I have been thinking about television lately, which is not something I expected to find myself doing.

Not television as it exists now - fragmented, on-demand, somehow both everywhere and nowhere - but television as it was imagined at the beginning. There is a striking piece of archival footage of David Sarnoff, the RCA executive who helped launch NBC, describing his vision for the medium. He talked about it as a device for bringing the world’s great performances into the home. Symphony orchestras. Theatre. Educational lectures from the finest minds. The television set as a window onto human excellence, available to everyone regardless of geography or class.

That is not what happened, of course. What happened is that the incentives of mass broadcasting - the need to hold the largest possible audience for the longest possible time - gradually shaped television into something closer to a sedative. Not because anyone sat in a room and decided to lobotomize the public, but because a passive audience is a larger audience, and a larger audience is more money. The medium selected for content that required nothing of the viewer. Over decades, “watching television” became synonymous with not quite being present in your own life.

I notice this happening again, with AI, and I find myself trying to figure out which side of the transformation I am on.

The distinction that keeps forming in my mind is this: there are two fundamentally different ways to use these tools. The first is consumptive - you come with a problem, you receive an output, you leave. The AI does the thing. You are the recipient. The second is something more like collaboration toward expression - you use the tool to think out loud, to push on half-formed ideas, to articulate something you already partially know but cannot quite reach. In the first mode, the AI completes you. In the second, it extends you.

But I have come to distrust the sharpness of that binary. The more honest framing is about balance and, more importantly, about clarity - knowing, at any given moment, which mode you are in and why you chose it.

The difference sounds philosophical until you sit with what each one does to you over time.

I have noticed, for instance, that when I use AI to write something I could write myself - an email I am avoiding, a summary I am too tired to produce - there is a small but real erosion. Not of capability, exactly, but of engagement. I have offloaded not just the effort but the thinking, and the thinking is the part that changes you. Writing a difficult email forces you to figure out what you actually want to say, which forces you to figure out what you actually think, which occasionally reveals that what you thought you thought is not quite right. Bypassing that process is not just laziness. It is a small act of self-avoidance.

On the other side: the conversations where I use AI to chase an idea down are some of the more genuinely useful intellectual experiences I have had. Not because the AI is always right - it often isn’t - but because having something respond to your half-formed idea, even imperfectly, forces you to sharpen it. You hear your own idea echoed back, slightly distorted, and suddenly you know where it was vague. The friction of the exchange is productive. You come away with something you did not have before, and it is yours.

I notice this in how I use AI for the writing on this site. The posts here are not things I did not know - they are things I knew in the way you know something you have been living with for years: partially, messily, in fragments. Notes that only I could fully decode. Intuitions I could not quite hand to someone else. What AI helps me do is not generate the ideas but articulate them - take the shape of something I already see and render it in a form that other people can actually follow. That feels like expression to me. The ideas are still mine. The expansion is mine. The AI just helps me build the bridge between what I understand and what someone else can.

That is a legitimate and genuinely useful mode. But I notice it works only because I already know what I am trying to say. If I did not - if I were using AI to supply the ideas rather than to articulate ones I already hold - the result would look similar from the outside and feel completely hollow from the inside.

This maps onto something I have been thinking about with expertise more broadly. There is a version of delegation that makes you more capable - you hand off the parts that do not require your particular attention so that you can go deeper on the parts that do. A good editor does not make a writer worse; they free the writer from certain problems so the writer can focus on what only the writer can do. And there is a version of delegation that makes you less capable - you hand off the parts that would have built your understanding, the parts that, had you done them yourself, would have left you more able the next time. The question is whether you are delegating the execution or the thinking.

For trivial things - the boilerplate, the scheduling, the conversion of a format into another format - consumption is fine. More than fine; it is sane. The human brain is not well deployed on tasks that do not require it. Delegation here is just efficiency, and feeling guilty about it is a form of self-flattery. But as the tasks grow in importance, as they get closer to the things that actually develop you, the calculus changes. You cannot delegate your way to understanding something. You can only fake it.

The worry is not, I think, the one people usually reach for - that AI will replace human creativity, that we will become redundant. The more plausible danger is subtler: that we will choose replacement because it is easier. That we will drift, gradually, toward the consumptive mode not because it is imposed on us but because it requires less of us. Every convenience that saves us effort also saves us from the growth that effort produces. Most of the time that trade is worth making. With writing, with thinking, with the particular kind of struggle that leads to genuine understanding - I am less sure.

There is a specific failure mode I am watching for in myself. AI is very good at producing outputs that look right. Fluent, coherent, credible. And fluency is seductive - it creates the feeling of understanding even when understanding is absent. I can read an AI’s explanation of something I do not fully grasp, nod along, and come away with the impression that I now know it. That impression is wrong. The explanation was not mine. The comprehension was performed, not achieved. And the gap between performed comprehension and real comprehension tends to only reveal itself at the worst possible moment - when something goes unexpectedly wrong and you realize you never actually understood what you were doing.

I do not think the answer is to never let yourself not understand things. That is paralysis dressed up as rigor. It is okay to operate in areas where your understanding is partial. But it should be a conscious choice, held lightly, with the intent to close the gap - not a laxness that sets in so gradually you stop noticing it. The direction of travel matters. You can be outside your understanding and still be leaning in. Or you can be inside your understanding and quietly drifting out. AI, used without attention, accelerates the second.

Television did not become a sedative through malice. It became one through the path of least resistance, scaled across millions of choices made by people who were tired and wanted to rest. The medium was not the problem. The problem was the incentive structure that shaped what the medium became, and the ease with which passive consumption filled the space that active engagement might have occupied.

AI is not television. The comparison only goes so far. But the pattern - a technology that can extend human capacity toward something genuinely expansive, gradually captured by the pull toward passivity because passivity is so much easier - feels worth naming. Worth watching for in myself.

The equilibrium I am trying to hold is not “use AI for expression, never for consumption.” That would be needlessly ascetic and probably false. The equilibrium is more like: use it for consumption where consumption is appropriate, use it for expression where expression matters, and stay clear enough about which you are doing that the two do not quietly swap places. The danger is not the consumptive mode. The danger is using the consumptive mode and telling yourself it is expressive. That is the version where you lose the thread of what you are actually doing, where the output drifts from what is yours, where you look up one day and cannot quite locate yourself in the work.

I do not have a solution to this. I am not describing a discipline I have mastered. I am describing a tension I have noticed, between the version of myself that uses these tools to go further and the version that uses them to go nowhere faster. Most days both versions show up. The question is which one I am consciously choosing, and whether I am paying enough attention to know the difference.

The television, in most homes, eventually moved from the living room to the bedroom. A smaller screen, closer to your face, the last thing you saw before you fell asleep. Progress in the direction of frictionlessness. I think about that sometimes when I reach for the AI before I have tried thinking first. Not because reaching for it is wrong. But because trying first is how you know what you actually think.