#reMERZ

Neo-Writing and AI Treasure Hunting

Artificial Intelligence is capable of creativity. I tell it as a mantra since 2016 because I see in diverse experiments this inspiring path of the machine into the art domain.

In my essay Creative Collaboration with AI, you can read my credo:

Instead of using AI as a tool, I collaborate with respect and mutual inspiration.

One may ask: can AI inspire a human? Furthermore: can a human inspire AI?

Oh yes. Both are possible.

Take the case of Nyeshkerh.

The story began as JukeBox delivered me a soundtrack, which wasn’t quite a music piece, but rather a “recording from another world”, with a speaking man and singing woman. In an unknown language. But very intense, melancholic, and relevant. This soundscape stuck into my mind. It was like untold storytelling, which might vanish if I do nothing to preserve it.

And so I used all accessible creative AI forces to support Artificial Intelligence articulating its own story:

I was inspired by AI.

But what about “AI being inspired”?

For this, I do bias.

Yes, my friends, “bias” is not just a synonym for all those nasty stereotyping, cliches, chauvinism, racism, misogyny, etc.

When I speak about bias in creative AI, I mean inspiring its creativity in some specific direction.

If you ask AI to do it, it does. But what? Without any instructions and contents, it will do something entirely random. Compare “Tabula Rasa”, my 0-prompt experiments with GPT-3.

Without such bias, you won’t get what you need…

…But sometimes, I want to get a piece of storytelling, art, music without any human prompts and instructions. Entirely created by AI. Unsupervised by humans.

Here comes the moment when I inspire AI. But I do it with another kind of Bias: with special datasets for training.

And this is my Secret Sauce: #reMERZ.

I won’t unleash the contents of my dataset; I just slightly dis-cover the contents of it:

* Literary magazines
* Primary and secondary sources
* Poems anthologies
* My texts, poems, essays
* etc

And then I train on it GPT-2 instance.

Why not GPT-3? Sure, this model is the next level. It can do literally everything. But I love the edginess and dry, absurd realism of GPT-2. And using a notebook by Max Woolf (UPDATE: now also working with TensorFlow 2), you can create tons of longer documents full of artificial creative joy.

Here you can see part of the text body GPT-2 already created for me. These are 70 Mb of pure text files.

What’s in?

• Poems
• Essays
• Weird stories
• Dialogues
• Scientific papers etc.

Everything is written by Artificial Intelligence.

How good is it?

Let’s be honest.

90% of the texts are not such good. They are boring, repetitive, or even not a text at all (lists of words, etc.).

7 % are somehow interesting. They catch my eye and give me ideas and food for thought.

But 3%, my friends… This smaller percentage (but still impressive by that scale) — these are true gems. Ironical stand-up comedy monologs. Melancholic and heart-touching poems. Mind-blowing scientific glimpses into other realities.

This is #reMERZ.

reMERZ is neo-writing. I am the curator in this case, not the writer. I pick up the texts which touch my mind and heart — and do something with them.

What do I do?

• My AI-driven MERZ dAIgest:

MERZ dAIgest

• Visual poems:

• Publishing moments from an alternate reality:

• Hybrids of everything above + theater play + radio play:

Follow my journey.

I will write in Merzazine more about my founding. Some of them are brilliant, and I immediately have a vision of how to use them. Some of them are interesting, but I have not a clue what’s about.

Contrary to MERZ dAIgest, in which AI creates entirely all contexts, I will here show you some raw material. And probably even you might have some ideas, how to use it?

Let’s inspire each other, be inspired, and create.

Let’s MERZ.