The Choosing Human

There was a moment at a university dinner recently when someone asked what was left for graduates in an age when fluent expert articulation is available on demand. What came out was less about my work as a software producer, and more from my life as it’s been lived: our value today is in knowing who we are, and exercising our right to make choices.

I have been thinking about that sentence since then. It is the answer to a question the AI industry has been asking the wrong way around.

What remains when the tools can articulate anything

The industry's pitch is that these tools augment your knowledge. They give you eloquent responses for any imaginable prompt. That pitch is true as far as it goes. It also positions the human as a knowledge-deficient creature being topped up by the tool, and that framing is doing damage that’s just starting to enter public discussion.

The inversion is this: AI tools don't make you valuable by adding knowledge to you. They make your value visible by stripping the knowledge problem away.

Now that anyone with internet access has fluent expert articulation on the growing library of anything, the differentiator that mattered for most of human history, knowing things other people didn’t know changes. What remains, underneath, is what was always doing the work. The chooser. The self that points the knowledge at the problem, decides what matters and knows when to refuse the answer's framing entirely.

That self was always doing the work. The knowledge was the wrapper. The wrapper was so consistently present that we mistook it for the thing itself. The tools have commodified and changed the wrapper and what's left is the part that was always actually valuable.

The wrapper and what's underneath

For most of human history, knowing things other people didn't know was a reliable form of value. Doctors knew medicine. Lawyers knew law. Scholars knew texts. Software engineers knew the languages and patterns of their craft. Each profession was a body of expertise that had been slowly accumulated and expensively learnt.

Underneath the expertise, in every case, was something else. The doctor's judgement about which question to ask first. The lawyer's sense of which argument would actually land in this courtroom, in front of this judge. The engineer's choice of which problem to solve and which to leave alone.

The expertise was visible and the choosing underneath it was invisible because it was bundled in with the expertise so there was no need to distinguish between them. You couldn't have a doctor without medical knowledge, so the question of what the doctor was doing apart from the knowledge rarely got asked.

The tools we have today now force this very question to be asked. They deliver the knowledge layer, fluently, on demand and customised. The question of what the human is contributing apart from the knowledge is now the question of our time.

Self-knowledge as the prerequisite

Choice without self-knowledge is reactivity. To choose meaningfully, you have to know yourself. What you value. What you want. What you can live with. What you cannot.

This is one of the oldest documented requirements. Every wisdom tradition that has thought about how to live well has put self-knowledge near the centre. The Delphic know thyself. The interior accounts of the Psalms. The Stoic practice of examining one's responses. The contemplative traditions east and west that train attention back toward what is actually happening in the person, beneath what they say is happening.

You can check this against your own experience right now. As you read this sentence, multiple things are happening to you at once. A thought about what I just wrote. A feeling about whether I got it right. A separate awareness of your body: tired, comfortable, restless, hungry. A half-formed response forming itself into words. A memory this article has triggered. An impulse to push back, or to agree, or to put the screen down for a minute. They are all there.

What you share of all that, to anyone, is only a tiny fraction. The choice of which fraction is the act that constitutes your interior as yours. The unshared things are the texture of the inner world that the shared things are selected from. The selecting is the work. The selecting is the self.

This is what having an inner world means, in the most ordinary sense. A constant, moment-by-moment surplus of thought, feeling and bodily signal, from which you choose what to give shape to and what to let pass. An AI tool has no version of this. It has no inner concurrence to choose from. There is no thought it had and decided not to share, or a feeling it weighed against another, or a memory it noticed and let go.

The tools threaten self-knowledge in a way that is easy to miss because the threat is so gentle.

They offer fluent articulations of any position. Pick a question, such as “What should I do about my career?” or “What do I think about this political issue?”, and the tool will produce a coherent-sounding, well-reasoned answer. The answer sounds like something a thoughtful person might say. See my article on the History of AI, for why it works like this.

What's gone, in that moment, is the task of working out what you actually think. The articulation arrives so fluently that the friction that used to be the price of arriving at a view: sitting with the question and noticing your own resistance can be easily skipped. You ended up with a position. You didn't end up with your position.

This is already happening, at scale, in the way people are forming views on every question the tools will answer. The self-knowledge erodes silently because nothing dramatic is being lost in any given moment.

The defence is an ancient one. Know yourself. Spend time with the question before reaching for the tool. Notice when an articulation feels right because it fits you, versus when it feels right because it is well-written. The two feelings are different and the tool only operates constantly in the second one.

Choice as the expression

Unexercised self-knowledge is just introspection. It is through our acted-out choices that the known self enters the world.

The tools threaten choice from a different angle than they threaten self-knowledge. They threaten it by offering recommendations on everything, and the recommendation is usually plausible enough that going against it feels unjustified.

Why override the tool? It has more information than you. It has read more about your problem than you have. Its answer is probably right, or at least more right than yours probably would have been.

The default becomes acceptance and acceptance becomes the shape of the human relationship to decisions that used to be theirs. The chooser steps out of the seat because staying in the seat now feels like an indulgence - a refusal to use the best available input.

What gets lost in this is the practice of choosing. The capacity to choose is maintained by exercise. Each decision retained is a small reinforcement. Each decision delegated is a small surrender of the muscle that holds the self together. Stop choosing, and the self atrophies because the activity that constituted it has stopped.

The stakes here are in our own personhood. A person is partly constituted by the pattern of choices they make over time. The shape of a life is the shape of its choosing. A life of accepted recommendations is a different life from a life of made decisions, even if the outcomes look similar on the outside. The interior is shaped by who is doing the choosing.

