GabyThe era of artificial intelligence (AI) is here, and neural networks will only get smarter. Their impact on art and pop culture is already significant. Convincing deepfakes are already worrying record labels.
But don’t dismiss AI’s potential benefits for the music industry. One likely advantage for some executives is predicting which songs could become undeniable hits.
How exactly do neural networks work? Will they be able to determine future hits?
Researchers measured the neural activity of 33 people who listened to 24 songs and combined that data with statistical modeling. The neural network could then almost perfectly predict which tracks would become hits and which would fail.
The lead author, Professor Paul Zak, says that by giving AI access to neurophysiological data, his team could nearly perfectly predict hits. He calls the result a breakthrough: the network predicted whether millions of listeners would respond to new songs based on the brain activity of only 33 participants. Zak says nothing like this level of accuracy has been seen before.
Participants, aged 18 to 57, wore heart-rate sensors and listened to a playlist of 24 recently released songs selected by a music service. A song counted as a “hit” if it had more than 700,000 streams. The playlist spanned genres and included 13 hits and 11 non-hits.
After the listening experiment, participants completed a survey about the songs. The survey asked whether the composition was offensive, whether they had heard the song before, and whether they would recommend it to friends.
But the key data were the neurophysiological reactions to the songs. Using “neuroforecasting” from the 33 participants, researchers predicted how other listeners would react—whether a track would be a success or a failure—without collecting data from thousands of people.
The neural network produced an extremely accurate forecast for millions based on the small group of respondents. Zak says the collected brain signals reflect activity in the brain networks tied to mood and energy.
Using a linear statistical model, the researchers achieved a 69% prediction success rate. That wasn’t bad, but the AI hit 97.2% accuracy on the same task. The team also found that when the network used only the first minute of a song, accuracy remained about 82%.
Forecasting hits not only in music
Zak says that if similar neuroscience technology becomes common, platforms could offer entertainment tailored to listeners’ neurophysiology. Instead of hundreds of options, listeners might get 2-3 song choices, making it faster to find music they’ll enjoy.
The researchers said music companies could use this to “easily identify new songs as potential hits that people will add to their playlists.”
The study had limits—the song selection and sample were narrow—but it’s easy to imagine a future where music, TV, and film skip traditional test screenings. Especially if AI models can predict user success with better than 80% accuracy after just one minute of media exposure.
That said, about 100,000 new songs are uploaded to the internet every day, so music fans’ choices aren’t likely to shrink soon. As the publication Newatlas reports, Professor Zak adds, “It is quite likely that this approach can be used to predict hits for many other forms of entertainment, including movies and TV shows.”
