Guest post from Tom Matcham, mathematician and founder of coAdjoint, who shares his recipe for using personalisation to make better educational games.
The elephant in the Games Industry room
For many years now, there has been an elephant in the Games Industry room, and he’s sat right in the middle of Educational Games Corner. The elephant has witnessed many educational games being designed and developed, most of them following the ‘tried-and-true’ linear formula, presenting every user with identical content independent of their performance in previous tasks. Occasionally, the elephant raises his trunk, waggles it around to get everyone’s attention and says: ‘Linear, static games really aren’t that great for learning, perhaps someone should try something different?’ The developers would usually grumble excuses: ‘sounds hard'; ‘we’d waste time making content only a few people would see'; ‘gamers are dumb, they wouldn’t understand’. ‘Oh,’ the elephant would say, and all the developers would go back to making their normal, impersonal games.
In my opinion, the elephant makes a very good point. As educational games designers, we are very good at producing fantastic user interfaces, exciting content and engaging game mechanics, yet we are very poor at accommodating the individuality of the people playing our games. I believe that far too many educational games are made without the personalisation of the learning experience in mind, and that we could all make better games if personalisation was at the heart of our game design considerations.
But how do we make a personalised educational game?
The most natural question to ask next is how do you make a personalised educational game? I believe the most suitable approach for educational game design is to mathematically model the skills of the user. By using the actions of the user as input to the skill model, we can understand the overall ability of the user and perhaps more importantly their strengths and weaknesses by using machine learning to classify the player’s skill-set. The output of the machine learning algorithm can then be used to pick the most appropriate content for the user with respect to their current skill-set. For example, if the user is weak in a particular area, you may want to provide the user with content that will help them improve in that area. Once your model informs you that their skill levels have sufficiently increased, you can provide more general content until a weakness is identified again. The advantage is clear: the user benefits from a personalised experience tailored to their needs.
Example of a mathematical model of arithmetic learning in the Unity game engine. Each box represents a skill relevant to arithmetic learning, such as the multiplication of large numbers. Each skill is modelled using a sigmoid function, as shown in the Skill Preview window. The parameters Q, B and v determine the profile of the sigmoid. Connections between skills model the fact that the skills are related. The directed value of the connection determines the strength of the connection. For example, the model assumes that ability in multiplying small numbers will significantly aid the learning of adding small numbers, whilst ability in addition and subtraction are only considered to be weakly related.
Clearly, the design of a game using a system like the one described above would be a very different process from the design of more traditional, linear game. Firstly, since content would have to be available at any stage in the game to accommodate for the skill-set of the player, levels and narrative would have to be built in a modular, non-linear fashion. Secondly, the designer has to spend time deciding what skills are relevant to the game, and then thirdly must create a mathematical system to model player skills at runtime.
The right tools for the job
Fortunately, tools already exist to help the developer implement these tasks. For example, Twine, a non-linear story creator, can greatly simplify the process of modular narrative design and furthermore has a easily parsed file type, making integration with other platforms relatively easy. Mathematical tools such as machine learning algorithms are widely available and, importantly, easy to use without an understanding of the underlying code and maths.
Example of a machine learning algorithm being trained to analyse user skill-sets in the Unity game engine. The designer is asked to classify a possible skill-set for a user that is generated by the software. In this machine learning problem, the designer is classifying based on weak areas of the user. The user simulated above is particularly weak at addition, and so would be classified as a weak adder. This result could be used in game to provide the user with a higher ratio of simple addition questions until his addition skill levels catch up with the rest of his skills.
Personalisation is an extremely powerful tool that’s also extremely under-utilised. If you’re still unconvinced, take a couple of examples from the ‘normal’ games community: The Sims, Minecraft and The Grand Theft Auto games are all personalised games and also sold incredibly well. The method of personalisation may be different to the one described above, but the reason for the excellent player engagement and sales is the same: these games make the player’s actions an intrinsic component of the story, if not the entire story. I believe that achieving the educational gaming equivalent, to make the player an intrinsic part of the learning process, could lead to some incredible results in educational gaming, and potentially change the way we learn forever.
Tom Matcham is a mathematician and founder of coAdjoint, a middleware company helping developers to personalise the content of their video games and educational software. He can be contacted through [email protected].
Comments are closed.