Data Science in Video Games
Video games are an amazing platform to gain insights into what people are, how they think, how they act, and how the influence one another. The idea that developers control the very reality, in which people exist, for potentially large portions of their life, is just magic. But there often seems to be a lot of hesitation, resistance, and misconception in what it takes to understand, and really dig into the root of what data is, as a science in games.
Data science, while not always titled as such, has been in practice for decades, in various amalgamations. Naturally, more limited inputs of information called for somewhat more rudimentary methods, however with more and more points of data to pull from online by users on digital platforms, the ability to dig into that data has created a lot of explosive growth into what can be learned and tested against. But with games, it's a whole other world of possibility.
Right now, games have a tendency to look at what is happening from one update to another. And often, its somewhat two dimensional insights. Looking at drops in users, or balancing topics, or showing a graph or chart that shows players spending vs units bought, or difficulty in scenarios. But the reality of this, is those changes and general analysis is not what data science is, and this lack of process often leads to MORE updates, more assumptions, and even friction between product developers/designers, and data side groups.
And really, why wouldn't it? Designers have thrived and been trained to trust their creative intuition to make fun from their own experiences. Products are pitched by stating "its like this, but also this game, with these features" to leadership so it becomes palatable. I have seen maybe two groups that can articulate the core behavior of their products, without talking about how the product is like other products. And to be clear, this is ok, and to be expected. But there is an opportunity for actual sciences in data to flex, and get people excited and on board. Still, good data science is an all hands on deck topic.
Good data science is about seeing the story of changes in data. How worlds are created, when changes occur, the impact of those changes, and what insights can be gained or tested against for the future. But getting there with accuracy, takes experimentation of what data is being tested against, and being able to test that against variables for various outcomes. And with a good process in place, and access to the right data, with the proper insights, you can literally change the way people behave. Let that settle in.... Change the way people behave.
The process is about more than just what you see at the surface level from player behavior in a set amount of time, its about defining a hypothesis for testing against different points of data, and furthermore, various perspectives of science. Looking at what happens with data while pulling one set of data, and adding others, can result in DRAMATICALLY different outputs. And that is not even considering the choice of models being utilized to gauge that data in the first place.
The reality of all of this is, data science in games, no matter how big or "data driven" you are, is still in its infancy, and its a bad idea to assume that games are anywhere near understanding what sort of impact data science has on games. Tools in the field are often not built to scale, the implications of big data and iterative small file analysis are HUGELY segmented. Deployment of changes into controlled environments is often not a thing. And frequency of data to gauge at runtime is pretty much not a factor with commercial tools, which is why we find ourselves plugging in and hacking data bases, visualization, and analytical tools to find outputs.
It's all very mystifying, and even exciting to get visual ideas of what data does, it's a great start, but even visualization of information is ONLY an interpretation and abstraction of data. From a mathematical and even psychological standpoint, there is really no good way of visualizing data in practice of data science, outside of surface level reporting. The devil is in the details, and analysis paralysis is not typically a topic of concern, so long as you have the capacity to dig deep, and understandably with production timelines, and buy in, that's a tough thing to accommodate.
This whole field applied to games is going to take baby steps, and a lot of foot work from leadership to really start understanding what data science is, how it can be used, and how it can impact their products and teams. Just the same, its always exciting to see more and more studio start to show interests, and im anxious to see what the emergence of AI and Machine Learning can bring to an industry that I have loved since I was a child.