AI Stuff

My thoughts on Artificial Intelligence is in the evolutionary stage. I have thoughts, but do not pretend they are my final thoughts.

Artificial Intelligence (henceforth "AI") at the moment of this writing is mostly jargon. Despite this, I don't think we can stop the use of AI in our culture at this point. It will persist indefinitely and at an ongoingly larger capacity as we march into the future; malleable creatures that we are. But it is entirely to be determined how we will actually use it. (Most of the marketing reflects an over abundance of word salad that confuses speculation with adoption. A lot of what has been presented is nonsense that most of us will never pursue.)

We will each choose how we leverage or do not leverage it. Those choices will beget something that none of us can reliably predict. Too early to tell.

Zero of the words that appear in my works are AI generated. I don't use AI for words. And I likely never will if for no other reason than I am stuck in my ways. AI fails to accurately express myself. The words I choose (despite using them to produce fiction) are idiosyncratic sacred expressions of myself and they flavor the tone that shapes a signature that I can readily trace back to my identity. That identity is heavily biological and also something resoundingly soulful. It is an identity that can only be inaccurately inferred in the digital domain.

Despite largely missing an entire data set, AI currently, at best makes statistically calculated random guesses by using an extraordinarily narrow collection of actions that are performed in a biased manner in a digital landscape. That bias seems to further bias itself towards larger data sets. E.G. This one is like that one. Put em together and say they are the same! Look! Trends! We know the answer!

Predictive trend lines are not future facts and there is no singular answer.

The AI I have thus far encountered fails to have the necessary purview of my personal bias. This bias most often thwarts the accuracy of the AI predictions. I rarely use them. Yes, it can get something right some of the time. But not reliably so as it is predicting an event that has not yet occurred in a very consistent manner.

Generally, when AI attempt to assert themselves into my orbit (e.g word prompting), I find they are reliably inconsistently wrong. They are most often pesky mosquitos in my digital experiences and rarely beautiful butterflies. At best they are a draw and suggest a word I was going to use anyways. Often they suggest a word that is similar but one I would not use. I may not use the prompt because it is not the word I prefer. I may not use the prompt because my word usage is unpredictably idiosyncratic and the word is entirely non idiosyncratic. Or I may not use the word because it does not meet some form of external aesthetic that the AI does not have within its dataset.

So, yeah, no AI where words are concerned.

The two dimensional pictorial works are also not generated using AI. I use the Procreate application for almost all of these works. Any photos do not use AI. I actually prefer some degree of traditionally rejected artifacts in a photo. E.G. a tourist attraction as it is with a million people in entirely unstated states of being over a tourist attraction with the people remove via AI I am in general in complete rejection of using AI to sanitize our world in an attempt to shield it from its actual state.

The audio (henceforth "tracks") I produce are increasingly complex, but the short answer is what I create will increasingly incorporate various levels of AI. At the moment of this writing the use of it in my tracks is limited and when used it is targeted and restricted as I will describe below. Perhaps I will update as I progress?

How much AI in these is too much AI? I dunno. Ultimately the levels cannot be so excessive that I am unable to see my reflection in the results I produce. But the levels can certainly exceed level zero as I do already use some AI based applications to support the creation of the tracks I produce.

Currently the tracks I produce reflect a substantial input from a variety of digital and analogue instruments that are clearly outside the bounds of AI. I do not expect this to change much as I enjoy the creative parts of these processes more than I enjoy the automated AI processes that seek to produce similar results. I have a fondness for the tactile interaction with non software instruments and I believe I the end result of a complete track benefits from that world.

Examples of digital and analogue instruments that I use range from guitar to software synths, to various external samplers to voice. I extend this line of thought to include guitar pedals and digital amplifiers and modelers.

For what it's worth, I do not use any form of AI for voice mangling. I don't prefer that approach as it removes a key signature for my tracks. I enjoy the overt limitations of using my own voice. I think encountering the discomfort of hearing my voice streaming back through the headphones is beneficial. It forces me to accept multiple perspectives of my work will exist. Many will be uncomfortable.

