Unless you’ve been purposefully hiding under a rock recently, you’ve probably been inundated with AI-related content. Two weeks ago it was everyone’s Lensa AI-generated portraits, and this week, it’s been ChatGPT, which I used to generate the intro for the most recent episode of Mid-Riff. I was inspired by a couple of other podcasts— Hard Fork and Today, Explained, which did the same, and it really is kind of bonkers.
Rather than just reading the AI-generated text, I decided to up the ante. I use a combination of Logic Pro and another program called Descript for editing. I’m assuming you are familiar with Logic Pro, which is a very popular audio editing software, but Descript, if you are less familiar, takes a transcript of your audio and allows you to edit not just from the waveform as you would in most audio editing, but also from the text. So, for example, if someone forgets what the question was, I can go to the text, just delete it, and it’s magically gone from the audio, too. It can also automatically remove crutch words, most of which I keep, but I do remove most ummms and uhhhhhhs (mostly successfully, but sometimes I have to help with the cuts and fading).
Descript also allows you to train your voice to generate, essentially, a robot version of yourself. You can feed it text and it will read it, in your own voice. I’d never really tried it, so, I decided to do this with the ChatGPT-generated text, and that was the intro, which honestly wasn’t too bad. All I did was type “write an introduction for a podcast about gender and music gear”. The audio could probably work well if you were feeding it just a word that you misspoke on or something, but as you can see here, it’s missing a bit of personality.
I do want to start with a caveat: as a psychologist by training, this is not my area of expertise. But I will place my area of expertise, gender, specifically, and diversity, more broadly, over it. Also, while there are implications here for many areas, such as education, I am focusing here on music and art.
MY EXPERIENCE WITH CHATGPT
When I first tried ChatGPT, I asked it to “write a blog post about stereotype threat and music gear.” Since those are fairly specific topics, I was assuming it would be too hard, but it actually did pretty well. So, I asked it to write something longer, and honestly, even though it didn’t have a lot of personality, I could have put that on my website, and no one would have batted an eye. Yikes.
Next, I wanted it to do a creative task, so I asked it to “write a song about being a mom.” This is where things got a little rough and where I think folks who writing creative output have somewhat less to worry about. It was pretty cookie cutter and awful, even after using a few different prompts, such as “write a metal song about being a mom,” which did start with the line, “I’m a mom of chaos.” So, there’s that. I also had it write a hilarious song about Providence, Rhode Island’s disgraced former mayor, Buddy Cianci, which was also hilarious. I found that when I asked it to write a poem rather than a song, the results were somewhat better… but not much.
So, that was my approximately 24-hr foray into ChatGPT before it reached capacity internally and I haven’t been able to access it since.
While some of the responses were hit or miss, overall, it is surprisingly competent technology. And in some ways, the fact that it is better than we might expect leads to our fascination with it. I will add, I saw just today that Canva now includes text-based AI, and as someone who blogs, I get ads for different text-based AI programs, like Jasper, who advertises itself as an AI copywriter, at least once per day.
HOW DOES GENERATIVE AI WORK?
ChatGPT and other generative AI works by scouring text databases they are fed by its programmers and creating patterns based on what it finds. ChatGPT used text from the internet up until 2021. You can see how art- or music-based AI would use different sources to generate its patterns and the how source material of any AI can impact the quality, relevance, and diversity of responses.
As you’ve likely seen, ChatGPT is just one space where AI has been the rage lately. As I’d mentioned, Lensa, the art-based generative AI program has blown up. I’ve also recently seen an AI-created guitar tabulature program, and there are programs now that actually create AI-generated music (Tencent Music in China has generated 1000+ songs with generated AI vocals that are getting millions of plays). And you’ve likely had AI generate a music playlist for you— maybe even today.
LEGAL and ETHICAL CONCERNS for ARTISTS
If we know this AI pulls from a piece of writing, artwork, or music, the next question often comes to ownership. But if thousands or millions of pieces are being scoured to generate a particular piece, whose is it? Is it the artists whose work is the source material? Is it the person who programmed it? Is it the AI generator? What if the source is public domain? How do we know that someone created a piece of work themselves or generated it through AI? These are concerns both legally and ethically.
