Rensselaer Polytechnic Institute’s Department of Cognitive Science is trying to find ways in which we can get better at video games. The Lead Researcher, PhD student John Lindstedt (under Principle Investigator Wayne Gray), is using an in-house coded version of Tetris to study the learning and cognitive training process within the context of developing teaching method that would increase a person’s skill in the games they play.
The research event that I participated in at Genericon XXVIII was disguised as a ‘Tetris tournament’. Players registered to compete for the highest score, in which the top three players earned cash prizes. It initially sounded intimidating, but the overly welcoming nature of the research students made it anything but.
“We used the tournament as a way to attract skilled players while, in reality, we want to gather players of all skill levels,” Lindstedt says. “That’s why we try to make the atmosphere generally relaxed and as fun as we can, so that when we say ‘Hey! It’s free, and it’s for science!’ We get people who think, ‘Ah, what the heck’, rather than ‘Oh, no, I couldn’t, I’m terrible, I’ll mess up your study.’”
After taking part, I can attest to the fact that they succeeded. Play sessions were individualised, with each participant paired with a desktop and a technician. All were given two playthroughs to compensate for any bad runs. But, given that the nature of the term ‘tournament’ and the opportunity to win cash prizes would almost inevitably attract more skilled players, the research team also holds sessions throughout the school year to observe enrolled students as well. All of this has gone towards gathering data over the past three years.
CogWorks Lab’s coded version of Tetris, otherwise known as ‘Meta-T’, is specifically designed to gather that data. Every key stroke, every block rotation, and in some cases, even eye movement (not studied in the tournament) is tracked and compiled into data with millisecond accuracy. Meta-T provides various experimental options, like narrowing the field of vision using eye tracking, and altering the Tetrominoes’ fall rate completely. Meta-T also allows for comparative data collection between skilled and novice players – helpful for identifying areas where less skilled players can improve.
One of the most fascinating aspects of Meta-T, however, is its support for computational cognitive modelling (or to break down those eleven syllables, CCM). Lindstedt explained this as an artificial intelligence that’s “still written in code, but its purpose is to make the computer behave in some of the same ways as humans”.
Computational cognitive modelling scales back from the computational processing that typical AIs use to come to a solution. Instead, coming up with an answer instantaneously, like humans, the process that a CCM takes in reaching a solution occurs in stages.
Here, Lindstedt highlights the difference between AIs and CCMs:
“An AI can churn through a billion numbers per second, simulating 50 moves ahead, all within a millisecond of the block appearing on the screen. A cognitive model does things differently. It has to shift its attention to the block (a couple hundred milliseconds), it has to focus on it for a moment to comprehend it (a few hundred more), it has to then look around to see how that block could fit into the existing pile (a few more), then it has to press its first key (150 milliseconds minimum), all still happening in under a second; it’s a snail’s pace compared to its AI cousin.”
The research purposes of developing such models is to allow researchers to tangibly study the human thought process. This is done by altering the CCMs parts to test various results. It would be impossible to do so any other way outside of some sort of psychic puppetry one could imagine in science fiction.
Interestingly enough, however, the team has also incorporated a system of AIs that sound similar to these CCMs. For the purpose of highlighting areas of improvement in player skill, the team developed AI systems operating as automated coaches. While they aren’t able to manoeuvre Tetris at the same level as highly skilled players, they certainly can teach less experienced players a thing or two.
“So, the AI systems we’re developing are actually being used right now in a study examining different automated methods of coaching (i.e., using suggestions from the AI to give guidance to players),” Lindstedt explains. “The systems are super simple (just using simple addition and multiplication to rate “good” vs “bad” options in the game), so no Skynet to be found here!”
But in efforts to further develop a proper framework used to coach and increase player skill, the subject’s cognitive capabilities need to be taken into consideration. When I asked if the research team has considered the classic trifecta of learning methods: visual, auditory, and tactile, Lindstedt said something that I wasn’t expecting:
“Those three individualized methods of learning have since been discredited over the years.”
“Really?” I ask.
“There really isn’t such a thing as learners who strictly learn through one or the other. It would be inefficient for a teacher to develop three separate ways to teach a classroom full of students.” He continues, “…if I can find an effective way to teach something visually, everyone in the classroom should benefit from that. You can’t really put anyone in a box.”
Lindstedt pointed me to this Wired article which discusses how the traditional learning styles is largely a myth.
Of course, while Lindstedt states that such discrimination in learning types doesn’t necessarily exist in the categorized manner we were all once taught, he is aware of the various factors that can influence one’s ability to learn:
“I guess a limitation of our experiment is not factoring in various physiological, psychological, and biological factors that can impact player-measured performance in our research. The real question, however, is that have those differences predisposed them to playing video games, or have playing video games helped them develop these differences and skill sets?”
Essentially, he’s asking which came first: the chicken or the egg. Over the years, to shake up the conventional ‘wisdom’ about video games having a violent impact on children, there have been several research studies and experiments correlating playing video games with other activities such as eye tracking and rapid decision making. Many of these have analysed the direct changes in efficiency right after playing games. But one might ask whether or not the physiological make-up of some participants, or the chance that some might be closeted gamers, had an impact on the results of each study.
With such an ambitious undertaking, there is the inevitable question everyone wants answered: “Can you train me to perform at E-Sports level?”
“It’s an impossible question to answer at this point,” Lindstedt says.
“When folks talk about getting better at games, the idea is to train, and train, and train. But for some people, that’s not enough. Folks may eventually hit a plateau, but we want to move past that. We want to meet them at their perceived limit and then find ways to grind them even further,” he adds.
We’ve all experienced the stark contrast between competing against ourselves, friends, within global leaderboards, and in online matches. It’s like there’s a dense cognitive barricade that’s preventing us to do what many skilled players do so easily.
When I hear that researchers are actively trying to develop methods to improve player performance, especially in a market where couch and online gaming are becoming more commonplace, it sparks excitement.
“Video games are an excellent domain for this sort of research, especially with respect to the immense cognitive skills these players often display…There’s near limitless enthusiasm for Tetris, both from its players and from fellow researchers hearing about its relevance to cognitive science for the first time— a researcher couldn’t ask for better.”