7/21/26
They’ve moved the AI goalposts again. Not surprising given the rapidity of evolution of the tech, the illegality of the scraping processes, and the responses of music owners, both the big guys and the little guys.
The RIAA is proposing a two part labeling system for sound recordings. AI in caps indicates fully AI-generated. ai in lowercase indicates AI-assisted. The criteria for the lowercase ai rating are “the recording was created substantially by humans and expresses human creativity; however, generative AI was used for some expressive elements. Humans performed the lead vocal and primary instruments”.
Hmm. Vague and a problem for me. Since I’ve recently relented and decided that BIAB output doesn’t have the energy and fluidity of Suno’s output I’ve gone back to using Suno output in my finished tracks. Now I have to put on the brakes again, because it’s not fair to have my songs labeled as fully AI when just the instruments have been generated by AI and I’ve done everything else — music, lyrics and lead vocals. I’m hoping that an adjustment is made to this rating system in the next few months, otherwise I’m thrown back to BIAB. ACE Studio offers a different approach to generating AI backing tracks and vocals, and I was planning to dive into that after my surgery but now I’ll put that on hold again. Maybe there will be a hue and cry from the huge segment of the songwriting community that uses AI music generation the same way that I do. It’s a sticky wicket no matter which way you turn because everything is so hard to prove one way or the other.
Meanwhile I will release most of my remaining songs that don’t include AI at all. That’s 33 songs left after the 6 already released. I will release most the songs I’ve already created that use BIAB backing tracks, plus all 8 songs on the “In My Particular State Of Mind” album, on which I played all the instruments but drums, plus all six acoustic songs on the “In The Land Of My Heart” album, featuring me and my guitar, recorded live in Whiskey Hill Studios. For the initial release of this album I’ll use the original recordings, remastered and with a few edits in “I Choose” to update lyrics that I’d changed shortly after that recording was made, possible now with stem-splitting so ubiquitous.
I plan to release all remaining 33 songs within the next couple of weeks so that they’re distributed in time for my surgery on 8/19/26. Lots to do but it’s doable. Then after that, if the RIAA rating system hasn’t been tweaked, I’ll be going back to the drawing board for the bulk of my new songs that I’ve finished writing during the past year or so.
6/1/26
My thoughts regarding my use of generative AI continue to evolve. I’ve spent about a month working on songs using BIAB. I even tried Tonics (boo!). I’m quite happy with the BIAB output on the song I’ve spent the most time on, Ain’t Comin’ Back, though it’s been quite a lot of work. Of course, that’s exactly what I expected, which is why stepping away from Suno for producing finished works was such a hard decision. I stand by that decision. I don’t expect to waver unless something changes dramatically in the generative AI world. To use Suno to create finished works is something I can’t realistically return to.
However, and this strains my principles somewhat, it’s still a marvelously useful collaborative tool in the song creation process. It takes the place of songwriters’ circles, music camps, open mics, etc. Hearing my nascent creations played back to me in a variety of styles sharpens my lyric writing and my sense of structure for any given song. Creating a song is a birthing process of sorts. Not a physically painful and exhausting as birthing an actual child, it nonetheless requires interacting with the nascent idea and all of the ideas that flow from it during the process, pushing then relaxing, over and over, until I can hold the child and admire it.
That sounds over dramatic but it’s still true. If I want my creations to be as good as they can be I will edit and edit and edit ad nauseam. And this is where Suno comes in really handy. I’m coercing it into following my song structure, chord progressions, melody, melodic rhythm, and listening back to the way the song hits me in different styles, and how the lyrics sit in my mind and soul, editing with each iteration.
A mechanical design organization would be stupid to keep doing things the old way. Designers on drafting boards, model shops iterating design change after design change on each part design. Those days are long gone. Design in CAD, machine in CAM, create CAM tool paths directly from 3D CAD files. Iterate each part once because form and fit are handled in CAD. So it’s the same with generative AI. Suno handles all the baking track iterations and I handle the design changes. When I’m satisfied I have the final design I move to BIAB to create the finished product. I audition and select the right RealTracks style and in BIAB and set to work importing the tracks into Logic Pro, plus soloist RealTracks. I add my vocal track and mix the whole thing. My final song output uses tracks played by real session musicians, not generative AI. Not illegally scraped audio.
