Cory Arcangel

Cory Arcangel on bot performance, machine learning, and online junk space

Cory Arcangel, /roʊˈdeɪoʊ/ Let’s Play: HOLLYWOOD, 2017–2021, custom-built high-performance computer rig, custom-built Deep-Q Learning RPG playing software bot with system sounds by Daniel Lopatin (Oneohtrix Point Never), Kim Kardashian: Hollywood Android game, Android phone, amplifier and speakers, various cables, dimensions variable.

Cory Arcangel’s latest exhibition, “Century 21,” continues his interest in the structural aesthetics of games, exhaustive virtual navigation, and sometimes punishing durations. Its centerpiece is a custom-built machine-learning computer that plays the mobile role-playing game Kim Kardashian: Hollywood. This work is exhibited along with an array of live and recorded bot performances, prints, and paintings the artist refers to as “flatware”: high-resolution scans of pants that are printed on IKEA tabletops. Here, Arcangel discusses the four-year process of bringing this exhibition—on view online and at New York’s Greene Naftali gallery through April 17—to life and how his work has changed over the past twenty years.

THE EXHIBITION CONTAINS two live bots and two recordings of bot performances. Related to Your Interests, 2020–, is a YouTube bot that scrapes super low-end trash clickbait websites twice a day and mashes texts, images, and articles together into videos where the text is read by an artificial voice and puts it on YouTube. It’s been running since last spring and by now there are hundreds of hours of video. They range from bizarre and short to really long and totally unwatchable. elleusa, equinor, equinox, etrade_financial, 2020, is a recording of an online performance where a bot has been programmed to go on the Twitter profiles of each of the companies in the title, scroll down, and like every post until the Twitter feed stops refreshing. we deliver / the king checked by the queen, 2020, records two bots playing a game of chess by tagging each other in comments on various corporate and celeb Instagram posts. 

Then there’s /roʊˈdeɪoʊ/ Let’s Play: HOLLYWOOD, 2017–21. The project originated from a note I made for myself on March 1, 2016 on my private channel for ideas: “Deep Blue playing Kim K game.” In my mind, I imagined a huge server that you could see playing the game. Working with Henry Van Dusen and Kevin Roark, who built the hardware and software, we realized we would have to make our own supercomputer. So we built a machine-learning software which runs on custom hardware—/roʊˈdeɪoʊ/—that looks at open-world role-playing games—in this case, Kim Kardashian: Hollywood—and tries to make sense of what it’s seeing. I’ll try not to get too into the weeds, but the short story is /roʊˈdeɪoʊ/ has a front-end AI that identifies what it sees—for example, a person or a plant or a door—and that information is evaluated in a kind of machine-learning black box, and then decisions are made by clicking and swiping. So /roʊˈdeɪoʊ/ basically clicks and learns, clicks and learns, clicks and learns. Nothing has been done with the code of Kim Kardashian: Hollywood. In the gallery you see /roʊˈdeɪoʊ/ playing the game on an off-the-shelf Android phone. 

Since it learns visually, the software could work on any RPG game. Because Kardashian is so open-ended and vague, we needed something other than the money and stars /roʊˈdeɪoʊ/ <em>has accumulated to keep it going. So </em>/roʊˈdeɪoʊ/ is also programmed to try to get to new places it hasn’t gone before. It looks at the space it’s in—the objects in the room, color temperature, etc.—to determine whether or not it’s in a new place. Sometimes, an ad pops up and bounces the game to YouTube, and then it can get loose on the net. We left that functionality in because it’s even more dystopian than watching it play the game.

Cory Arcangel, we deliver / the king checked by the queen, 2020, dual-channel screen recording of a live bot performance on Instagram, February 26, 2020, dimensions variable.

So much of how we navigate the internet now is branding and advertising. For years I’ve been working on a series I call “flatware,” which are scanned pants that are printed on IKEA tabletops. I installed a large one called 3-(blk), 2020, across from /roʊˈdeɪoʊ/. It’s pretty much all black except for three recognizable Adidas stripes. I see celebrity, fast fashion, branding, and supply chains as connected and part of internet/IRL junk space today. How far away are those three Adidas stripes in any given Instagram feed, or for that matter, just outside our doors? With fast fashion, like corporate online space, there’s no archive: It exists for a second and then the whole store changes. A lot of this exhibition is about holding onto experiences—virtual and IRL—that aren’t meant to exist for a long time. 

A lot of my earlier projects came out of TV. It was my dream to have work shown on public-access cable when I was a teenager and in college, and that interest in television continued in the early 2000s. Things like Totally Fucked, 2003, or Super Mario Clouds, 2002, are really dry narratives, something that you could see as experimental TV. In a way, so were my earlier self-playing games. Eventually, Web 2.0 happened and scrolling became a way of life. In 2014, I made the “Surfs” series, where I navigated corporate websites for the brands Subway, Office Max, Starbucks, and Dunkin Donuts. Those also had to do with my interest in capturing online spaces that so often just vanish. 

With /roʊˈdeɪoʊ/ Let’s Play: HOLLYWOOD, I’m no longer the author of the narrative. Neither me nor the programmers know what it’s going to do or how it’s going to go. In that sense it’s like having an animal in a work or a performance in the gallery. It’s not playing a cached version of the game or anything like that. If the game needs an update, that has to happen in the exhibition. Likewise, if the game goes offline or gets pulled from circulation, the whole thing will stop. It’s really as live as it can possibly be. I’ve been interested in ultra-durational things for as long as I’ve been making art, so it’s appropriate that I don’t necessarily know how long this will last.