The people working on Steam (and more generally, the team at Valve) seem to be on a transparency kick at the moment. There are multiple blog entries which try to open up particularly opaque bits of the company so users can understand what’s going on. Presumably there’s an element of using that understanding to defuse criticism when it comes to subjects like the terrible reputation of Steam’s support system and whether it’s warranted if you add in some more data. The most recent entry zeroes in on how the store itself works and aims to share the “thinking” the system has used to come up with game recommendations:
Robin Walker notes “We want to show you more of what it’s doing and why – and we have some features planned to help with this, starting with one we’re launching today: an algorithm section on game pages that states why the Store thinks this game will (or will not) be interesting to you.”
Recommended games already had a snippet which said that I was getting a game in this feed because it shared tags with other games I’d played. But now when you click through to those games it goes into a bit more detail.
I’ve run through the first few in my list to see what its current hit rate is like and the results are as follows:
In Prey, you awaken aboard Talos I, a space station orbiting the moon in the year 2032. You are the key subject of an experiment meant to alter humanity forever – but things have gone terribly wrong. The space station has been overrun by hostile aliens and you are now being hunted.
Steam picked this because it’s similar to Bioshock 2 and No Man’s Sky, the user reviews are generally Very Positive, it’s in the top sellers, 21 friends own it and 5 friends want it.
I have 0.1 hours in Bioshock 2 and it was because I was taking a couple of screenshots for an apartment feature in multiplayer mode not in the main game. Conversely I have 67 hours in No Man’s Sky. The system doesn’t seem to take into account those relative weightings or that I have no interest in other games on the BioShock side of the equation.
In terms of the user reviews and top seller stuff it makes sense that a game getting positive coverage and which is proving to be popular across a wider audience might override those concerns even if it was registering them in some way. From a commercial point of view it makes sense that Steam would show me this game because I might buy it to be part of the conversation. It could have shown me a smaller title – a walking sim of some kind instead – and I wonder how it makes the decision not to. I might be an anomaly here but I’d be more likely to buy the more obscure game. Prey chat is at saturation point all around me and I actively seek alternatives.
On the friend front it’s a lot of game journos who own the game, so colleagues as well as my regular gaming buddies. I have no idea whether there is any way to factor in things like press keys – I assume not – so that’s a potentially distorting factor both for me and for anyone who is friends with people working in the industry.
The friends who want Prey are a mixed bag in terms of how their interests align with mine. Mostly I know them through Dota and Counter-Strike so they’re people I encounter in a very specific vein of competitive multiplayer.
Game about the difficult work of people that manage emergency lines and services. Your task is to answer incoming calls and to react properly – give first aid instructions, advise, dispatch correct number of firemen / police / ambulances, or sometimes – just ignore the call. Play on ANY CITY in the world!
I’d never heard of this one but I’m getting it as the second recommendation because it’s similar to Youtubers Life and Planet Coaster, the user reviews come out as Very Positive and 7 friends own it.
Real talk: I played Youtubers Life for work for an hour and a half and it was rubs, and I spent half an hour on Planet Coaster before the lack of a tutorial meant I gave up. Three of the seven friends who own the game dont’ actually seem to have booted it up because when I click for their names it says that they have it in their library but only four have actually played. There’s no-one in that list who jumps out as being a kindred spirit.
It’s actually the game description which appeals but it’s because it puts me in mind of Cook Serve Delicious but with ambulances rather than any of the reasoning shown by the machine.
Outlast 2 introduces you to Sullivan Knoth and his followers, who left our wicked world behind to give birth to Temple Gate, a town, deep in the wilderness and hidden from civilization. Knoth and his flock are preparing for the tribulations of the end of times and you’re right in the thick of it.
This one is because I accidentally left Five Nights At Freddy’s running after I thought I’d quit and had left the room and because I have a smidge of time with Narcosis – a deep sea horror game that I was playing as a spider crab befriending sim. It is “currently popular” and 4 friends own it. Two are work colleagues and two have never booted it up.
User tags include “survival horror”, “horror”, “psychological horror”, “violent”, “dark” and “parkour”. I actually clicked to see the list of tags and it invited me to submit some of my own starting with the ones I’ve used on other products. Those are “origami”, “papercraft” and “paper” which should give you an inkling of how far off the mark Outlast 2 actually was as a recommendation.
Something worth noting here is that they system requires users to engage in the feedback loop and tell the system where it’s going wrong so that either the team behind the algorithms or the algorithms themselves can adjest accordingly.
I tend to get stuck in uncanny valleys of recommendation engines because I tend to give systems the bare minimum of data needed to use them. I need Steam for work and to play Dota so I have to agree to it tracking some elements of my behaviour and I’m human so sometimes I’ll follow a link from something it’s tweaked and either buy or not buy. But because I don’t like feeding in extra data I don’t tend to use the discovery queue and if I do browse it it’s more that curiosity about how close to the mark it is rather than because I’m looking to buy or play something.
For me the promise of less friction on a shopping/entertainment platform doesn’t outweigh the discomfort I have about sharing more data than I absolutely have to. Obviously I still participate by having a Steam account and using areas of the service so it’s about the degree to which I offer this stuff up willingly rather than anything else. Your personal boundaries will vary.
The other interesting thing here is just about the phrasing. Walker’s blog post positions the Steam store as an entity:
“The Store is constantly trying to balance all the different interested groups of players and developers. It knows that it has a limited number of spaces it can use to show games to a player. It has some knowledge of the player, if the player is logged in and has a purchase / play history. It has some knowledge of the game, based on what the developer has told it and what previous purchasers of the game have said & done. It chews on all that data, and finally, decides which games it should show the player in all the various sections of the Store.”
I assume it’s shorthand for referring to the set of algorithms and types of intelligence Valve are deploying to manage the storefront’s variable behaviour but I enjoy seeing how these things gradually take on a fledgling identity through our communication with them and as we scrabble for ways to quickly explain what they’re doing.
Valve are working on another two big explainy blog posts. Next they’ll gab about “some ways the Store is being exploited”, which sounds like it might dig into changes to gifting and keys and whatnot. The third will get into Steam Direct, the upcoming scheme to replace Greenlight with more-open publishing, and talk about the fee Valve will charge for that.