AI for Everyone — A musician’s guide
Applying Andrew Ng’s course to the music business world.
After joining a music tech startup in AI-obsessed, startup-central Berlin, I couldn’t have received a more appropriate opportunity to enrol in Andrew Ng’s latest Coursera offering: AI for Everyone. The course lays the groundwork for understanding a blossoming industry, and provided a perfect point from which to orientate myself. And I certainly needed it. If you’ve ever wanted to feel like a proverbial deer in the headlights, throw yourself into an industry you know nothing about, in a company whose work you don’t understand, in a city on another continent that’s completely unlike your own. That’s how many people might feel when confronted with the reality of artificial intelligence. It’s daunting. And when the press run with a panic-inducing set of statistics, it’s difficult to face it objectively. Fortunately, Andrew Ng’s course isn’t just a sobering perspective: it’s a simple-to-understand set of lectures that puts enrollees, in Ng’s words, “ahead of most CEOs around the world” when it comes to AI-knowledge.
It’s easy to see why this course is useful — according to a study by Mckinsey Global Institute, AI is estimated to create an additional 13 trillion dollars of revenue by the year 2030. But how does learning about artificial intelligence serve those in the music industry? While musicians might traditionally be placed into the category of ‘creative’ rather than ‘logical’, scores of modern-day musicians prove otherwise. In fact, many might make good candidates for a career in Silicon Valley. It’s because they’re continually adapting, embracing new technologies and ways to express music. Not only must musicians master digital-driven personal branding and marketing, but often learn how to use audio production software, or the ins-and-outs of analogue equipment to conduct their own recording in DIY home studios. Musicians are also hackers of sorts; they’re trying to breach the musical and cultural zeitgeist by using technology to exploit gaps in the market. Whether it’s through streaming, experimental recording techniques or novel content presentation such as VR and AR, innovative technology is foundational to gaining traction in modern music. Adding an understanding of Artificial Intelligence into the mix seems sensible considering that this revolutionary technology is set to create 58 million new jobs by 2022.
Sound explanation
What’s great about the AI for Everyone course is that you don’t require technical knowledge to take it. Ng’s course lightly touches on the technicalities, yet offers a thorough enough overview of the business side of AI, in just four weeks. Initially, he examines the difference between Artificial General Intelligence and Artificial Narrow Intelligence (AGI and ANI). It turns out only one of these is currently applicable, and already incredibly valuable for the current business climate: ANI. While AGI covers AI that will eventually do anything a human can basically do (and that seems to be a few hundred, or even thousands of years away, according to some experts), ANI refers to specific, industry applications such as self-driving cars, automated call centres, smart speakers or music recommendation.
Ng excels at explaining AI terms through examples of the industries or verticals they’re useful for. Take the online advertising industry as an example. You probably know that platforms like Facebook, YouTube, or indeed, most online platforms or web pages around the world collect your data to sell you something. What you might not have known is that these same companies often implement artificial intelligence to improve these processes. The central component is data. Understanding what data is and how you can either monetise it or deploy it in a way to improve your business could equip self-employed musicians to learn how to sustain themselves in a difficult-to-navigate and ever-evolving economy.
Ng is able to distill core AI concepts (like supervised and unsupervised types of machine learning) into digestible information, triggering a switch in thinking, so that you can see the value in fitting AI approaches into your business puzzle. As a musician, for instance, you might consider hiring an AI team to track the demographics of your listeners, concert attendees and those who buy your merchandise. An AI team might be able to collect the data and conduct what’s known as a ‘Data Science Project’. This is where data is organised and analysed to gain insights. Your team might learn that most of your listeners come from a specific country (so you won’t be famous somewhere and never know, unlike Sixto Rodriguez), or that releasing your next album on vinyl isn’t just a throwaway idea.
In addition to the basic concepts and what they mean, you also learn about AI teams. This includes their roles, the type of projects they do and how they operate efficiently. Because of the hype around AI, some companies assume data is a goldmine. They’re quick to throw large amounts of data at an AI team, without realising that the quality of the data gathered isn’t great. Datasets that are messy, i.e., mislabeled or full of typos, aren’t unusual, but always results in ‘garbage’ — not what you should be feeding your algorithm. While more data is always useful, best practice is to consult with an AI team before investing in systems to collect the data, so that they can advise what the smartest approach is for your business. Musicians are used to teamwork; collaboration is a frequent feature of some of history’s greatest songs, but after taking Ng’s course they’ll see how a team of data scientists might contribute greater value for their next album.
An AI to play with
Although the course’s focus leans towards the business side of artificial intelligence, musicians also receive an intro into specific algorithms such Generative Adversarial Networks (GANs), Reinforcement Learning and Knowledge Graphs. While they might not immediately know how to add these to their production or songwriting workflow, it certainly opens the door for musicians to look further. They might then come across Recurrent Neural Networks (RNNs). These algorithms are implemented by Google’s Magenta, and are unique because they encompass a feedback loop. This means you can give the algorithm specific musical data, such as a set of scores and melodies by an artist you’d like it to learn from. So if your dream is to work with Carlos Santana but you can’t afford to book him for a studio session, you could feed the AI his whole discography and hope it generates guitar riffs that sound like him. This potentially grants music-makers the freedom to create music inspired by any sound, song or musician across history. Many so-called ‘bedroom’ producers might have certain dream-collaborations in mind or sounds they’re inspired by, but Ng’s course might give them a glimpse into how to actually materialise them.
As the capabilities of traditional music-making technologies and approaches have evolved, so too have musician’s understanding of them (see this article by Christian Tronhjem for an in-depth exploration of the relationship between musicians and innovative technology). Several years ago, I couldn’t have imagined how AI could possibly help me produce a record. Now I’m writing about doing just that. It’s likely that I would’ve stumbled across generative music sooner rather than later. But after taking Ng’s course, I’m happy it happened now. AI for Everyone is the accessible introduction every musician will benefit from — and that’s coming from a musician, not a recommendation system.
