AI does everything else, may as well let it entertain you

Want to find the next big song before it becomes cool? AI can help with that!

---

Your rotating cast of Back Page correspondents has written about AI coming for your job on multiple occasions. 

In recent times, this newfangled technology has been reported to perform better than humans on mock ob/gyn examinations, provide higher quality and more empathic answers to medical questions, read ECGs faster and more accurately than humans, find solutions to all of Medicare’s problems and even generate abstracts for research articles.  

Despite this ever-increasing list, here’s one job this Back Page contributor hadn’t considered AI might come for: record company executive. 

Researchers from Claremont Graduate University in the US claim to have developed an AI model that can correctly identify whether a song will be a hit (or a flop) with near-perfect accuracy – something that has challenged those who make and sell music for decades.  

As part of the research, published in Frontiers in Artificial Intelligence, 33 participants listened to a mix of 24 “hits” (songs with over 700,000 streams) and “flops” from a variety of genres released in the six months prior to the study while their heart rate and other neurophysiological data were recorded.  

After looking at the above list, this Back Page scribbler (who only recognises one of the “hits”) has to put their hand up and accept they may no longer be “with it”. 

The neurophysiological data was measured to identify how “immersed” someone was in the music, with the theory being that listeners will be more immersed in “hit” songs than “flops”.  

Researchers created a much larger “synthetic” dataset of neurophysiological responses (10,000 observations) based on the data collected from real participants, as small datasets can highly bias machine-learning approaches.  

This approach is also known as “neuroforecasting”, where data from a small group of people is used to predict population-level effects – without needing to collect data from hundreds or thousands of people.  

Using the synthetic data, the machine-learning model accurately classified the song type 97.2% of the time – 96.6% of the time for “hits” and 97.6% for “flops” – a vast improvement from the 69% accuracy of the normal logistic prediction model. Impressively, the AI model correctly identified “hits” 82% of the time when only using data from the first minute of each song.  

“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,” Professor Paul Zak, a neuroscientist and senior author of the study, said in a statement.  

But importantly, when applied to the data collected from real human participants, the AI model accurately classified hits 96% of the time. 

“If in the future wearable neuroscience technologies, like the ones we used for this study, become commonplace, the right entertainment could be sent to audiences based on their neurophysiology. Instead of being offered hundreds of choices, they might be given just two or three, making it easier and faster for them to choose music that they will enjoy,” said Professor Zak.  

While AI may be able to tell you what music will be a hit, and the radio plays what they want you to hear, this Back Page scribe is confident it will be some time before we have robotic rock patrols cruising our streets in self-driving Black Thunders and handing out icy cold cans of Coke.  

Sending your favourite song to penny@medicalrepublic.com.au will get you a banging reply.