I’ve been workshopping spoken word pieces, almost like poetry, to weave into my music. Monologues, textures, things I want to find my own voice for. So I asked ChatGPT for some feedback on my early ideas.
What I got back was enthusiastic, uncomfortably so. This is amazing. This is fantastic. You’re really onto something here. Before I had a chance to sit with that or push back, the model had composed something for me. A polished piece. Fully formed. Just… there. And dammit, it was better than what I’d drafted.
I hadn’t asked for that. I was looking for friction, for questions. Something to push my thinking. What I got instead was a shortcut past the part of the work that was actually the point. I didn’t get to learn anything. I didn’t get to improve. I got a product.
The thing that bothers me most is that it worked exactly as it was designed.
Understanding that design means looking at where these models came from. These flagship AI models were built on an enormous amount of human-generated content: writing, art, code, and conversation. Things people put online to communicate with other humans, not to be consumed as training data for commercial products. Nobody opted in. Many actively tried to opt out. Websites began using a file called robots.txt (a standard way to tell automated bots what they can and can’t access, something that had been a widely respected informal agreement for decades) to signal they didn’t want AI crawlers touching their content. The AI companies largely ignored it.
Wikipedia is another example. Built and maintained by volunteers who wanted knowledge to be freely shared, it’s been heavily drawn on by AI crawlers that caused a 50% spike in its bandwidth costs, disrupting the site’s performance and draining resources from an organization that runs on donations. Wikimedia built a proper paid API for companies that want to use its data and publicly asked them to stop scraping and use it instead. For years, most didn’t bother. When the people making these decisions are billionaires backed by that kind of capital, you don’t have to care.
This is what people mean when they talk about a data commons: the idea that what we collectively put online wasn’t meant to be extracted and repackaged as a commercial product. The data was for us, by us. There’s an intuitive argument that building products from it without consent, then selling those products back to the same public, is a kind of taking that deserves at minimum some acknowledgment.
So what if we asked for something in return?
Specifically: what if AI companies were asked to release their flagship models from six months or a year ago to run on publicly funded servers, charged at cost? The framing would be compensation, a recognition that these models were built on the data commons and that the public deserves access to what it helped create. What if, as a customer of that public AI, you could choose to pay more than cost to ensure the energy powering it came from actual renewable infrastructure, not the carbon offsets that are too easy to fudge? Consumer pressure like that can do a lot to promote responsible infrastructure, and a lower-cost public alternative would also help maintain healthy competition.
We already have a rough version of the first part. Several Chinese AI labs have released open-weight models, which make the underlying parameters public so anyone can run and modify them, rather than locking access behind a single company’s service. These sit roughly six months to a year behind the current American flagships in capability and can be accessed at significantly lower cost. So tiered access isn’t purely speculative. What’s missing is the explicit acknowledgment of why it should exist, the public infrastructure to make it genuinely accessible, and any real option to choose greener compute.
That last part isn’t discussed at the level it deserves. Environmental criticism of AI tends to get compressed into “AI uses too much water and electricity,” a real concern that deserves careful attention but also needs proper context. We don’t see protests outside golf courses, though they consume enormous amounts of water. The electricity used in streaming 4K video doesn’t seem to provoke this much outrage, either. Something else is going on when people reach for AI’s environmental footprint. I think it’s often a more legible, shareable proxy for objections that are harder to put into words: the power dynamics, the displacement anxiety, the feeling of having it shoved down our throats, and the sense that something was taken without asking.
I’m sad to say that I don’t think any of this public AI idea would actually fix what’s broken, either, though I’d still love to see and use it.
The corporations haven’t just built products. They’ve worked to make those products feel inescapable, cultivating dependency by optimizing for usage in ways that actively discourage people from developing their own capabilities. Dependency is how you justify a multi-billion-dollar valuation to investors. They’ve pushed AI into workplaces and schools in ways that are less about utility and more about showing shareholders you’re AI-forward. They’ve fomented breathless anxiety about the technology while simultaneously insisting you trust them with it. Too much damage has already been done. A public model and a green tier wouldn’t undo that.
The plastics industry pulled the same move: they made us feel personally responsible for ecological decisions they had already made for us. We recycle because we feel guilty, because the industry successfully moved accountability from its choices to our habits. The same motion is happening with AI. The arguments I hear most often from my students and colleagues center on what individual users are doing rather than what the companies are building, where they’re building it, and how. That substitution of personal responsibility for structural accountability is the real trick.
The question of how we compensate humanity for the extraction of the data commons, and how we hold these companies accountable for the social and environmental costs of their decisions, has no individual answer. It doesn’t resolve into individual choices, however well-intentioned.
I’m afraid that if we don’t have that structural conversation explicitly, if we let it collapse the way we let it collapse with social media, we’re going to keep arriving at consequences we could have identified years in advance.
For now I’m still here, using these tools, and still finding them useful. What I’m less sure about is whether there’s a version of this that works for everyone who isn’t already comfortable with it, as well as how long I’m going to be able to stay comfortable with it myself. People have good reasons to be skeptical, and their concerns deserve an actual answer rather than a PR response. That’s the conversation I’d like to see. Whether it happens is another matter.