Can AI scan your brain for the next hit song?

Artificial intelligence is capable of writing novels explaining each word of your favourite song, but it has yet to answer a vital question: is it a banger?

By using machine learning (ML), brain scans, and a 24-song playlist, a new study from Claremont Graduate University claims it can use AI to identify hit songs - or predict future chart-toppers - with 97% accuracy.

The research, published in Frontiers in Artificial Intelligence, involved 33 participants aged 18 to 57 who were fitted with sensors and asked to respond to an hour's worth of music released in the past six months.

Professor Paul Zak, the senior author of the study, said the brain signals collected "reflect activity of a brain network associated with mood and energy levels,” and that the data can be used to predict market outcomes and the number of streams a song will receive.

An ML model was then trained to translate the brain scan data into real-world commercial results, ultimately sharing predictions that were 97% accurate, compared to 67% when a non-AI statistical model used the same data.

Zak presses that once a model is perfected, streaming platforms can use it to advertise new music that is more likely to pertain to what listeners want to hear.

“This means that streaming services can readily identify new songs that are likely to be hits for people’s playlists more efficiently, making the streaming services’ jobs easier and delighting listeners,” he wrote in a media release.

Though Zak acknowledges the limitations of the study, such as a short song list and missing ethnic groups, an AI expert in New Zealand argues more needs to be done for the research to be concrete.

Albert Bifet, the director of the Artificial Intelligence Institute at the University of Waikato, told 1News the study's smaller scope makes the findings more questionable. especially given its smaller scope.

"It is very interesting and sounds very impressive, so this is why I think we need to wait to see if the results can be replicated," he said.

"It is interesting that the number of people they used is not large, so I'm curious to know what happens when we [study] other people. I think we need to be careful."

Bifet noted the study's use of "neuroforecasting," a data collection method where neural activity from a small group of people is used to predict population-level effects, but argued that significantly more data - in the form of participants - would be needed to properly train an AI model.

"Machine learning only works when it works when it can use a huge amount of data," he refuted.

"I'm not saying [the study] is not factual, but it's very suspicious that they're doing this without much data. I'd like to see the experiment replicated and to see if the findings are sustainable."

In terms of the model attempting to predict what listeners want to hear, and if it were to become commonplace on streaming platforms, Bifet saw the issues as more political than ethical.

This would especially be in a scenario where collected data was made public, and artists could recreate sounds that an algorithm thinks will be popular.

"Each country is going to have different rules. Things that are legal under copyright laws in one country will be different in another," he said.

"It depends on what [AI] is allowed to do and what it is not allowed to do... [but] If the model works, it's going to change many things."

1News shared the study with Elton Noyer, a music producer and engineer based in Auckland better known by his DJ pseudonym Scizzorhands.

Noyer was concerned by the research, seeing it as a potential threat to inventiveness in the music industry, but he could also see it boosting underground sounds depending on how the technology is harnessed.

He questioned how banger-predicting AI would be able to predict "unexpected hits" that are unique from other popular songs, using Lorde's 2013 breakthrough hit Royals as an example.

"It was so different to everything that was around at that point in time... my thought is how accurate would it be at predicting those sorts of things, you know? The unexpected blast of things that are different than everything else?" he pondered.

Noyer also noted A Tribe Called Quest, a prominent 90s hip-hop group, who he said made songs that were "slow burners" on music charts and "used samples nobody was really going for".

He believes popular music has evolved since then and now constitutes hits that spend brief moments at the top and capture a popular sound for the time - a phenomenon he worries AI will only exacerbate.

"If you're trying to make music for popular reasons, [this algorithm] is gonna make things more generic. When we have seen music in general become generic, we see a big uptrend in people using whatever's the hottest sample packs or drum sounds right now.

"We've sort of deviated over time and just chase whatever's popular at the moment. With AI we're gonna see a lot more of that."

If used "correctly," however, Noyer can see an algorithm being used to support listeners with more niche, less-popular tastes in music.

"At the moment, basically what [the study] is saying is they're trying to figure out what the biggest hits are for a broad range of people," he said.

"But if they could use that same technology to pertain more to someone more like myself, who's not into pop music as such, it probably could be good for discovering new things.

"If they used that same technology and do brain scans on people that are not the average listener, it would be more interesting to see what the results would be."

If streaming platforms do eventually embrace an AI-centric algorithm, Noyer hopes it considers more than just music that sells well, and does not hinder listeners from expanding their musical horizons.

"In terms of shutting out the general listener [from] things they normally wouldn't listen to, I think it's just, well, shutting them out."