Vote 2020 graphic
Everything you need to know about and expect during
the most important election of our lifetimes

Steam's New Customizable AI Tool Scans Your Playtime And Tells You What To Buy Next

Illustration for article titled Steams New Customizable AI Tool Scans Your Playtime And Tells You What To Buy Next
SteamedSteamedSteamed is dedicated to all things in and around Valve’s PC gaming service.

Just days after the end of a summer sale that laid bare many of Steam’s issues with helping people discover new games, Valve has announced a new section dedicated to “experiments around discoverability, video, machine learning, and more.” It’s called Steam Labs, and the first bits of mad video game science to emerge from it are promising.

Advertisement

For now, Steam Labs has three active experiments: micro trailers, an interactive recommender, and an automated, daily half-hour show about games. Micro trailers are pretty straightforward: Steam presents you with selections of six-second trailers organized by genre, curator selections, or other categories. If you’re intrigued by what you see of a particular game, you click on it to visit its store page. It seems to be inspired by long-running Twitter account “Steam trailers in 6s,” a proven and useful game-discovery tool.

The automated show is assembled from similarly brief clips of games, but with multiple micro trailers for each game assembled in a quad display set to music. I tried watching the first episode, but I got bored a couple minutes in, given that it was basically the micro trailers feature except I wasn’t in control. The goal, originally, was for Steam to automatically generate voice-over descriptions from games’ store pages, too, but then Valve remembered that robots are weird.

Advertisement

“In text-to-speech tests, THE. COMPUTER. GENERATED. VOICE. WE. USED. SOUNDED. A. LITTLE. STILTED, so we tabled that for a bit,” the company wrote on the new feature’s Steam page. “We’re working on that, though.”

The main event of this impromptu lab tour is easily the interactive recommender. It’s a “neural-network model that is trained to recommend games based on a user’s playtime history, along with other salient data” based on “many millions of Steam users and many billions of play sessions.” But it’s not just an automated list of games; while it initially crunches numbers by chewing on your most-played games, you’re also able to tell it what kinds of games you want to see. The majority of my most-played games, for example, are RPGs like Divinity: Original Sin 2, The Witcher 3, and Fallout: New Vegas, but I was able to adjust sliders so that I could see progressively more or less niche and newer or older games. With the sliders alone, I was able to configure it so that I got a selection of intriguing games across multiple genres, many of which I didn’t know (and only a few of which, like YIIK, I knew were definitely bad). In addition, you can restrict games by tags.

Initial response to these new tools from both users and developers seems positive—a far cry from the torrents of doom, gloom, and confusion that have followed recent Steam announcements. Valve notes, however, that these tools are works in progress, and some might never escape the lab’s blue-hued blackness to see the light of day.

“Some of them may turn out great,” Valve wrote on the Steam Labs landing page. “Others, we may toss out. We hope that most will be improved with your feedback and go on to be a part of Steam. This is the way of Steam Labs.”

Kotaku senior reporter. Beats: Twitch, streaming, PC gaming. Writing a book about streamers tentatively titled "STREAMERS" to be published by Atria/Simon & Schuster in the future.

Share This Story

Get our newsletter

DISCUSSION

Ravenous Sophovore

I always wonder how much the algorithm for these sort of “you may also like” features are influenced by outside factors like, say, how much money the owner of the product is paying for their product to get featured more often.

Amazon’s “based on your browsing history” seems full of things that don’t seem to have a lot to do with my tastes but do seem to be big sellers. Not sure if, statistically, people who bought what I bought are also likely to have bought some popular thing or if some popular thing got (or is staying) popular because they bought their way into the recommends of many people.

Maybe I’m just being paranoid.