How will AI and chatbots have an impact on the music industry?


AI and bots are two of the technological buzzwords currently at large in the music industry, but what impact will they really make on the way we discover, listen to and share music – as well as connecting with the artists that make it?

This morning, Music Ally hosted a conference panel at the by:Larm festival in Oslo to kick those topics around.

The panel included Ieva Martinkenaite, VP of Telenor Research, heading up its AI lab; Syd Lawrence, founder of The Bot Platform, which is powering Facebook Messenger bots for a growing number of musicians; Lucy Blair, head of international marketing at The Orchard; and Gregor Pryor, partner at law firm ReedSmith. The moderator was Music Ally’s Eamonn Forde.

“Artificial intelligence has become a hype in many industries,” said Martinkenaite in an introductory presentation. “Everyone is saying AI is going to change the world, AI is going to be the most important technology in the 21st century.” But she warned that the first step towards delivering on this hype is to understand what it is.

And? It’s a system of technologies, methods and applications, enabling machines – computers – to do what humans usually do: intelligence. “The interesting part of AI is for computers and machines to do things that humans easily do,” she said. “Machines can sense, they can comprehend, they can act upon, and they can learn. And the last is the most important, in research… a machine can learn and adapt its behaviour based on external incidents.”

Martinkenaite added that artificial intelligence research started back in the 1950s, but that the real drive around R&D has come since 2013/2014. “First, the computational power – the processing of the computers – has become massively better… Second, our behavioural patterns have changed tremendously. We sleep with mobile phones, we eat with mobile phones, we date with mobile phones… so there are massive data points. Big data is coming and making a difference. And third, emerging change: self-learning machines. Computers based on behavioural patterns.”

Her last point: “It’s important to understand that AI is a hammer. It really depends what you want to do with it. It’s not magic: you have to understand the problem to be solved,” she said. Attracting and retaining customers based on a better understanding of their behaviour – whether you’re a telco or, say, a music-streaming service – is one of those problems ripe for solutions.

“There is a tremendous interest now,” said Martinkenaite, who said that Telenor’s AI lab will be launching on 8 March to redouble the telco’s efforts around these technologies. “If you’re not really investing long-term… you’re probably going to start losing big-time, pretty soon.”

Lawrence warned about the dangers for the music industry of chasing the AI hype. “The music industry isn’t charging in as fast as a lot of other people, like the tech world. Technology is there to solve a problem, and there’s been far too much hype over the past 12 months especially with regards to AI chatbots. I hate that term with a passion!” he said.

“First they’re not AI, and second, no one ever wants to chat with a bot. They’re there to solve a problem… All industries are guilty of diving in to the hype. We’re getting a lot of people thinking AI is something you can plug in, like a thing.”

Pryor talked about the investment landscape. “Savvy investors understand that AI for AI’s sake is not worthy of a lot of investment. The music industry has been very heavily disrupted by technology, and the distribution chain is constantly the tail that’s wagging the dog when it comes to the rightsholders,” he said.

“To the extent that the distribution is influenced by artificial intelligence, you can certainly see that has interest from investors, because it potentially has real value for the rightsholders.”

AI panel

Martinkenaite talked about the growth in the number of phD students and researchers in AI. “But we also have to see the other side of that: the Google, Amazon, Facebook and the rest, who have been heavily collecting data on their platforms, not yet used commercially. So it’s a question of when and how they are going to strategise around it,” she said.

“These big giants that are setting the stage having lots of data, training their models, but not really selling and commercialising them. It really depends on long-term research, and that’s coming especially if you look at the US and China in particular. But it really depends on the other side: what is that AI going to be used for?”

What about music-industry uses? Pryor cited AI music composition. “The labels have experimented with this,” he said. “The problem that’s ostensibly solving is if you are a YouTuber and you want to use music in conjunction with your video, often there’s a licensing challenge that you can’t overcome. With artificial intelligence… the machine will just spew out something that’s fit for purpose, and as far as the video producer’s concerned, it’s royalty-free.”

He admitted there may be ownership issues around that. “The companies that created the machines that then create the music have taken a pretty hands-off approach when it comes to how royalties get paid. I on’t think the law has properly taken account of who would own that: some people think it’s the person who created the machine, some people think it’s the person who provides the inputs for the work it creates,” he said.

