Are We Outsourcing Too Much Thinking?

A question comes up at dinner — when a particular film came out, what a word actually means, how a law works — and a decade ago it might have sat there unresolved for an evening, maybe longer, prompting a small back-and-forth of guesses. Now someone reaches for a phone, or increasingly, types the question into an AI chat, and the answer arrives fully formed within seconds. The evening moves on. Nobody remembers, a week later, what the guesses were.

This is a small example of a much larger and genuinely open question: as it becomes easier to hand thinking to something else — a search engine, an algorithm, a language model — what happens to the thinking we used to do ourselves?

Cognitive offloading isn't new

It's worth being honest that outsourcing mental work is not a new phenomenon invented by AI. Writing itself was, in a sense, the first major cognitive offload — Plato worried, in the Phaedrus, that writing would weaken memory by giving people a place to store thoughts outside their own minds. Calculators offloaded arithmetic. Calendars offloaded remembering appointments. GPS offloaded the mental map-building that used to be a genuine skill for anyone who drove regularly.

In each case, something was lost and something was gained. Fewer people can do long division by hand or navigate a city from memory, but this freed up attention for other things. The pattern is consistent enough that it's tempting to assume AI is simply the next entry on the list, no different in kind from a calculator.

Every generation outsources something. The question is whether this generation is outsourcing something different in kind, not just degree.

What might be different this time

There's a reasonable case that large language models offload something further along the thinking process than earlier tools did. A calculator offloads a mechanical procedure with one correct answer. A search engine offloads finding information that already exists. An AI chat tool, used for drafting an argument, summarising a document, or working through a decision, can offload something closer to the actual formation of a view — the sifting, weighing, and provisional judging that used to require sitting with a problem yourself.

That doesn't automatically make it worse. Plenty of professional and creative work has always involved outsourcing parts of the thinking process to other people — editors, collaborators, research assistants — and nobody considers that a loss of cognitive capacity. What's new is the scale and speed at which one person, alone, can now do this for almost every kind of thinking task, with no other person involved and no visible cost.

What the evidence actually says so far

What this does not mean

None of this is an argument against using AI tools, and this isn't a piece that concludes with a warning to use them less. AI genuinely removes friction from tasks that didn't need to be hard — formatting a document, restructuring a messy first draft, checking a plausible-sounding fact. Treating every use of AI as a moral or cognitive compromise is its own kind of overcorrection, and not one supported by good evidence. The more useful question isn't "should I use this," but "what am I using this for, and is that the part of the task I actually wanted to keep doing myself?"

The distinction that matters: process versus outcome

A useful way to think about this is the difference between wanting an outcome and wanting the process that produces it. If you need a document formatted correctly, you almost certainly just want the outcome — there's little lost in outsourcing that entirely. But if you're trying to form a considered opinion about something that matters to you — a career decision, a difficult conversation, a piece of writing that's meant to represent your own thinking — the process of working through it is often part of the point. Skipping to the AI-generated conclusion can feel efficient while quietly removing the part of the experience that was actually valuable.

• Outsourcing formatting, transcription, or first-pass summarising: usually low-cost, high-benefit.
• Outsourcing the first draft of an argument you'll refine yourself: moderate — useful if you still do the refining.
• Outsourcing the entire formation of an opinion on something personally significant: worth noticing, at minimum.

In real life

Two people are each asked to write a difficult, honest message to a friend. One drafts something themselves, badly, crosses most of it out, and eventually lands on three sentences that are actually true to how they feel. The other asks an AI tool to write it, tweaks a couple of words, and sends it. The second message might read just as well, or better. But something about actually locating the true sentence — the discomfort of drafting it — was arguably part of what the task was for, separate from the quality of the final text.

Neither side of this argument is fully right

It's tempting to resolve this with a clean verdict — AI is either a dangerous crutch or a straightforward upgrade to human capability — but neither framing survives close contact with how people actually use these tools. Most people move fluidly between offloading trivial cognitive work (genuinely fine) and offloading meaningful cognitive work (worth noticing) many times a day, often without registering the difference. The skill worth building isn't AI avoidance or AI enthusiasm. It's the habit of briefly asking, before reaching for it, what kind of thinking this particular moment actually calls for.

A thought to keep

Not every unfinished thought needs finishing by something else. Some of them were doing something useful precisely by staying yours, and a little unresolved, for a while longer.