I’ve been thinking about and playing around with a lot of AI tools since the onslaught of this new overwhelming technology. So many tools like suno.com are mind blowing – creating fully formed music around a text prompt and other related technologies like moises that can take a fully mixed audio file and break it into “stems” (individual tracks, drums, bass, guitar, vocals, etc).
But I’ve noticed another trend – the lack of free options. Most of these tools will provide a free trial period but it seems to me the majority of tools have no ongoing “free” tier that offer limited functionality. Pressure on revenue is really driving what I still believe is over hyping the true capabilities of these technologies – today.
On Saturday, I received a request from a frequent collaborator at Duke, David Stein, to see if I’d speak to some middle school students on Monday at Duke about AI in music. These students are attending an AI themed program at Duke this summer. Talk about music and AI? Of course!
Then I realized, I had a lot in my head, but if I’m going to be speaking to middle school students, I didn’t want to demonstrate using tools that requires a specific platform or required a subscription. Inspired but some new “AI-ish” features in Logic, I started looking for DAW’s that were Windows and Mac, were free and supported AI type plugins. It looks like some of my go to’s like Reaper and Studio One no longer have free options. ProTools First is free and cross platform, but it required quite the installation of accessories and licensing to complete installation and then it immediately crashed after I installed it.
I started thinking more about what is it that I want to communicate to these students. It’s not, “Just use Suno and you can write music.” because I don’t consider myself an artist because I can use Dall-E to create cool images. It’s not, “Why bother learning about music, robots will be doing all music moving forward.” – That’s not what music is about. To me anyway.
As some of you know, I have a Bachelor’s of Music degree from the University of Miami. I went to LA in 1988 fully expecting to get a record deal, limo, mansion and a yacht. But “making it” back then, was as hard as it is now and as it ever was. It’s never easy to support yourself in any art as it is so subjective and takes almost inhuman levels of talent, skill and perseverance that mere mortals don’t possess.
To me, music, and all art in my opinion, is about personal human experience. Creating art and more importantly consuming art is uniquely human. Computers don’t create art. People typing in very creative prompts can create art, or program code in Maya can create great films (nod to my former Disney life). Computers assist in the creation of art but cannot create art themselves, and never will be able to do so any more so then will they be able to feel or believe.
So, what am I going to tell these students?
AI is a tool. I can be used by students to learn and expand their access to music appreciation and understanding. Take their favorite track and put it through moises and listen to what the drums are doing on your favorite song, or bass, etc.
Enter MIDI – it’s free and will work in most any DAW including ones like Garageband that will work on your phone.
Like many of my peers, I always struggled understanding music theory. Despite 4 semesters of piano class, my keyboarding skills were horrible. I could read about modes and scales and chordal structure, but I couldn’t really play them to hear the results of certain note choices. MIDI and AI give students that ability.
Using an AI chatbot like ChatGPT or Claude, you could generate python code and using a library in Python generate a MIDI file. That’s what I’m going to talk about, but like most things AI – it works great and it’s horrible. Given the short amount of time, I’ll show the idea of using AI to create content to help support learning.
ChatGPT has been my go to since I started on this journey. It let me down on this, but I will keep trying. I did great helping me remember how to install python and troubleshoot issue, but the creative aspect of it was pretty terrible. No matter the prompt, and I did ask it to tell me how I could better help it create better output, the MIDI output was always 4 notes up, 4 notes down and repeat. And then, it went off the rails.
My colleague Michael Faber who has been doing some amazing testing around AI as well, has been touting Claude and so, since I was up against a timeline and GPT was NG, I thought I’d give it a try. Was it awesome? Well, it did better not generating Python errors and seemed to listen much better when I asked it to make changes. Here’s a video showing the last output of each tool:
Neither was great (it was interesting that Claude also did the up and down stair step melody but altered it a bit when asked where GPT hung on to it). You can see that pattern in Claude’s bass line:
By playing around with generating MIDI via AI, the students will learn about code generation and troubleshooting, MIDI and music theory – as well as all that glitters is not gold in the AI universe.