The only avenue I can think of that I would consider for using AI for voice would be something that would better enable me to approximate a more feminine version of my voice into my works. That could be very interesting. That said, I won't pay for a subscription for such a solution as I find that impedes my purposes. And I absolutely have zero desire for any solution that is simply mimicking the voice of others. I would find that simply boring.

I do, on occasion grab a royalty free loop that as a voice. Those can add an interesting flavor if you mangle them. Loops can be fun, but I prefer using the loops I create using other processes.

I often use presets when they are available to the instruments/pedals/Interfaces I use. If I exclusively used only tones that I created from scratch I would be forever dumpster diving into tones and would fail to produce tracks. My focus is on complete tracks rather than sound creation. The division between the two practices can be unclear at times. Bottom line, presets are endless these days and presets save time when you are in the initial stages of seeding a piece. Besides, they exist. By the millions. Back in the day people romanticized the piano as it was deemed the commonality for all tone and that all pieces could use that tone as a reference point. Everyone wanted a common sound. Now there are some who believe only those sounds that never existed can be worthy and seem to infer using a preset is a bit of a cheat. I am not one of those folks. It seems odd to avoid a preset if it produces the sound you are looking for in your track.

It saves me time.

When using presets I simply look for something that fits a tonality I am seeking in a track. I may or may not modify it as possible. Often I barely change anything. Maybe something fundamental and slight. I may play some notes manually. I may use a sequencer. I try to limit arpeggio triggers. But that is only to force me to learn new things and to not get carried away with over using them.

So in general, I rarely create a tone from complete scratch. You could say I use the "ready to bake" approach.

At the other extreme, I generally adjust pedals by hand rather than by foot when it comes to sequencers and voice recordings. And even guitar. It is done in a cautious and limited performative manner. I do this to introduce uncertainty and to make each track unique. I can reremix, but never repeat a track. I guess I could use Midi for some of this. But I change things in and out so often I would have to use a lot more time sorting MIDI parameters than just doing it by hand.

These performative actions are fully considered and I undertook them as accepted risks when I was in that very moment. It may sound great. It may sound bad. I may have to remove portions from the final track. I may walk away from the effort entirely and do something else. Who knows what will occur? Finding out is my idea of fun.

This sound often travels through a variety of sequencers that move as a snake through the external affects into my Interface and into Logic X. Through this pathway and effort and a certain degree of creativity I derive all of the voice samples as well as the backbone for many of the tracks I produce.

I heavily rely upon this practice of sampling. I use short samples, long samples, overly repetitive samples, infrequent samples. Nothing is out of bounds. I do also use samples/loops created by others, but it is less rather than more common.

For what it's worth, I do not use samples or loops that knowingly violate copyright.

I do use a Digital Audio Workstation (Henceforth DAW). For clarity I will say it Logic Pro. But I don't think the specificity of the DAW I use shapes my opinion on AI. Logic Pro just happened to be where I started and it is just what I know. I use it to collect and consolidate sound artifacts into tracks. Investing in learning complex software takes time away from creating tracks so I don't use multiple DAWs. My priority is creating completed pieces. Anything that distracts from that already difficult process easily stalls the effort. I approach this like a dumb fool climbing down Everest telling him self "whatever you do, you must keep moving forward. If you stop you die!" Learning new software kills your tracks.

Logic Pro does increasingly incorporate AI features into its solution. So far it is not demanded that I use it. I could entirely avoid using that functionality if I so chose. That approach is necessary for those of us who create works using this application. Interestingly, my current machine is unable to use many of these features. I do use a couple of the features and is in track creation where I do use AI. That is a choice I have actively made.

Logic Pro includes session player functionality. You want a bass player playing some reserved notes, you can have one. Need a drummer to keep a consistent beat? You can have one. Want a pianist playing chords with both hands? Sure thing.

I use them to create samples to hang about the tree like garland and ornaments. They are usually mapped back to something I created using external instruments or internal midi instruments. Most of the time I substantially modify the session presets and initial suggestions that are offered by the session player functionality. I most likely have never used the first offering presented. The initial sounds they provide are often overly masturbatory and showboaty and don't often provide the right feel I am looking for in a sound.

