Recent years have seen much speculation about both the potential and limitations of art generated by artificial intelligence. Technologists and novices alike look at the output of tools like Deep Dream and DALL-E, debate their aesthetic merits, and wonder what similar algorithmically based programs will be able to accomplish in the future. The new documentary The Computer Accent reiterates many of the talking points within these debates without adding much to them, but it does offer one helpful contribution to the subject: an extended, in-depth look at the actual process of creating art using AI.

In 2016 the pop-dance band YACHT — musicians Claire L. Evans and Jona Bechtolt — decided to use AI to help compose their next album. Directors Riel Roch-Decter and Sebastian Pardo and their crew followed the band for several years as they pursued this vision, learning about neural networks and strategizing how to devise a method to extract usable music from a machine. Ultimately, Evans, Bechtolt, and their technological and musical collaborators notate YACHT’s entire back catalogue in MIDI, which is then fed into MusicVAE, a tool by Magenta (Google’s open-source project researching machine learning). With the model “trained” on this info, it can then generate its own MIDI snippets based on YACHT’s preexisting music. The band take what they can from this output — riffs that intrigue them, passages that sound good — and then finesse and combine the material into various songs. The result is their 2019 album Chain Tripping. 

Still from The Computer Accent

Frustratingly, The Computer Accent, like much discourse around AI and art, operates from a flawed understanding of its own subject. Chief among such misunderstandings is the idea that any of these tools truly constitute “artificial intelligences.” Algorithm-based programs are nothing more than highly complex aggregators. While the art they can generate looks increasingly like art made by humans as the technology has been refined, this is not an intelligence getting “smarter” but a more sophisticated form of imitation. 

The general assumption is that feeding more information into these programs will help them better approximate the human mind, as if cognition can be quantified. But pattern recognition and association are elements of sentience, not its basis. On their own, these qualities don’t build to complex thought; they form biases and stereotypes (as evidenced by AI’s continued problem with perpetuating racism). Early in The Computer Accent, one expert cautions against anthropomorphizing these tools. Yet even if Evans and Bechtolt only refer to MusicVAE as a “collaborator” in jest, the film still posits man and machine in these terms.

Still from The Computer Accent

This is a shame because it elides what’s truly interesting about what YACHT does with Chain Tripping, which is find a novel technique for songwriting. The idea of “an album composed by AI” might inspire images of a computer spitting out the music, but the process entails a huge amount of human curation and remixing. Still, the results sound mostly like YACHT retreading familiar ground … because, of course, all the AI can do currently is remix what’s put into it. Still, YACHT’s experiment was one worth attempting, even if its results were mixed. As a document of that experiment, The Computer Accent helps illustrate the prospective uses of AI, along with how much, much further it has to go.

The Computer Accent is currently in theaters.

The Latest

Avatar photo

Dan Schindel

Dan Schindel is a freelance writer and copy editor living in Brooklyn, and a former associate editor at Hyperallergic. His portfolio and links are here.