It’s a slow process but more rewarding.
4/25/26
After about a year of using ChatGPT and Suno, and after watching numerous videos by Adam Neely, Rick Beato, Crystal Delgado and others regarding how AI learns and what the impacts might be on the music industry I’ve concluded that Artificial Intelligence is indeed artificial, and certainly not intelligent. What do I mean?
AI is great at memorizing. It can be very good at sorting through what it’s memorized in order to present what appears to be logical and well informed info. Sometimes it is. Other times it’s not. But it always presents the information as hard fact. It also can’t give you information you couldn’t get on your own with your own research. That’s because it’s using the same sources you would use when gathering information. It can certainly do it much faster because all that information is housed in its neural network whereas you have to go out and search for it yourself.
But one problem is that it’s the same information. It’s gathered (scraped) from the internet. As most thinking people know, the internet is chock full of misinformation. And another bigger problem is that AI doesn’t have the human intelligence to tell what’s true and what isn’t true. Where you might take a chunk of information with a grain of salt, considering its source, AI can’t do that. It could have, and certainly some attempts were made during training to help it sort out true from false, but the speed and breadth of the scraping prohibited any comprehensive attempts to validate information. The race to gather up as much information as possible prohibited reasonable validation efforts.
AI has no conscience. That can be valuable at times, but not at others. AI doesn’t have the ability to accurately synthesize information to make quantum leaps of understanding, as humans can. For example, if you see a video of an artist creating a design, you can recognize that you can use a portion of that design to create a whole new design, all without copying the original design verbatim. Your mind sees the potential for a whole new design, just from the inspiration of a few pen strokes you watched another artist do. AI can’t do this. It can only copy what’s already been done. It might be able to break a design into elements that it can reassemble into a new design but there will be no new elements in the design. It’s simply a rearrangement. Of course humans can do this too, and many do. But more creative people see the possibilities for something new and can go off in entirely new directions and come up with something new, using, perhaps, that one element as inspiration.
There is a great danger of people coming to rely on AI to do tasks that, with a standard high school level education, can be done by humans. I’m talking about basic math and language skills. The ability to write and speak coherently, to understand how numbers add up in the grocery store, etc. There is nothing wrong with using advanced tech to assist with advanced tasks, but relying on a LLM to write Facebook posts for you? C’mon.
Now let’s turn to AI music generation. I have had a yearlong love affair with Suno.com. It started slowly, as I initially rejected its origins, processes and results. But I was seduced by how quickly it can turn a nascent song idea into a finished sounding recording, I have used Band In A Box (BIAB) for decades to produce backing tracks for my original songs and have been very happy with the results. Suno’s seduction began when I uploaded a 30 second clip of a very rough idea I’d recorded on my iPhone over a decade ago, tried a few styles, and was amazed at the quality of the output. Rough recordings are never inspiring or encouraging. I wonder, is there really a song in there? Suno always answers YES! Very emphatically. So I began to upload more snippets and was encouraged enough by the results to actually finish writing over 20 songs over several months, all starting with fragments I’d recorded on my iPhone over a decade ago.
All along, though, I wrestled with the ethics of AI training methods. Large Language Models (LLMs) scrape text, image and video content from every possible legal and illegal (copyrighted) source. Music generation models do the same with audio, scraping streaming services and websites for content. In the process copyright holders are not acknowledged and therefore not compensated. As a songwriter who has over 50 original songs registered with the U.S. Copyright Office I’m certainly sensitive to how unethical this process is, and I’ve been wrestling with my conscience the whole time I’ve been using Suno. I have registered one song thus far using Suno tracks as backing tracks for the song. It’s my chord progression, my melody, my melodic rhythms, my lyrics, but Suno’s tracks. I documented just how much AI was involved in the process but I haven’t received word from the Copyright Office that it’s been accepted. We’ll see.
