Is AI Making Us Smarter, or Just Better at Outsourcing Thinking?

Is AI Making Us Smarter, or Just Better at Outsourcing Thinking?

AI and human cognition

The brain has always outsourced things. We stopped memorizing epic poems once we had books, stopped doing long division once we had calculators, stopped navigating by stars once we had GPS. Each time, someone worried we were getting dumber. Each time, we just got good at something else. The question now is whether generative AI is a bigger version of that same trade, or a different kind of deal entirely, one where what we’re handing over isn’t a task but a habit of mind.

The Memory We Used to Keep in Our Heads

Phone numbers used to live in your head. Addresses, birthdays, the plot of a book you finished a month ago. Most of that has already moved to a device, and nobody mourns it much. What’s newer is handing over the first draft of a thought itself, not just the storage of facts but the shaping of an argument, a eulogy, an apology text.

Cognitive scientists call the phone-number shift “transactive memory,” the same mechanism that lets a couple divide who remembers the dentist appointments and who remembers the wifi password. AI expands that system from a handful of people you trust to a tool that never forgets and never gets tired of being asked. The difference is that a spouse pushes back. A chatbot, by design, does not.

That matters because memory was never just storage. Recalling something, struggling for a word, reconstructing an event from fragments, those are small workouts for the same mental muscles used in problem-solving generally. Skip the workout enough times and the muscle doesn’t vanish, but it does get softer.

When ChatGPT Becomes Your Second Brain

There’s a particular feeling of relief in typing a half-formed thought into a chat window and watching it come back organized, articulate, done. It’s useful. It’s also a little bit like hiring someone to lift weights for you and expecting to see results at the gym.

The “second brain” framing is popular because it sounds empowering, an extension rather than a replacement. And for some tasks that’s exactly right: drafting a meeting agenda, summarizing a dense report, checking a formula. Nobody’s cognition is enriched by manually formatting a spreadsheet, so automating it frees up attention for things that actually need a human brain.

The trouble starts when the second brain quietly becomes the first one. Researchers studying knowledge workers have noted a pattern where people stop generating their own first draft of an idea at all, going straight to the model and then editing its output. The editing feels like thinking. It isn’t quite the same thing, because editing works from a menu of options someone else already chose, while generating starts from a blank page where the options are still yours to invent.

Learning at Machine Speed and What Gets Lost in the Shortcut?

Ask an AI to explain quantum entanglement, the French Revolution, or how mortgages work, and you’ll have a competent answer in seconds. Compare that to the old route: a textbook chapter, a confusing lecture, maybe three Wikipedia tabs and a YouTube video before it clicked. The old route was slower and also, frustratingly, often stickier.

Learning speed and learning depth aren’t the same metric, even though they get treated as if they are. Desirable difficulty is a real concept in education research: material that’s slightly harder to acquire tends to be retained better, precisely because the brain has to work to encode it. An AI answer removes the friction, and the friction was doing something.

None of this is an argument against AI tutors, which can be genuinely good at meeting a learner where they are rather than where a standardized curriculum assumes them to be. The site has covered that ground in more detail in the piece on AI in personalized learning, which looks at how adaptive systems can tailor pace to the individual rather than the classroom average. The shortcut isn’t the enemy. Mistaking the shortcut for the destination is.

Can an Algorithm Really Simulate Empathy?

Empathy gets described as if it’s a single skill, but it’s really two things stitched together: reading what someone else feels, and caring that they feel it. AI models have gotten startlingly good at the first part. A language model can detect frustration in a customer’s phrasing, grief in the structure of a message, sarcasm buried in a complaint, often faster and more consistently than a tired human reading the same text at the end of a long shift.

What it can’t do is care, because caring requires something to be at stake for the one doing the caring. A model producing a comforting message isn’t relieved when the comfort lands, and it doesn’t carry the conversation with it afterward the way a person does. The output can be indistinguishable from empathy on the page. The experience behind it isn’t there, and that gap doesn’t show up in the transcript.

The interesting risk isn’t that people will be fooled by fake empathy. It’s that outsourcing the emotional heavy lifting of a hard conversation, drafting an apology, writing condolences, responding to a friend in crisis, might dull the muscle that notices someone else’s pain in the first place. Noticing is a skill too, and skills atrophy the same way memory does, through disuse rather than damage.

Neuro-AI Interfaces and The Leap from Assistant to Extension

Everything above describes AI as a tool sitting outside the skull, something you open and close like an app. Neural interfaces are the development that collapses that distance. Brain-computer systems that read intention directly from neural signals, already working in limited medical contexts for people with paralysis, point toward a future where the line between “I thought it” and “the device thought it for me” gets genuinely blurry.

That’s not a far-future hypothetical confined to labs. Work on decoding neural signals into language is already active enough to have its own research threads, including efforts to translate brain activity into sign language output for people who can’t vocalize, covered in more depth elsewhere on this site. The leap from assistant to extension isn’t a single dramatic moment. It’s a series of small integrations, each one reasonable on its own, that add up to a different relationship between a person and their own cognition.

The philosophical question underneath all of this is where the self ends. If a note-taking app holds your memory, most people agree you’re still you. If a neural implant holds your intention before you’re consciously aware of forming it, the answer gets murkier. Nobody has a clean definition of the boundary, and the technology isn’t waiting for one to be settled.

The Cognitive Trade-off Nobody’s Pricing In

Every technology that offloads a mental task is, in effect, a trade: time and effort saved now, in exchange for a capacity that gets used less and therefore weakens. That trade isn’t inherently bad. Few people lament losing the ability to calculate a square root by hand. But it’s a trade that should be made on purpose, with some sense of what’s being given up, rather than by default because the option was there and convenient.

Right now, that pricing isn’t happening. Nobody opens ChatGPT to a dialog box that says “this will save you four minutes and cost you a slightly weaker first draft of your own opinion.” The convenience is visible. The cost is diffuse, showing up months later as a vague sense that writing something from scratch feels harder than it used to, or that holding a complex argument in your head for a full conversation takes more effort than it did.

The people who seem to navigate this best aren’t the ones avoiding AI, they’re the ones using it deliberately: letting it handle retrieval and formatting while keeping the generation, the first attempt, the rough draft, as their own. That’s a discipline, not a default setting, and defaults are what most people actually live by.

So the honest answer to whether AI is making us smarter or just better at outsourcing thinking is probably both, applied unevenly, to different people, for different tasks, without anyone keeping score. The brain doesn’t send a notification when a skill starts to fade. It just quietly stops being there when you reach for it.

If this got you rethinking what your own brain still does unaided, check the AI and cognitive science reading picks linked through this piece, and then ask yourself: what’s one mental skill you’ve quietly handed over to AI this month?

Thank you for questions, shares and comments!

Share your thoughts or questions in the comments below!