Steam's announcement of Steam Labs today came with some interesting tools that may impact the future of selling games on Steam. Key among them is the interactive recommender---a new tool for game discoverability that, per Valve, uses a user's "hours played" metric and some machine learning to make game recommendations based on user inputs.
After some concerns over the performance of the 2019 Steam summer sale and the growth of the Epic Game Store as a curated storefront, developers have plenty of reason to pay attention to what the future of Steam recommendation might look like. Fortunately, several developers have already weighed in with positive---if cautious, impressions.
Where the Water Tastes Like Wine developer Johnnemann Nordhagen succinctly encapsulated the range of responses to this new feature in a pair of tweets first expressing frustration over "hours played" as a metric for game recommendation, but then noticing the tool didn't seem to be blindly suggesting titles based on that number alone.
Steam's new "machine learning" recommendation engine appears to be based on playtime, a metric which will get a lot of things wrong for a lot of people, and is highly biased against single-player narrative games— Johnnemann ðŸŒ¹ (@johnnemann) July 11, 2019
I played around with it and it's way better than I feared! It uses more than playtime, it seems, even though that's what is front-loaded in their explanation https://t.co/876rxpLmtD— Johnnemann ðŸŒ¹ (@johnnemann) July 11, 2019
Veteran indie devs like David Pittman and Robert Zubek quickly weighed in to say the tool already helped them surface some interesting games they hadn't heard of, with Zubek seeking more questions about the game's machine learning, and what inputs developers should be paying attention to when their games surface in the tool.
Steam's new interactive recommender seems cool, I've already discovered a few newish games I'd never heard of before! https://t.co/7SwRKzKXpw— David Lindsey Pittman (@dphrygian) July 11, 2019
I'm loving these interactive recos, they seem to be right on! :)@dejobaan could you go into more ML details? Blog post says it's using only play histories, not tags, scores, etc. Does this mean it's treating users as points in N-dimensional space of "playtimes per game"... 1/2 https://t.co/iFfy8Y7Bpv— Robert Zubek (@rzubek) July 11, 2019
NYU game center chair Naomi Clark joked that the tool seemed to be measuring the 500 hours she'd mistakenly left The Wolf Among Us running in another window, but ultimately said the recommended games were still a good selection, and that the slider between "popular" and "niche" was a particularly good part of the tool.
Jokes aside, this does actually seem to work pretty well for preferences, since in various configuration & genres it recommended a lot of games that I played and enjoyed, including some shown. I imagine it has lower bounds, but niche slider is great.— Naomi Clark [æš—æ‚ªãƒ»ç›´ç¾Ž] (@metasynthie) July 11, 2019
Some developers, however, remained pessimistic about Steam's use of algorithms to drive game recommendations at all. Observation programmer Andrew Haining described this behavior as Valve having "no interest in listening to developers or players."
It's just another reminder that valve have no interest in listening to developers or players, they'll keep shoving algorithms down our throats whether we like it or not.— Andrew Haining â˜â˜¸ï¸â˜®ï¸ (@1stCLord) July 11, 2019
Many developers selling games on Steam right now are intimately familiar with how well front-page placement on the store, no matter how it's handled, can be to their games' sales. At this time, Valve has stated it would make a blog post if there were plans to bring the recommender to the Steam front page.