Martinkenaite: “It’s data… Software is cheap, data is not. And what you do with that data… for these deep-learning algorithmic experiments, when a machine can learn and give weights to an incident… you need massive data points, continuously gathered in real-time,” she said. Lawrence said AI composition will need this too: the ingestion of a big catalogue of songs to understand how they work, training its model for creating new music.

“If I take a Rolling Stones song and use that as inspiration, that’s fine, but if I use bits of it to create a new song, it’s breaking the copyright. From a machine’s perspective, it depends how they create the new songs out of the previous songs. But you still have to train it, so there’s a grey area there,” he said – in terms of who would be responsible for a machine’s copyright infringement.

Blair talked about how the music industry feels about another aspect of AI: machine-learning to power music recommendations on streaming services. “Obviously AI is going to learn what your listeners’ preferences are, and what music they like to listen to when, in different contexts,” she said. “You have mass-personalisation now, with Spotify’s Discover Weekly and Release Radar playlists.”

She added that music marketers are still “drowning in data but starving for insight… AI is fascinating, because I’d love to be able to get to that point where ‘this person consumes this music at this time in these particular contexts. What can you then do with that data… to better connect that person with the artist?”

Blair also wondered whether more use of machine-learning risks losing some of the serendipity around discovering music: can algorithms be trained to throw the kind of curveball recommendations at us that, say, a DJ, a journalist or a friend might?

The conversation moved on to smart assistants like Amazon’s Alexa, and their limitations in their current form. “You use it for certain tasks. I do use it for music, whether I ask it to play a certain radio station or a playlist. I use it to turn on and off my lights, and I use it for various other bits and pieces. But they’re all very task-driven. I walk in and ask it to turn on the kettle, and then ask it what my To-Do list is,” he said.

“I really think that it is such an over-egged topic, and those without any knowledge of AI think that AI can do everything, but everything has to be trained… I do believe that these things are amazing, and they don’t need or don’t have real intelligence.”

Could AI have a role in the A&R side of the music industry? “It’s all about business. In non-music businesses, when it’s customer acquisition and working out what customers want, you can use various bits of tech to be able to establish how good your customer is, how likely your person is to become a customer, and therefore what type of money you can make from them,” said Lawrence.

“Ultimately that’s what an A&R’s job is. I don’t think that’s a good thing: I would love it to be more about the art… one of the things that machines can’t do is emotions, and music is very emotional,” he continued.

“There is no science behind emotions. But I think certainly data – which already plays a major part in A&R – will help, and that will grow. And I think it will be a shame once machines take over A&R and it is all just machines-driven. In some cases it will work, and it is already happening now.”

Pryor talked about music curation, citing Apple Music’s emphasis on human curators. “Already you can sense the machine is nowhere near as intelligent as the human being when it comes to certain aspects of music selection. And the other thing machines can’t do is be social, and music is inherently social: that peer recommendation of your friends saying ‘have you checked out this certain band?’ he said.

Lawrence chimed in, citing Spotify’s machine-driven Discover Weekly playlist. “It’s all based off playlists that humans have created,” he said. “A lot of it is smoke and mirrors, and I think it’s amazing.”

How can the music industry create partnerships with the research side of AI? Martinkenaite gave a perspective based on Telenor’s new AI lab.

“The lab is going to be open, and it essentially has to be open for other partners in other industries to come in.It has to become a magnet,” she said. “The interesting part of research with data will be different sorts of data. It could be mobile data, music data, combined with e-commerce data…” Something that music partners could provide.

“And second is real problems. In music, distribution, acquisition, context… we need researchers to get a good understanding of the problems to be solved,” she said.

Pryor said that the music industry will have money to invest in this area, thanks to the streaming-fueled resurgence of revenues for the major labels. “They’re going to have cash to invest, so I think they definitely will invest in it,” he said.

“They understand that AI can have quite a dramatic influence on their revenue. If there’s one pool of money from streaming that’s finite from those consumers, what the labels and publishers want is a bigger share… they’ve set up funds, they’ve started incubating, and they’re supporting startups. If AI can affect their business, they’ll invest.”

He warned that it won’t be about which company has the deepest pockets to invest, but about which can identify the smartest, most relevant partners in the AI world.

Written by: Stuart Dredge