Or would you rather be a fish?

My last blog post was almost four years ago - it may as well have been from another life. Blink, and time disappears. People come and go, places come and go, and on some days you're not entirely sure you know the stranger in the mirror. That's self-reflection for you, pun intended.

But I've kept this blog alive. I've spent silly money to keep the lights on. It's not that I've stopped writing, of course not. I've written many words over these quiet years, but they were more akin to journal entries than narrative pieces. Quick scribbles on busy days for an audience of one. I kept the lights on because I believed that one day when I was done living through some adventures, I'd like to write about them. That the scribbles would somehow come together in a narrative. I don't think my adventures are done yet, but I suppose today is that day.

If you grabbed my box of tales right now and gave it a good shake, you'd hear all the new ones rattle about inside. Places I've been, things I've seen. And if you were to ask me to pick one out and begin, I'd start with Ron Padgett's The Line:

There’s an old song
my grandfather used to sing
that has the question,
“Or would you rather be a fish?”
In the same song
is the same question
but with a mule and a pig,
but the one I hear sometimes
in my head is the fish one.
Just that one line.
Would you rather be a fish?
As if the rest of the song
didn’t have to be there.

I've found poetry here in Seattle. There are poems in the grass by the Elliott Bay trail; I hear them when I go for a run. There are some poems that fall from the sky during ember sunsets; I hear them when I walk in the rain. If you sit in Suzzallo for a few hours, and your ears get used to the silence, you can hear them whisper in the stained glass panes above the bookshelves. I've learned to notice these poems - feel their presence - but they're slippery things, and not very easy to capture or convey. Once or twice a week, some of us who listen get together at Hugo House and try to express what we've heard. It's difficult, and it's fun.

As someone working in Natural Language Processing, I know we're witnessing a new age of AI development. In late 2022, OpenAI released ChatGPT for public use, and Pandora's Box was opened. A new class of auto-regressive models trained at a hitherto unfathomable scale on a hitherto unimaginable volume and quality of scandalously acquired data. The model seems 'intelligent' and produces human-like output. While the world was distracted by this magician's act, legitimate concerns were brushed under the carpet. Look at those fireworks in the sky, the thief tells you while stealing your wallet.

Privacy concerns, copyright violations, energy needs, risks of misuse, and implicit biases - all suppressed or ignored at worst and transformed from ethical issues into legal ones at best. "Is this the right thing to do?" replaced by "Can I get away with it?". That's how tech works, right? Disrupt the world. Go break things. We'll see what happens later.

We are at later, now - from the point of view of a poet, the later is a world where everyone wants to create AI art, and no one wants to consume it. When one appreciates a poem or a painting or is moved by a song or a scene - very little of it has to do with just the rhyme scheme, the photorealism, a rich bass, or the amount of money put into VFX. Most of it is about being human.

The why of a poem is often a key part of the what. AI imitations through generative models fail to capture this aspect. We like to anthropomorphize AI and assign it sentient features - a human tendency since the days of Eliza - and we forget that these models have no agency and cannot ever create any meaningful art. Ironically, the engineers and scientists who put the data and training recipes together for these models - the closest thing to a collective agency such a model can possess - didn't have the domain expertise or ability to truly understand their own data, either. We've been led astray by our own nomenclature - generative models provide no value and inflict a lot of harm when generating poetry. That is clearly the wrong way to go about things.

Which brings me back to my own why for this blog. I'm trying to engage with more art - poetry, in my case - and understand how exactly the new age AI models can help creators and consumers of art - if at all. If I am convinced that a generative approach to poetry is the wrong way - what would I say is the right way? Does it even exist? I'd like to find out. And I'd like to share what I find helpful or interesting from this particular perspective. Nothing that follows would exist if I were on just one side of this debate - mistrusting and hating AI models with a blanket passion or creating and using them with a blanket indifference - but that's not who I am or where I am in my life.

Which brings us back to Adam Driver sitting on a bench in Paterson, New Jersey, looking all solemn and poetic and reciting Ron Padgett's lines with quiet sincerity. Would you rather be a fish? he asks.

As if the rest of the poem didn't have to be there.