League and Overwatch,” Jens Hilger, CEO at Dojo Madness, told me in an email.
Each companion app is integrated within the gaming environment, rendering itself into a sort of early predecessor to an R2-D2 that generates customized insights for each player based on their individual historical behavior.
And this is just the beginning.
Imagine a match between two teams where the maps are revised just prior to the start, based on pattern recognitions of the characteristics, preferences, and the tendencies of each individual team member on either team. Or subtle changes are made to static conditions during the actual gameplay in order to challenge and engage players and teams to play more consciously. Machine learning can intelligently manipulate the playing fields by adding in customized handicaps tailored for the specific teams of each match.
Now dive deeper and imagine a game that can do more than just look up your history but can read and respond to your real-time data and use it to anticipate your next move. The experience would be one of spontaneous creation of scenarios, characters, and world maps that are devised not by developers but by the game’s “mind” itself.
A whole host of features can be implemented and brought into play that will allow eSports to continually evolve, perhaps even by itself, in ways that transcends well beyond the confined limitations of traditional sports.
The real-time “telemetry” data, is the information you’re constantly receiving and sending back to the game. It’s what could offer an individualized experience that machine learning tailors to each player, broadcaster or even the audience pool itself.
Telemetry data allows for a spontaneous experience based on intelligent anticipation, referencing and augmented by historical big data, and may later lead to a type of co-creation experience that makes one think of the Mind Game played by the boy protagonist in the movie, Ender’s Game, but in multiplayer.
Machine learning is going to blow our minds, literally, and it will change the face of competitive gaming.
Amir-Esmaeil Bozorgzadeh is the co-founder at Virtuleap, a sandbox for creative developers to showcase their VR concepts to the world, which is currently running the world’s biggest WebVR Hackathon.
