News about how artificial intelligence can potentially enhance the listening experience often takes a back seat to worries about it ruining music and artists’ livelihood. Spotify wants to change that.
Finding your groove
Like ChatGPT with a specialty in thoughtful tune curation, Spotify Technology SA’s new AI Playlist tool instantly whips up a soundtrack to your life from a text prompt. The feature, which is currently in beta in the UK and Australia, supposedly goes beyond Spotify’s AI DJ launched last year and the personalized daily mixes it already pushes to its listeners. AI Playlist, we are told, knows what you want to listen to based purely on an idea.
Want some instrumental jazz while you tinker with pivot tables, or a high-tempo musical backdrop for a power walk around Hyde Park in the spring? Want to feel like Russell Crowe in Gladiator? Yeah, you do. AI Playlist has you covered, says Spotify.
As with most generative AI experiments, persistent human input is key to getting the most out of this tool. In trying it out, I found the initial playlist generated from my prompt often missed the mark of the precise musical vibes I was after. Mercifully, a ‘Refine’ button allows you to work with the AI to hone the list and better match your mood.
This is an important development. Spotify has long used machine learning to tailor products like Discover Weekly to the listener’s taste. AI Playlist allows you more control over the shape of the listening experience than any of Spotify’s similar features, and it responds surprisingly well to human coaxing.
A familiar lament from listeners is that Spotify’s algorithms are known to generate “personalized” selections that essentially feed you the same songs you already listen to in a slightly different order. After some experimentation, I would say this latest tool relies less on my top songs than the others. That’s surely good for getting out of the comfort zone and expanding beyond those five Beyoncé anthems you’ve had on repeat for days.
That said, the more I refined the playlists, the more my old favorites crept back in. Which just goes to show what an unwieldy challenge this entire subjective business is for software makers.
Also interesting is that much of the new music the AI suggested came from big names, which resurfaces a chronic issue with music discovery online. When I asked explicitly for a soundtrack made up of “little known London artists,” the results were dominated by Arctic Monkeys and Kasabian (plus Vampire Weekend, who I’m pretty sure are very New York). “Unsung Indie bangers” generated the exact same list.
Perhaps all this will improve as the product advances out of its testing stage. The streamer firm notes that “the most successful playlists are generated with prompts that contain a combination of genres, moods, artists, or decades.” But whatever I tried, it was hard to see much air time given to musicians of lower and less mainstream profile, even with broader prompts.
Spotify says there are “measures in place around prompts that are offensive.” True to policy, a request for R. Kelly songs was swiftly rebuffed as were any attempts toward an overly political vibe. A prompt for some music to “Make America Great Again” had AI Playlist gently directing me to celebrating music “across a wide spectrum of genres and styles.” I guess better to leave that stuff to the podcasters on the platform.
Bizarrely, a desire to listen to “woke songs” was transformed into “a playlist to keep you energized and informed.” Its big hitters were mainly Kendrick Lamar, 2Pac and J. Cole with a smattering of Eminem.
Given the overwhelming choice of playlists out there, the novelty of the AI version is its ability to compute the weird and wonderful, and conjure up something to suit your whim. To make it more than a curiosity, Spotify needs to iron out the obvious wrinkles, including handling niche requests with greater sophistication and ensuring the inclusion of smaller artists.
Source: Bloomberg