As I've mentioned in various places, I make a point of showing up on the forums while developing IMFMSynth and taking users' opinions on board. Something I noticed there is what prompted me to write this.
The "AI slop" problem in the audio industry
From what I gather, the industry has a problem with so-called "AI slop" (in Japanese we call it AI ポン出し, roughly "pop it out with AI and ship it"): low-quality products churned out in bulk, and forums worn down by the fallout. It's a touchy subject, so I'd rather not name names, but I've seen fights break out after someone posted a release of that kind, and not in just one place.
As the developer of IMFMSynth, an FM and wavetable synthesizer plug-in, I'm sure some people look at me through that lens too. It isn't something I'd put in the news section, but I wanted to lay out where I stand and how I think. My case is a bit complicated, because the developmental disabilities I live with, ADHD and ASD, come into it.
Honestly, how I feel about AI
Some people consider AI training, or making things with AI at all, to be theft. I think it depends on the case.
Manga artists and anime animators, after all, study the traditional or trending art styles of Japanese anime and manga, digest them, and draw in their own way. As long as they don't rip anyone off, that's naturally seen as a good thing, and society consumes the results. AI training is no different: you learn, and you turn it into something. If it's a rip-off, people get angry.
Which brings us to the question of whether there's any creativity or ingenuity in making things with AI. I think that depends on who's using it.
Let me explain with a bit of vocabulary from the software world.
What do you want to make? Whom do you want to help? What do you want to make easier? Putting that into words is what we call the "needs" (要求). In my case, it was this:
"FM was complicated, and you had no choice but to lean on presets. With the technology of the day, nobody could build a UI that let you really master it. Couldn't we solve that by combining today's web technology with good design? Couldn't we build an FM synth that even a beginner can use to design sounds on their own?"
That was my starting point.
Next comes the question of what kind of product, with what specification, would meet those needs. That is the "requirements" (要件). In my case:
"Be compatible with DX7 presets. Bring in modern elements such as wavetables. Deliver on those needs through a patch-based UI that is both traditional and intuitive."
That was the list.
What it comes down to is whether you, as the person making the product, can take responsibility for the needs and the requirements. That's anything but meaningless. If anything, it's probably the most fundamental essence of human creative work.
Now let me show you what would have happened if I had deliberately skipped the needs and the requirements, and irresponsibly tossed IMFMSynth at an AI and shipped whatever came out.
And here is the UI of the current version of IMFMSynth.
It has grown more complex, and yet I think the design hangs together. A product maker should let the product do the talking, so I'll leave it there.
My ADHD and ASD
I have a diagnosis of ADHD and ASD. My secondary conditions, depression and a sleep disorder, are severe, and I hold a Grade 2 Mental Disability Health and Welfare Handbook (精神障害者保健福祉手帳).
Since a lot of my users aren't Japanese, let me explain. The psychiatric category of the disability handbook has three grades: 1, 2 and 3. You can read the abstract criteria on a government office's website, so let me just give you my own understanding:
- People at Grade 3 are often treated as what you might call "barely normal." They live with the hardship of the social skills they exert by fighting their disability going unseen and unappreciated, and being taken for granted.
- People at Grade 2 mostly can't hold down a standard full-time job of eight hours a day, given the severity of their disability. They live with a kind of social pain that differs from Grade 3.
- I worked eight-hour days myself, but I simply couldn't endure it, and as a reasonable accommodation I now work six hours.
- People at Grade 1 generally need nursing care, and most of them live in a hospital ward or in the care of family or a facility.
In the interest of honesty, though: this is only my own impression. I'm a patient, not a doctor or a civil servant, so please take it as one person's opinion.
Now, to be concrete about what becomes a problem in daily life and as a developer: I'm abnormally inattentive. I come from a background of academic research in mathematics, Fourier and harmonic analysis in particular, and I'm good at optimization. Yet, wildly out of proportion to that, my basic lapses of attention are staggering.
When an AI writes code and streams it into the terminal, I have the knowledge and experience to call things like this:
"Hold on, right there! In DSP you don't need to call an exponential function every time. It's discrete, so when a Voice starts you can compute a constant from the decay time and cache it, and then all you do is multiply by it each step, right?"
"Hmm, this is off too! An adjacency matrix should be computed with bit operations, using a type that matches the CPU's register size."
I'm also the kind of person who starts by implementing the trigonometric, logarithmic, exponential and power functions from scratch. And I don't think of myself as disciplined, or rather I don't feel I have the right to call myself that. But looked at objectively, I suppose I'm someone who's very particular.
On the other hand, I forget to initialize variables all the time. Or I write something, it doesn't behave the way I wrote it, and it turns out I never called the function. Or I decide on the cause of a problem too early in debugging, miss the real issue, and waste time. In short, I burn time and mental energy on problems like these and on fixing them. This was a ball and chain on my creative work.
Then again, the hyperfocus, impulsiveness and fixations that come with ADHD and ASD have a good side too: they let me commit to a product like IMFMSynth.
What filled in that pain and frustration, made things easier for me, and on top of that let me carry real responsibility as a working adult was the AI assistance of recent years.
I don't make products for the money. I make them as self-expression, as a working adult.
Me and AI
I use AI as a means: to put into words, give shape to and make real the dreams of me and all of you users, and for my own self-realization. At the same time, precisely for that reason, I gladly accept the responsibility that truly matters.
These words come from everything I've described and from my love for my product. If you can see that I don't take this lightly, I hope that settles it.