When the game plays you: DotaPlus and surveillance capitalism

Software reliant on machine-learning algorithms has become a central character in our everyday lives. In simple terms, machine learning can be understood as a form of artificial intelligence that quasi-autonomously improves performance of a specific task based on the accumulation and parsing of data. For example, Instagram and Facebook ‘learn’ to curate what content its users are most likely to engage with using techniques based on prediction and probability.

A common term used to describe these kinds of services is ‘platform’. Broadly speaking, a platform can be understood a technical infrastructure that facilitates a range of human activities (buying cat food, playing videogames, posting images and so on). Platforms leverage an economic logic operating within a capitalist economy, as described by the scholar of political economy Nick Srnicek in his 2017 book Platform Capitalism. According to Srnicek, platforms link together different markets through rapid outward expansion, deriving significant economic value in doing so. We might think here of Amazon’s foray into smart homes with assistant tools like Alexa and their attempts to forge relationships with home builders.

In seeding these infrastructural roots, platforms harvest data which is used for further development and optimisation of the platform itself. Accumulating massive amounts of data is key to the operation and economic viability of platforms, especially those that use machine learning. My participation in a Facebook poll, my Instagram followers’ likes, and my Google browsing history of Red Dead Redemption 2 topics have all become tradeable commodities in what Shoshanna Zuboff calls the ‘behavioural futures’ marketplace, where they are sold off for use in systems of behavioural prediction – that is, systems that seek to understand our behaviour and likely desires, with the goal of shaping our consumption.

Platforms, in the current moment, are fecund sites of what Zuboff calls ‘surveillance capitalism’. That is, they are technologies and techniques of data collection that work to shape human behaviour on a large scale, with the end goal of deriving economic value and seizing market control.

Recently, machine-learning based platforms reliant on the acquisition of user data have taken hold in videogames. In March 2018, the US videogame developer Valve launched DotaPlus, a US$4/month subscription service for their game Dota 2. In the tradition of ‘battle passes’ – a monetisation model popular in free-to-play games such as Fortnite and Rocket LeagueDotaPlus offers users a range of analytics tools premised on players improving their skills at the game. The objective of Valve’s surveillance differs, however, from Zuboff’s formulation. Where Zuboff is focused largely on behavioural manipulation through predicting our actions and desires, DotaPlus uses prediction to minimise in-game ‘risk’ (that is, the risk of losing matches) and thus – potentially – a negative experience of the game. (For a discussion of the minimisation of negative emotions in gaming and attempts by developers to hit an emotional ‘sweet spot’, see Jame Ash’s The Interface Envelope.).

DotaPlus was launched at a time when Dota 2’s popularity seemed to have peaked: more accessible games like League of Legends had amassed larger player bases and newer genres and videogame formats – think Fortnite – had exploded in popularity and begun to dominate the market of online, free-to-play multiplayer games.

Valve’s move is an interesting one. As writers like Patrick Lemieux and Stephanie Boluk have argued, the company has historically focused far less on creating original games and more on acquiring and revamping existing ones (Counter Strike, Team Fortress, Portal, Dota) and tying them to Steam, its digital distribution platform. As Will Partin suggests, Valve has an abiding concern with creating games that make users amenable to using – and ideally spending money through – the Steam platform.

What makes DotaPlus interesting, then, isn’t that it represents a new stage in the ongoing platformisation of videogames, but rather – I would suggest – that it reflects an economic and infrastructural model reliant on the quasi-autonomous capture, sorting, and presenting of user data. This model has transformative implications for the players’ experience of gaming.

The key feature of DotaPlus is driving consumption of the game by helping players improve at Dota 2. Its operation is akin to a virtual coach. Players subscribe for platform features that sort, track and arrange data ostensibly designed to help them improve and that and enabled in the first place by Valve’s surveillance of user activity.

The Plus Assistant, for instance, presents users with real-time suggestions on how to perform in-game – namely, about which user-controlled characters (‘heroes’) or skill points it is best to select. As Valve puts it in its advertising:

Use the Plus Assistant to help you make build decisions by utilizing real-time item and ability suggestions. These suggestions are based on data gathered from millions of recent games at each skill bracket, ensuring your builds stay current in the ever-evolving meta.

Being good at videogames is not a skill that simply occurs naturally. Our actions and cognitive responses are trained through ongoing engagement with hardware and software. When players become competent, they draw on their memories of past play – often in a way that is very fluid or fast – to respond to situations, and act in an anticipation of what’s likely to unfold in the future.

The Plus Assistant changes this dynamic by offering statistically determined prompts for what players should do in certain contexts. Take the choice of hero in Dota 2. Selecting the right hero is very important, especially at higher skill levels, as different heroes have unique strengths and weaknesses against other characters, and a considered hero choice can mean the difference between winning and losing. In one match, say, I’m playing against the Huskar hero, whose main feature is that he deals damage inversely proportional to his health points. The game suggests I pick the Ancient Apparition hero, who has the ability to kill enemy players if they have low health. I no longer need the accumulated knowledge of hundreds or thousands of hours to know that I ought to make a particular choice in a particular situation. The new interface augments and transforms my actions along with my experience of each moment of gameplay, reducing the need for deliberation and increasing my chances of success – which in turn might influence my decision whether to continuing playing or not.

Beyond how it feels to play a particular game, there are implications when it comes to thinking about the cultural politics of gaming – challenging but also reinforcing many of the enduring and normative ways we think about videogames and gaming.

For instance, DotaPlus potentially makes Dota 2 –a difficult and time-consuming game – more accessible, particularly to those without sufficient economic or social capital to invest the many hours necessary in order to master it. This is key in challenging some of the normative and deeply exclusionary values associated with gaming and what it means to be a ‘true gamer’. Of course, it would be naïve to think that Valve’s motives are altruistic, but altering this dynamic is important, especially within the context of the often toxic and meritocratic cultures of multiplayer games – where often the prevailing attitude is ‘get good’ or get out.

Not all reactions to DotaPlus have been positive. A look at the game’s sub-Reddit suggests that many players see its dynamic and statistically-driven coaching tool as tilting the playing field in favour of those who can afford it – what some have referred to as ‘pay to win’.


The platformisation of games is nothing new in and of itself, nor is the creation of economic value through the surveillance of user activity. What is novel here is the creation of data-driven, statistical interfaces, where users negotiate the game through the prism of their own actions and perceptions, but also through pay-for-use data interfaces made possible in the first place by Valve’s surveillant collection of user data.

Much like other ‘assistant’ tools of the age of the Internet of Things, DotaPlus collects and analyses massive amount of data to not only predict but effectively shape future experience. However, while DotaPlus traffics in user data, it doesn’t quite play the same game as platforms like Amazon, which have a stake in the ‘behavioural futures’ markets. Instead, it is designed to encourage a dwindling player-base to continue playing, purchase microtransactions and – perhaps most importantly – keep people on Steam. More broadly, therefore, DotaPlus represents an example of how more and more aspects of everyday life have become sites for the ever more efficient extraction of profit.


Image: ‘Hand of Midas’ patch for Dota 2 

Ben Egliston

Ben Egliston is a PhD candidate and sessional lecturer at the University of Sydney in the Department of Media and Communications. He researches and teaches about games and new media. His current research is focused on user experience in games. He has published on topics such as e-sports, data analytics, and livestreaming for a range of academic and general interest publications.

More by Ben Egliston ›

Overland is a not-for-profit magazine with a proud history of supporting writers, and publishing ideas and voices often excluded from other places.

If you like this piece, or support Overland’s work in general, please subscribe or donate.

Related articles & Essays