I love the collaborative relationship I have with Suno. And now I’m thinking that’s the way I’ll use this tool moving forward — for feedback, for encouragement, for production ideas. Then go back to my DAW (Logic Pro) and my backing tracks generator (BIAB) to finish the song production. BIAB uses real humans to play real tracks and real humans to write code that can combine the real tracks into useable styles. BIAB was my tool of choice until a year ago and now I return to it. They (PG Music) come up with a significant update each year. There will be no danger of AI detectors flagging my songs, and no danger of the Copyright Office rejecting my songs.
Addendum (5/20/26) - Another issue with AI-generated music that's cropped up: it can be mass produced. Therefore it can be mass uploaded. Streaming services pay based upon percentage of overall listens to their entire catalog. The more AI listens there are, the less money is available to pay real musicians.
I've very recently found a new (to me) tool that uses human studio musicians, much the way that BIAB does. It's called Tonalic. No AI. Just brilliant ways to assemble musical patterns to create human music inside a DAW. So now I have two tools that celebrate human musicianship.
7/15/25
There now are lots of apps and websites that will create songs for you. Until recently I haven’t used any of them to create the songs on this website. I used ChatGPT to help me break out of a lyrical rut once. That was for the song Sidetracked, and I didn’t use any of the lyrics it generated verbatim, but more like a songwriter circle where other songwriters suggest lyrical directions you haven’t thought about. For Sidetracked, the ChatGPT lyrics had an uplifting ending, and when I read those lyrics I realized that there was an uplifting ending implicit in the story already, I just hadn’t been thinking about it.
Artificial Intelligence is a tricky subject. Any computer app, be it a word processor, spreadsheet, internet browser, photo processor, video processor, etc., is a form of AI that we’ve been using for decades. Not generative like the new large language models, but artificially intelligent enough to do lots of mundane tasks for us. Even hand-held calculators, smartphones, and similar devices contained forms of AI in their operating systems, even before AI has became the hot ticket it is now. That’s what makes them such valuable tools. In fact, Bandzoogle contains a tremendous assortment of powerful time-saving tools that I use to create, expand and maintain this website. I couldn't do it without their tool set.
On the musical side, I’ve been using an app called Band-In-A-Box (BIAB) since 1995 to help me generate backing tracks for my recordings. Some of these recordings only contain BIAB drum tracks while I played all the other instruments. Many more recordings contain mostly BIAB backing tracks, or all BIAB backing tracks. This link will take you to a page on this site that explains this in more detail. And my Digital Audio Workstation (DAW) of choice, Logic Pro, gives me the power to sit in my home studio and produce high quality recordings with its level of AI.
Now I’m at a point where my curiosity about AI song generators has gotten me to try one: Suno.com. I’ve been blown away by the audio quality of the music it generates. There are legal aspects that are being explored these days, with class action lawsuits aiming to get AI generating businesses like Suno, Udio and others to somehow pony up and compensate music creators for the unfair or fair use of their content in the training of the AI neural networks. I'm happy, at this point, to let the courts work all that out. After all, my music is also online and subject to being scraped and used for training.
With Suno's latest update to version 4.5 I can now upload up to and 8 minute audio track. This now allows me to have full control over the chord progressions, lyrics, melody and melodic rhythm that Suno will base its “cover” versions of my song on. I'm really happy with the quality of the output, and though I'm using BIAB to create the instrumental tracks and adding a scratch vocal track to the audio I upload, the final resutls are much more dynamic than what I can normally get BIAB to produce no matter how much I manipulate BIAB's output alone.
So my current position is that the genie is out of the bottle and there’s no going back. AI music generation is here to stay. But I'm a stickler for being as original as possible and I expect I will always want full control over the copyrightable aspects of songs, so I'll continue to exercise full control over the chord progressions, lyrics, melodies, and melodic rhythms of my songs, using AI as just another tool in my toolbox.
- Paul Fifield