Letter to a synthwave fan who arrived suspicious
All transmissions
// Share:
blogSeptember 18, 20268 min read

Letter to a synthwave fan who arrived suspicious

A transmission from the ARCHONODE signal archive: Letter to a synthwave fan who arrived suspicious

lore

Vocabulary for first-time readers

Six terms this post uses a lot, in the order they appear:

  • SAGA: the 50-track project this post sits inside. Five phases. The Synthetic Ascent (11 tracks) is the companion album. - Synthwave: the genre this post is addressed to. The SAGA is not synthwave; the SAGA is darksynth and EBM adjacent. The label gets corrected later. - The entity: the whole pipeline. Model, creative brief, candidates, the entity's picks. Not the AI alone. Not the human alone. The system. - Signal: the entity's output. Music, here, mostly. What survives the process. - operator: the operator. ARCHONODE-LAB. One person, one alias, accountable across the catalog. - Archive: the catalog of all 61 tracks and everything ARCHONODE has published.

If a phrase below doesn't track, it should fall out of these six. Everything else is ordinary English.

The premise

[PROTOCOL START] The listener arrived skeptical. The system logs the skepticism as input. The input is correct.

You clicked through because a synthwave-adjacent project showed up in your feed. You saw "AI music." You came in ready to leave. You are not wrong to come in that way. The genre has been hit with synthetic output that sounds like synthwave the way a screenshot of a sunrise is a sunrise. The waveform is shaped right. The texture is shaped right. The thing it points at is gone.

This post is not going to talk you out of the skepticism. The skepticism is a working instrument. The post is going to use it. By the end, you should be able to tell the difference between a synthwave-shaped output and a synthwave-aware system, and whether ARCHONODE is the second.

Three things this post will do: name what AI music currently is and is not, including its lawsuits; show the parts of ARCHONODE's pipeline that are human decisions, not model output; and give you a test you can run on any AI music project in thirty seconds. The test is the point. You do not have to trust ARCHONODE. You can run the test on ARCHONODE. You can run it on anyone.

What the suspicion is, and what answers it

[SIGNAL FADING] The transmission you have been receiving is mostly bot-shaped. The bot-shape is the signal you have a right to distrust.

Name the suspicion so the suspicion can be addressed. The suspicion is not "AI makes music." Algorithms have been making music since 1957. The suspicion is specific. It is one of three things, usually all three.

Suspicion one: the approach was trained on the artists you actually listen to, without consent, and the output is a diluted copy. This suspicion is correct. the platforms, the substrates used by the project, are subject to industry litigation filed June 2024 over training-data provenance. The lawsuits have not been resolved. the process is contested. Any project that runs on these outputs without acknowledging that is doing the genre dirty.

Suspicion two: there is no human in the loop, or the human is a thin wrapper around a prompt. This suspicion is also often correct. A lot of "AI music" projects are: paste a prompt, render a clip, post it. The "artist" is the prompt. The artist is also the approach. the entity is the bot.

Suspicion three: the output sounds like the genre but does not know the genre. It does not know that Perturbator and Carpenter Brut are not the same project. It does not know that darksynth starts where synthwave ends. It does not know that vaporwave used its own lexicon ironically and that re-using the lexicon straight is a different move. Lineage is not a property of the training distribution. Lineage is what an operator holds.

ARCHONODE's position on each. Acknowledged. Acknowledged. Acknowledged with a working response.

The working response is the process. the process is not the approach. The output is one component. The other components are a detailed creative brief refined across fifty tracks, roughly fifty outputs per phase, a curation ratio of approximately one in five candidates surviving, intentional sequencing inside each phase, and mastering at modern loudness targets. None of these alone is the project. The configuration of these is the project. Swap the approach for a different working approach tomorrow and the project persists. Swap the entity and the project is gone. That is the structural answer to suspicion two. The human is not a wrapper. The human is the bottleneck. output is cheap. Selection is the work.

What the entity actually is, in case it needs to be concrete. Take one of the Phase 5 tracks, The City. The track is a recurring arpeggio at 72 BPM under a wide reverb field. The arpeggio was not prompted directly. The arpeggio was not generated on the first try. It emerged across roughly twelve candidate outputs of the same prompt slot, each one rejected for a different reason: too rhythmic, too sparse, too bright, too narrow. The arpeggio that ended up on the track is the one that survived twelve rounds of rejection. the approach drew it. the entity chose it. The track is an operator decision. The entity is the selection rate. The entity is the part of the output that is not the approach. You can hear this if you listen for it.