So I consistently modify them into something more reflective of what I wish to hear. I then further mangle those results using effects. EG a delay pedal. Or adding reverb. Or something really weird and glitchy. Sometimes I do this to jarring extremes. Most of the time it is to highlight a specific sound that I am hearing through the mix. Most often session players are secondary to something else in the mix. Sometimes they are merely serving as a metronome duct taping the chaos surrounding the mix. I use them in a very utility oriented manner. I don't give them solos as they lack ego. I don't need to appease them. They need to do a job and get along. Sometimes happy accidents occur, though, and the sample can be surprisingly pure in a state that require little modification and other than some minor nudging of a knob here or there we leave it short at one or two or a few bars, but mostly intact. Then we move on.

What I do not do is just stack session players one upon another willy nilly and then, bounding like a rabid lassie across an empty field, without modification, just let them run independently playing off each like monster eating pac mans e saying "ta da!" I don't get any joy from that. I also find they have their own idiosyncrasies that get unnecessarily tiresome over a long stretch, ATMP. They can be melodic. They can be technically sound. But I find them oddly sterile most of the time. Left unchecked they soon behave like a house band having to vamp for an indeterminist amount of time waiting for the prima donna headliner to finally come to the stage. So they riff, endlessly in an uninteresting perfectly timed manner as though they are looking over their shoulders waiting and waiting. Even their humanism features suffers this lamentation.

Oh, but except that one time. And that other time. Oh, and that time where I felt that such a use produced an effect type of feel so I left it as is. I feel there are multiple ways to approach these features. One way says abandon ship. And another says we maybe can just patch the hole? This is an obtuse way of me saying that I often test my assumptions to expose my innate human bias by using AI functionality regardless of my bias in an attempt to find what I call "The Line of Demarcation". The Line of Demarcation is that line where I clearly know I am not creating but only the AI is creating a track. I am forever looking where the line that tells me exactly where "Not enough me!" resides. I have yet to find that critical line of delineation. It's out there, I know when I am clearly in its darkest AI waters. But I search and search for where the actual line of critical demarcation resides. Because I want to be on the "There is enough me!" side of that line.

Everything I incorporate into a track must be trimmed and groomed to serve some other driving element that originates from within my head when I am behaving as a producer. Arguably, the exclusivity of not using human session players is the primary tragedy of the track I produce. But arguably the tracks I produce are an entirely different creative process that isn't really bound by the mores imposed upon traditional music. I certainly don't pretend that they belong in a traditionalist category.

I do use Logic Pro's remastering process. The software in question self defines this as an example of generative AI. Using it provides me a quick ability to check the state of the sound of things in the mix as the piece progress. In the end that process is, for me, mostly a back end mostly post creative pre production task. It influences how the work is presented but it is not the potatoes that are found in the tots. If that is objectionable, that is ok. That objection doesn't change what I do. Sometimes I do reject the results. The results sometimes leave a bit too much sheen on my tracks. There are other options that could resolve that, but my current computer can't run those portions of the process due to technical constraints.

Regardless, I often use what I can for remastering. The time it saves me is put back into the portion of the process I believe most influences my work. Perhaps, as I mature in experience I will choose otherwise? Perhaps as I progress onwards I will find myself being even more precious of my time and use even more AI for these types of proceses? But for now, I require the time that some of this functionality saves.

I have not yet begun to investigate moving visuals. But when I move into that direction I will evaluate accordingly. Part of the learning process is learning what is moving the needle for efficiencies of time vs what is inhibiting the creation of an expressive action. I have a pretty high bar (though it objectively may not actually seem like it most of the time, ha!) as to what I actually want to present on Canned Parasite. It has to have the right feel. It has to have the right content. It is very much a feel. But I am entirely ignorant on those processes and will have to figure it out my perspective as I go along. As such it would be premature to express positions of any kind other than "I seek to limit the use of AI to tasks that fall outside of the creative process." Whatever the fv<K that means.