The skill of today is staying in the seat where the question of accept-or-not is being asked. The moment I stop asking, and start accepting by default, my active choosing muscle begins to weaken and by the time I notice, I may have built something that doesn’t reflect a true expression of my own choices.

Skill with these tools

Skilled users of these tools treat the output as their own input material.

Unskilled use of these tools treat the output as the answer.

The deeper skill is knowing what you want and knowing when the output is right for your purpose and when it isn't, being willing to refuse something fluent and ask for the awkward thing instead.

I use these tools frequently. I find them both a joy and a pain to use at times. They enable me to create work I could not otherwise do. They give me drafts to react against, options to choose among, articulations of positions I then have to decide whether I actually hold. The tool externalises the knowledge work so I can spend my capacity on the choosing work.

The difference between good and bad usage of these tools is not visible from the outside. Two people can be using the same tool for the same task, and the work coming out can look superficially similar. The difference is in who is in the chooser's seat. One person is using the tool to produce material they then evaluate against their own judgement. The other person is accepting the tool's material and shipping it. The work in the first case improves over time. The work in the second case converges toward the grain of the current models.

The more time you spend around these tools the more you can see the grain when you start looking for it. A significant amount of the marketing arriving in my inbox now reads as if written by the same hand. There are tells that change as each model improves. The grain of the model is visible in writing produced by people who have never met. Writers who use these tools without staying in the chooser's seat lose the thing that made their writing theirs, including at times, ironically, my own.

The skill is to use the tool while keeping the work yours. That requires the same capacity at the centre of everything in this article. Knowing what you think and choosing what to keep.

What the tools cannot reach

In The Person Behind the Data, I argued that the industry's premise, that AI could build a stable model of your inner life from your data, is a category error. Lisa Feldman Barrett's work on constructed emotion shows that there is no fixed inner state for the tool to model. Emotions are constructed in the moment from interoceptive signals, history, context, and what the individual’s human system is predicting. The same situation produces different feelings on different days. There is no this is how I always feel about this, because the feeling is built fresh each time, in the body that is having it.

The same logic applies to choice.

Choice is not a stored value the tool can retrieve. It is an act performed by a self that is partly constituted by performing it. The tool can offer options. It can describe the considerations. It can predict what someone like you might do. What it cannot do is be the one in the seat where the act of choosing happens, because being there requires a body that has stakes in the outcome and a self that knows what it is choosing for.

This is why the tools do not replace humans, even as they replace huge volumes of the work humans used to do. The work they replace is the wrapper work. The thing underneath, the choosing, by a self that knows itself, is the thing they cannot reach, and it is the thing that was and is always the actual value.

The threat is not that the tools will become human. The more pressing and genuine threat is that humans will stop being human, because the tools decreased the friction of choosing.

The human stays in the seat

The practical question for anyone using these tools is the same question, asked in a hundred small forms throughout the day.

Am I in the seat where the choosing is happening, or have I stepped out of it?

The tool produces a draft and you ship it without reading it carefully. The tool recommends an approach and you take it without considering alternatives. The tool articulates a position and you adopt it without working out whether it’s actually yours. Each of these moments is an abdication of the self - a diminishment.

Staying in the seat is also undramatic. You read the draft and decide what stays. You hear the recommendation and decide whether it fits the situation as you understand it. This is harder than it sounds, because the tools are designed to make stepping out feel reasonable. They are fluent and helpful. It’s hard to choose to continue staying in the seat.

I wonder if the ways we work today will be sustainable for the future. When I know myself well, choosing is easy, and when I’m unsure choosing is one of the hardest things to do.

When we avoid choosing, we don't just skip a decision, we skip the act through which the self is formed. The cost isn't immediate. It shows up later, in the gradual erosion of the inner capacity that choosing used to exercise and sustain.

What this asks of us

We are at the beginning of a generational shift in what human value is for. The tools haven’t created the shift, choosing was always the core, knowledge was always the wrapper. The tools make the wrapper visible as a wrapper, and force the question of what’s underneath.

For most of human history, one could be valuable by knowing things. That era is ending. The thing that remains is what was always underneath. The self that knows itself, and the choice that oneself makes.

This asks something of us that the previous era did not. It asks us to do the inner work that the knowledge layer used to make optional. It asks us to know who we are, separate from what we know. It asks us to choose, actively, in moments where accepting a recommendation would be easier. It asks us to maintain the capacity that constitutes us, by exercising it.

It also offers something the previous era did not. The work of being a person, which was always the most important work, is now visibly the work. The wrapper has been pulled away. What we do with the chooser's seat is no longer hidden inside our profession. It is the thing itself.

The tools will keep getting better at the wrapper work. They will keep producing more fluent, more capable, more useful articulations of expert knowledge. They will keep diminishing the differentiator that used to come from knowing things.

What they will never do is be the one in the seat where the choosing happens. The seat is ours, and most of the incentives are arranged to coax us out of it, and there's no guarantee which way it goes.

Part three of a three-part series on today's AI systems and what they mean for how we work. A History of AI as Manufactured Expert Judgement traces how these models got here — through the systematic encoding of expert human judgement into model weights, rather than through intelligence emerging from scale. The Obedient Chooser argues that intelligence has two halves — knowledge and choice — and that what these systems do, however fluently, is only the first. This essay argues the inverse: that we are valuable because we have what these systems structurally lack, and the era of fluent expert articulation on demand makes that value visible rather than diminishing it.