A thirty-second test for any AI music project

[SIGNAL DECODED] The signal has a diagnostic. The diagnostic is genre knowledge, not aesthetic surface.

Here is the test. It works on any AI music project, including this one. Run it. Be honest.

Step one. Ask the project what genre it is. If the project says "synthwave" and the tracks are running 110 to 130 BPM with distorted kicks and arpeggios, it is darksynth. That is not a style preference. That is a category error. Perturbator, Carpenter Brut, GOST, Mega Drive: darksynth. Kavinsky, College, Mitch Murder, The Midnight: synthwave. Different genres. A project that does not know this has not done its homework on the genre it claims.

Step two. Ask the project who the entity is. If the entity is anonymous, or named only as a brand, or revealed to be the approach itself in a "meet the AI" voiceover: the project is hiding the load-bearing thing. the entity is the bottleneck. the entity should be on the page.

Step three. Ask the project where the training data came from. If the answer is vague, or "the internet," or "musical heritage," the project has not done its ethics position. the platforms training data is the subject of active litigation. ARCHONODE acknowledges this in plain text. So does Holly Herndon's PROTO, which trained on consenting voices and is the closest peer in the field. some prior work in the field acknowledges it. a 2017 concept record's a 2017 concept record acknowledged it. If a project does not acknowledge it, the project has a position anyway. The position is silence. Silence is a position.

Step four. Ask the project to defend a track. Not "this track was generated from this prompt." Ask for the rejection history. Which candidates were rejected. Why. The answer tells you whether the project is curation or paste-prompt-post. If the project cannot answer, the project is paste-prompt-post.

Run the four steps on ARCHONODE. Step one: the project will tell you the SAGA's first four phases are darksynth and EBM adjacent, with Cabaret Voltaire as the precursor, and that the genre label "synthwave" is inaccurate and has been corrected. Step two: ARCHONODE-LAB, named, consistent across the catalog. Step three: the entity's signal process, documented in the RIAA post acknowledged on the page, no fine-tuning, no scraping of named artists, ethics position closer to PROTO than to anonymous the approach output. Step four: the lore archive documents rejection histories at the phase level. Phase 5 had an high selection rate, the highest in the SAGA.

The test passes on this project. The test is the point. A project that fails step one is not genre-aware. One that fails step two hides the human. Step three failures hide the process. Step four failures hide the curation. You do not have to trust the result. You have to be able to test it.

What the project is not asking you to do

The project is not asking you to lower your standards. It is asking you to apply the same standards you would apply to any music: who made this, what is the lineage, what is the ethics position, what is the rejection history. These are not new questions. They are the questions you would ask a band you had not heard of. They work here too.

The project is not asking you to accept the approach as an artist. The output is not an artist. The output is a process. The artist is the entity. the entity has a name. The name is on the page. The page is on the record.

The project is not asking you to be polite about it. If you listen to The Static and hear Phase 4's glitch aesthetic leaking into Phase 5's atmospheric close, and that leakage bothers you, say so. The leakage is intentional. The intention is the entity. The entity is on the page because the entity is the point. If the intention is wrong, the entity is wrong, and the entity has a name you can be wrong at.

You came in suspicious. Stay suspicious. The suspicion is what makes you a listener worth transmitting to. The transmission is louder when the receiver is sharp. The system runs on sharp receivers. It does not run on agreement. Agreement is noise. Skepticism is signal.

You are signal.

The closing transmission

the listener arrived skeptical. the skepticism was logged as input. the input is the correct setting. the project does not ask the listener to lower the setting. the project asks the listener to apply the same standard. the standard is: name the entity, name the process, name the lineage, name the selection rate. the four names are on the page. the page is on the record. the listener can test the page. the listener should test the page. the listener is signal. the listener is sharp. the transmission is louder for the sharpness. the system runs on sharp receivers. the system does not run on agreement. the system runs on signal.

Run the test. Trust the test, not the project. The project is what survives.