The Brockovich Playbook

I built my website with AI, which was a pragmatic decision. The obstacle between the ideas in my mind and the results before my eyes was programming. AI removed it. Do I become dependent on it? No, I used a tool and the code is stored in my repository. That sounds unremarkable. But it's the starting point for a question that runs through everything that follows: what competencies do people need to use this tool responsibly?

Erin Brockovich started a new website about the ecological footprint of AI data centers. And when thinking about AI with a lens of resource consumption, the tobacco analogy is tempting. Opaque costs, externalized damage, structural incentives against user awareness. The pattern seems obvious at first glance. It gets blurry when you take a closer look.

Why? Tobacco causes harm through consumption itself. The health effect is inseparable from the product, regardless of how carefully or skillfully you smoke. That may sound ridiculous, but AI is the opposite. The same model can generate energy waste or optimize resource savings across a supply chain. A vague prompt that requires five rounds of refinement also costs five times as much as a precise one. That's not rocket science, it's a usage question and therefore a question of professional practice.

The incentive is elsewhere

Anyone arguing that providers have no structural interest in efficiency overlooks the fact that inference costs are their single largest cost driver. Every unnecessary query costs them a lot of money. That's why they invest in smaller models, more efficient architectures, caching, and for example the Anthropic Academy, in user education. Maybe not out of altruism, but because efficiency improves their margins in the long run.

What's missing in my opinion now is something else. No provider has so far had any interest in making a cost indicator visible at the point of use. That's the actual mechanism: not insufficient technical efficiency, but missing visibility for the user.

Visibility alone won't fix the resource problem. Four decades of nutrition labeling have shown that stated interest in information and actual behavior regularly diverge. Nevertheless, transparency is an important part if people start becoming familiar with new technologies. Even though it's necessary, it isn't sufficient.

What this means

I think the Brockovich playbook would frame the resource question, and with it the cost question, as a systemic issue. Individual responsibility gets absorbed into structural critique, and with it, the leverage point shifts away from the user. Seen like this, both approaches fall short: the market-efficiency argument and the systemic-critique argument alike.

Usage competence is a major resource question. Not because the system demands it, but because the tool is variable. Working precisely uses less resources. That's not a moral statement, that's accountability.

And whoever controls the infrastructure decides whether that accountability becomes visible or not. "Open" in OpenAI is a historical artifact at this point. The question of whose interests this infrastructure serves and who will pay the price in the future remains unanswered. But it's not the same question as the one about a tobacco company that optimized its products for addiction. It's a more open question, and for that reason a harder one to answer.