On AI in Music Composition and Image Generation

Back on March 4, Derek Thompson interviewed composer Mark Henry Phillips on his podcast “Plain English” about how AI might change the music industry. It was an interesting interview (to me) because Phillips himself discussed how impressed he was with the music it created from written prompts, how he had used it to complete some of his own unfinished compositions he’d abandoned when he got stuck, and a variety of other examples. Phillips was fascinated and, of course, worried that it would put him out of a job, while also noting that, by learning to use AI himself he could increase his productivity and be inspired to consider directions he might not have otherwise. What really intrigued me was an experiment he and Thompson ran.

Apparently, Thompson is an amateur pianist and composer himself. He found himself in a recording studio recently, and he decided to take advantage of a keyboard that was in the room to record, using his iPhone, a short composition he'd sketched that was inspired by the style of the Russian composer Dmitri Shostakovich. He then sent the recording to Phillips and asked him to "extend" the composition beyond the bare bones he'd played. Phillips did so, building out and elaborating on Thompson's musical sketch, and the results were pretty nice. Phillips said he could have done a lot more if the initial recording had been done on something more hi-fidelity than an iPhone. Thompson sent the product to his wife, who really liked it.

Yesterday, I received a Substack post by Ethan Mollick, whose book,[ _Co-Intelligence: Living and Working with AI](https://a.co/d/eIGKhOv)_, helped me to understand AI better, while allaying some of my fears. The post is called "[No Elephants: Breakthroughs in Image Generation](https://www.oneusefulthing.org/p/no-elephants-breakthroughs-in-image?r=3358ve&utm_campaign=post&utm_medium=web)." Mollick is not an AI promoter, nor a naysayer. He is an academic trying to understand what AI can and can't do, and every time there is a new release in AI world he takes it for a test run. This post showed the recent advances in AI image creation.

Mollick not only shows us the images, but also describes in detail the words he used to generate the images. So it's not just product, but process as well. Several things occurred to me as a result of these two explorations.

First, that **the new Most Important Person in the Room is going to be people who have a command of descriptive language and knowledge of past art styles and artists.** In other words, art and music history majors, or more broadly, the humanities. I'm looking forward to seeing universities who have zeroed out the arts scrambling to reconstruct new departments focused on AI, but which will _require_ all the traditional courses. We'll still need programmers and engineers for more high-level work, of course, as well as artists who can make original rather than derivative work.

Second, that **creativity will become more democratized**. Did you hear the joy in Derek Tompson's voice when PhilIips played for him his enhanced composition? More people will be able to have the experience of _making something themselves_. Instead of creativity being confined to a small number of designated professionals, people will be empowered to express themselves more gracefully. I can imagine parents and grandparents, aunts and uncles, who previously had to buy whatever birthday cards and party favors were available at Target, being able to generate things specific to their kids' likes and interests. People who never could draw or play an instrument will, through the power of words, be able to express themselves. Kids will develop an interest in art and music (and film, when that technology matures) that they wouldn't have had if they'd needed to buy a piano or a raft of art supplies. People will more easily be able to create an image for a blog post, or for their web page. In other posts, Mollick asks the AI to create an interactive game for him to teach certain skills that he teaches in his business classes. Teachers will be able to create presentations and teaching games tailored to the kids in their own classroom. This is like the transition from having to know HTML to create a website to being able to do so through drag-and-drop elements in Wordpress. It's similar to using Adobe Photoshop without the steep learning curve, or at least, a _different_ learning curve. Will this reduce jobs for commercial artists or composers? Maybe at first, but at some point professionals, who are skilled in working with AI, will be in demand because they have joined an understanding of AI with their own knowledge and creativity to make fabulous things that someone else would never have imagined.

Is AI going to be disruptive? Definitely, in the same way that self-driving cars will be disruptive, that drones will be disruptive, that robots continue to be disruptive. How many of you use tax software to file income taxes? Are there fewer employees at H & R Block now? What happened to all the typesetters who were necessary to publish a book before desktop publishing took over? Some of them didn't make the transition, but there are still typesetters using their skills to lay out beautiful books and magazines -- in fact, the manuscript of the third edition of my textbook, _Introduction to Play Analysis_, is in the hands of one right now, and their work makes my amateurish attempts in _Building a Sustainable Theater_ look kind of lame. But there is no way I could have found a publisher to publish a book with such a small audience, nor could I have afforded to have a freelancer lay it out for me. But it is now out in the world, and selling in the mid-single digits even as we speak.

I totally understand the anxiety that writers, artists, composers, and filmmakers have about AI. I remember many people having similar feelings on campus when MOOCs started to be created by well-known professors at prestigious universities. People worried that administrators would use MOOCs and online courses in general to provide courses at much lower cost than employing an actual, live professor who may or may not have been as "good" as the star prof from MIT. And yet, it didn't happen. In fact, conventional wisdom is that online education may have severely damaged the learning skills of the COVID students. Sure, I can now go on Udemy and take a class in HTML or day trading or some topic that interests me, and I can go on YouTube to learn how to build a picket fence or replace a faucet or fix a bike, and that's awesome, but is it a replacement for all education? Well, if you've ever ghosted an online course as I have, you know the answer.

My main point is that there is another side to AI than what is being promoted by people online, and we ought to at least imagine those benefits before going into wholesale condemnation mode.

[I'm not going to address the whole issue about LLMs being "trained" on "stolen" material -- that's a topic for another day.]