There have been instances where someone used Lensa to generate images with commands using specific artists’ names, which would lead the program to create a piece using just one artist as source material. What if you, like, I did, asked ChatGPT to write song lyrics in the style of Fleetwood Mac or Sonic Youth, specifically? With hip-hop and sampling, we have some legal precedent here, though it took quite some time to establish. However, most samples pull from just one source, rather than combining many together, which makes things slightly more complicated.
A friend of mine is a visual artist and every time someone shares their Lensa images, one of the generated works looks eerily similar to her work. As you can imagine, she, and many other artists, have been really upset about what they view as theft. Since this is all so new, we don’t yet know where this lands, legally, and laws will have to catch up with technology.
Part of the legal, ethical, and artistic concerns here are based on how we define creativity.
Is it creative? Is the work that’s made actually even good?
There are a few commonly used definitions of creativity, but Runco and Jaeger (2012) delineate two, characteristics that they identify as part of the standard definition, which is originality, or novelty, and effectiveness, or utility. The language has varied slightly over the years, but it’s those two main traits that are most important. Something both new and useful.
AI’s IMPACT ON MARGINALIZED GROUPS
As you can imagine, there are a variety of potential issues related to AI and creativity, musical echo chambers, or flattening creative output, for example— like in my song about being a mom. And, while that is a concern for me, I am more concerned about the likely disproportionate impact of this technology on oppressed groups.
While there’s already been evidence of gender bias in music playlist algorithms, in this case, much of this has to do with source material. If, due to the discrimination they’ve faced, a particular group is less represented in the source material used to generate a song, for example, the creativity and perspectives from that particular group’s experiences won’t be present in the AI-generated creation. And, if stereotypes are present in source’s data, they will then be found in what’s created. Essentially the representation problem is multiplied.
Recently, an AI-generated rapper, FK Meka, who is presented as Black, was created by two men- one Asian and one white. After becoming the first AI artist signed to a major label, the artist was recently dropped from Capitol Records after accusations of its recreation of racial stereotypes and appropriation, using racial slurs and sharing AI-generated images of the artist being beaten by police.
ChatGPT has already noted the potential for issues with sharing untrue information or harmful speech and has included some tips about how to prevent it. It looks like they’ve been attentive to it, but that’s going to be a long hall, I imagine.
Once legal issues start to roll out, the first issue that might have a disproportionate impact is likely that of ownership. In the music space, this might show up in a few ways. First, if an artist is less likely to be picked up by a label due to their identity or don’t have the money or information needed to copyright their work, their work is less likely to be protected. And then, even if they do have the rights to their work, they might be less likely to be able to afford to sue to uphold their copyright, if it is stolen.
So, issues of representation, ownership, appropriation, and stereotypes will likely all be major concerns moving forward. You can read more about AI and colonialism here, or find additional readings related to these topics here.
CONCERNS IN MUSIC GEAR
There is no shortage of pedal clones or fights about pedal clones in the industry. Lawsuits are common. Companies who are particularly litigious in nature, like Gibson or Rickenbacker seem to be continuously vigilant against trademark infringement, so much so that there is an entire era of “lawsuit” guitars.
With this, questions have arisen- is it acceptable to copy any design? Just the most popular/ubiquitous designs, like a tube screamer or a strat? Just designs that are no longer made? Does it matter who they were originally made by? Is it okay as long as they are given proper credit?
What, happens then, as we enter the potential realm of AI-generated schematics, guitar bodies, amp modeling (maybe we’re there?), plug-ins, or industrial designs?
Whose ideas will be sourced, who will get credit, how will they be compensated, and who’s left out?
I don’t have the answers here, but I hope that manufacturers and programmers alike are taking these questions into consideration as we move into this new landscape.
HOW DO WE HANDLE NEW TECHNOLOGY?
Of course, every time a new technology enters our cultural purview and use, there is an initial panic. This makes sense, especially since we can’t entirely predict how it will be used.
In the music space, you could see that with electric guitars, with synthesizers, and probably the closest analogy here- with sampling. Of course, people have found amazing ways to use all of this technology to push creativity forward, including the creation of entirely new genres of music. As with most new technology, there is enormous potential for musicians to use AI to collaborate and push creativity further.
However, it’s important to be mindful of the impact of this new technology, especially on marginalized groups in both its development and usage.
In the meantime, I hope you come to see my AI-generated noise project, “Mom of Chaos.”