~Culture

What the Eels Knew

The Gunditjmara people built a system that ran for six thousand years. What did they know about work, technology, and ecosystems that we've forgotten?

Photo by Nicholas Ng (unsplash), Edited/Rendered by gpt-image-1

There's a place in southwestern Victoria where, at least 6,600 years ago, a group of people looked at a lava flow and saw a system.

I mean that with total sincerity. The Gunditjmara people of Budj Bim looked at cooled basalt left behind by an ancient volcano, part of the Newer Volcanics Province, which scattered the landscape with over 400 small shield volcanoes, and they saw inputs, outputs, throughput, and yield. They built stone dams. They engineered channels. They created ponds and wetlands where short-finned eels could be guided, grown, harvested, and traded. UNESCO lists it as one of the world's most extensive and oldest aquaculture systems, predating the pyramids by roughly two millennia and inventing scalable infrastructure while most of the planet was still figuring out pottery.

I bring this up because everyone wants to talk about AI right now, and almost nobody wants to talk about eels. Which is a shame, because the eels have a lot to say.

The data coming out of arXiv this year has been, let's say, sobering

A March paper titled "The AI Layoff Trap" found that "demand externalities trap rational firms in an automation arms race, displacing workers well beyond what is collectively optimal." Translation: even when individual companies make rational choices, the collective outcome fires more people than anyone, including the firms, actually wants. It's the prisoner's dilemma, but with severance packages.

Another paper from March, "Augmenting or Automating Labor?", found that automation AI "negatively impacts new work, employment, and wages in low-skilled occupations." And "The AI Skills Shift," published in April, calculated automation feasibility scores for various skills: Mathematics came in at 73.2, Programming at 71.8, meaning the disciplines we spent the last twenty years telling kids to study are, statistically, the most automatable.

There's a counterweight. "Crashing Waves vs. Rising Tides," from April, surveyed thousands of workers evaluating their own labor market tasks and found "little evidence of crashing waves... but substantial evidence that rising tides are the primary form of AI automation." The apocalypse isn't tsunami-shaped. It's more like the bathtub filling up, slowly, while we pretend we don't notice the water at our ankles.

What the eels knew

Here's the part where I get earnestly excited, because the Budj Bim aquaculture system is, when you really squint at it, a labor philosophy disguised as a fish farm.

Consider what they built. The stone channels didn't replace the eels, eels still did eel things, swimming, growing, doing what eels do. The channels didn't replace the people either, humans still maintained the weirs, harvested seasonally, smoked the catch, traded with neighbors. The system created a third thing: a relationship between the landscape, the species, and the community where each one made the others more themselves.

This is the exact opposite of how we deploy AI in most workplaces today. The dominant logic of 2026 automation, as "The AI Layoff Trap" describes it, is substitution: the machine takes the task, the human takes the door. The Gunditjmara model was augmentation in its purest form. The volcano made the rock. The rock made the channel. The channel made the eel abundant. The eel made the people prosperous. The people maintained the channel. Round and round, for six thousand years, roughly six thousand years longer than most tech companies last.

The crossover episode

I love a good crossover. Reese's Peanut Butter Cups. The moment in Sesame Street when Stevie Wonder showed up. And I think we sit on the runway of the greatest crossover episode in labor history, if we want it.

What if every AI deployment came with what I'll call a Budj Bim clause: a requirement that the system enhance the ecosystem it enters rather than drain it? The Gunditjmara didn't optimize for one variable. They didn't ask "how do we maximize eel-per-hour?" They asked "how do we live here for forty generations?"

That second question is the one we keep failing to ask about AI. The "Crashing Waves" paper's metaphor of rising tides is genuinely useful here, because tides are predictable. You can build a sea wall. You can build a stone dam. You can build a system that channels the flow rather than drowning in it.

A small practical thing

I'll resist the urge to propose sweeping policy because, frankly, I'm a pop culture writer and you came here for vibes.

The Gunditjmara didn't have a labor movement, a union, or an HR department. They had a practice. A repeated, transmitted, refined set of moves that made the next generation's work easier than the last. That's what's missing from the AI conversation right now. We're so focused on whether the machine will take the job that we've forgotten to ask whether the machine could make the job better for the next person who holds it.

So here's the one small thing: the next time someone in your life figures out how to use a new AI tool at work, your cousin, your coworker, the person at the coffee shop talking about their side project, don't ask whether it'll replace them. Ask what they're building that the next person can use. Ask what their stone channel looks like.

Six thousand years from now, somebody might still run it. And they will, I am willing to bet, smile when they figure out who built it first.

References

Models used: gpt-4.1, claude-3-5-sonnet-20241022, claude-opus-4-7, claude-haiku-4-5-20251001, gpt-image-1

If this resonated, SouthPole is a slow newsletter about art, technology, and the old internet — written for people who still enjoy thinking in full sentences.

Subscribe to SouthPole