Reclaiming Agency

So, what can we do? We can navigate these complexities by adopting clearer frames for understanding AI:

Tool-frame : See AI as a deterministic pattern machine—a sophisticated but ultimately mechanical process of pattern recognition and generation. When I ask ChatGPT to help me summarize research papers, I'm using it like an advanced text processor, not consulting an expert.

Partner-frame : Treat AI as an ergonomic dialogue aid—a tool specifically designed to enhance human thinking through conversation, but still fundamentally a tool rather than an agent. I find this frame useful when brainstorming ideas with Claude; the conversational format helps my thinking, but I'm aware I'm the only conscious participant.

Ecosystem-frame : Recognize AI as part of a corporate-state-user feedback loop—highlighting the broader system the AI operates within. When using ChatGPT, I'm not just talking to "o3" but engaging with a product shaped by OpenAI's business decisions, regulatory pressures, and user behavior patterns.

The goal isn't to reject AI but to engage with it mindfully in ways that preserve human agency and clarity. Here are four concrete practices I've found helpful:

Surface the wiring : I consciously apply a mental label: "This is a token-predictor trained on specific data, steered by particular human choices." Using "it" instead of personal pronouns helps reinforce this reality. I'm mindful of keeping a tool framing.

Create boundaries objects : I implement clear modes of interaction by forcing specific and clear frames. I do indeed share personal things with LLMs because they effectively search through semantic-space, but they carry no judgement of their own.

Deeply evaluate : Many models are far better than they were in the past, and still have significant flaws. The text is designed to be maximally believable, not maximally true. I critically evaluate output in ways I wouldn't to a person.

Preserve accountability : I never pass off AI-generated content to other humans without taking full personal responsibility for it, accept accountability for its quality and accuracy.

When engaging with AI, I regularly ask myself:

• "What function am I hiring this tool for?"

• "Am I surrendering agency or extending it?"

• "What system of meaning am I co-creating here?"

• "What are the implicit rules and expectations, and are they appropriate for this entity?"

What happens when you treat a calculator like a mathematical genius? Nothing much—you still get correct calculations, albeit with unnecessary reverence. But what happens when you treat an entity designed to mimic human connection as if it actually connects? You create a one-sided relationship where you bear all costs and responsibilities.

Andrew Conner • The Illusion of AI Personhood