Artificial intelligence has an Achilles’ heel. It can’t decide what’s relevant.
It just so happens that this is a crucial skill where creativity is concerned. Take computer-generated art. Such work has been well received in many prominent settings over the past few years—Ernest Edmonds’s wonderfully colored interactive pieces (shown alongside Mark Rothko’s canvases) in the 2007 “ColorField Remix” exhibition in Washington D.C., for example, and Richard Brown’s Mimetic Starfish, commissioned for the opening of London’s Millennium Dome, which the Times of London described as “the best thing in the Dome.”
But those artworks didn’t depend on a subtle appreciation of relevance. The Edmonds work is abstract: vertical stripes of ever-changing colors, with no representational content whatsoever. The Starfish, which brings to mind real-life animals and movements, and even natural reactions such as curiosity and alarm, has no specific cultural associations.
Or in the realm of music, consider the creativity of a DJ (see “The Hit Charade”). What a DJ does is purely “combinational” creativity, or putting familiar ideas together in unfamiliar ways. DJs make no new music. Rather, they combine and order familiar pieces in unfamiliar ways. The value depends not only on the novelty of the DJ’s choices but on their aptness: their capacity to remind us of musical or cultural associations that wouldn’t have occurred to us otherwise.
The wider cultural associations are especially relevant when the music has lyrics. Think of “Eleanor Rigby” by the Beatles. However haunting the music, it would be less valued, and less memorable, without the words. The harsh discordance of the music and the near-savage sounds of the cellos reinforce the bitter sadness of the lyrics. They conjure up the loneliness and despair of Father McKenzie, as well as Eleanor Rigby herself, with extraordinary depth and richness.
A good DJ can take such things into account. For instance, a bitter song such as that one could segue into a sickly-sweet one, with the human audience enjoying the irony.
Pandora can’t do that. AI’s natural language processing is hugely limited by relevance blindness, the result of a computer that lacks semantic understanding or literary knowledge. Computers have written “novels,” but the prose is horrifically bland. And computer-generated soap opera plots (which can ignore verbal and grammatical elegance) will win no Tonys.
We still need people for that.
Margaret A. Boden, a research professor of cognitive science at the University of Sussex, is the author of Creativity and Art: Three Roads to Surprise.
This new data poisoning tool lets artists fight back against generative AI
The tool, called Nightshade, messes up training data in ways that could cause serious damage to image-generating AI models.
Rogue superintelligence and merging with machines: Inside the mind of OpenAI’s chief scientist
An exclusive conversation with Ilya Sutskever on his fears for the future of AI and why they’ve made him change the focus of his life’s work.
Data analytics reveal real business value
Sophisticated analytics tools mine insights from data, optimizing operational processes across the enterprise.
Driving companywide efficiencies with AI
Advanced AI and ML capabilities revolutionize how administrative and operations tasks are done.
Get the latest updates from
MIT Technology Review
Discover special offers, top stories, upcoming events, and more.