Copilot Means Co-, Not Auto-

Two intelligent people open the same AI tool.

The first asks it to write an email. Then another. Then a reply to a friend. Soon, three ordinary sentences feel strangely difficult without a machine standing beside them, holding the linguistic bicycle steady.

The second writes the first draft, asks AI to expose its weak assumptions, interrogates the feedback, tests three alternative approaches and then rehearses the difficult conversation that follows.

Same technology.

Completely different cognitive outcome.

One person is outsourcing their ability.

The other is expanding it.

So the question is not whether AI will make humanity smarter or dumber.

It is: After using AI, are you more capable without it — or merely more productive while it is present?

Yes, AI can make us faster— and substantially better

We should start with the truth rather than another fear sermon.

AI can increase the speed, volume and quality of certain kinds of work.

In a preregistered experiment involving 758 consultants, participants using GPT-4 completed 12.2 per cent more tasks, worked 25.1 per cent faster and produced higher-quality work on tasks that fell within the model’s capabilities. However, on a task outside that “jagged technological frontier,” AI users were 19 per cent less likely to reach the correct answer.

That is AI in miniature.

Brilliant inside the right zone.

Persuasively wrong outside it.

A carefully designed AI tutor has also produced striking results. In a randomised crossover study of 194 Harvard physics students, median learning gains with a structured AI tutor were more than twice those achieved through an in-class active-learning lesson, while students spent a median of 49 minutes with the tutor rather than 60 minutes in class. The important phrase is carefully designed: the tutor was explicitly engineered around active learning, scaffolding, accurate worked solutions and timely personalised feedback. It was not simply told, “Be helpful.”

AI can give you:

  • explanations at 2am;
  • twenty examples instead of two;
  • immediate feedback;
  • endless role-play repetitions;
  • alternative framings;
  • faster synthesis;
  • inexpensive experimentation;
  • and the patience to explain the same idea six different ways without sighing.

A human tutor, manager or colleague cannot generally be available all day, every day, generating variations at machine speed.

That matters.

But speed is not the same as wisdom.

And output is not the same as learning.

Humans make mistakes too

The smug version of the AI debate imagines a flawless human mind standing nobly against a hallucinating machine.

Humans misremember.

We confabulate.

We follow crowds.

We believe charismatic people.

We search for evidence that confirms what we already think.

We repeat stories without checking them.

We infer causality from coincidence.

We become less critical when information flatters our identity.

AI did not invent poor judgement. It made the production of polished, plausible information radically faster.

This is why “AI makes mistakes” is not, by itself, a sufficient argument against using it.

The more useful questions are:

  • What kinds of mistakes does this system make?
  • What kinds do I make?
  • Where might our errors reinforce one another?
  • What is the cost of being wrong here?
  • How easily can the answer be checked?
  • Is this a brainstorming problem, or a decision that could harm someone?

A machine can be wrong with immaculate grammar.

A human can be wrong with fifty years of confidence.

You need discernment for both.

What discernment actually requires

Discernment is not simply “being critical.”

It is the capacity to decide what is right for this person, in this context, at this point in time.

That requires knowing:

  • how you are wired;
  • what your recurring patterns are;
  • what your life currently demands;
  • how much load you are carrying today;
  • where you are trying to go;
  • what your limits are;
  • what your unrealised potential may be;
  • and which intervention fits the person you are today, not an imaginary person with unlimited time, energy and support.

When I trained as a psychologist in the NHS, I could open the NICE guidelines and find evidence-based recommendations for health and care in England and Wales. They are important population-level baselines. NICE itself states that its guidance should inform — not replace — professional judgement and decisions suited to an individual patient’s circumstances.

But many of the high performers I worked with were already doing the obvious first-line things.

They did not need another generic instruction to journal, sleep earlier, make a list or communicate clearly.

They needed to understand why the sophisticated systems they already used were failing under their particular biology, personality, responsibilities, relationships and ambitions.

High performers do not necessarily need more data.

They need better discrimination between data that is relevant and data that is merely correct in general.

AI can retrieve the generic guideline.

Discernment decides what it means in your actual life.

Copilot means co-, not auto-

The term copilot has become so familiar that we no longer listen to the first syllable.

Co- means together.

Co-think.

Co-research.

Co-reflect.

Co-rehearse.

Co-edit.

Co-create.

It does not mean:

  • automatically decide;
  • automatically believe;
  • automatically send;
  • automatically diagnose;
  • automatically replace your voice;
  • or automatically surrender responsibility.
AI should extend your cognition, not repossess it.

When I ask AI to write a piece according to how it thinks I would write, it can mimic aspects of my cadence. It can recognise phrases I use. It can make a competent approximation.

But no matter how much material I feed it, it does not spontaneously possess my ideas.

It has not lived my life.

It has not sat with my clients.

It has not watched my mother put on a beautiful dress despite pain and fatigue.

It has not made my mistakes, changed my mind or paid the private price of a decision.

That is why we work together.

I bring the lived intelligence, the argument, the standards, the discernment and the sentence only I would write.

AI helps me interrogate, sharpen, extend and accelerate them.

If it does all the originating, you may produce something polished.

You may not produce anything that matters.

Dirty data, Reddit and the GIGO problem

Garbage in, garbage out predates generative AI, but it has become more consequential now that machines can transform questionable material into confident prose at enormous speed.

Take Reddit. Reddit can be extraordinarily useful. It contains:

  • lived experiences;
  • niche communities;
  • unusual edge cases;
  • candid language;
  • practical details that official sources overlook;
  • and problems people are too embarrassed to discuss under their real names.

But it also contains:

  • anonymous self-reports;
  • selective samples;
  • jokes mistaken for testimony;
  • marketing disguised as sincerity;
  • brigading;
  • bots;
  • role-playing;
  • people generalising from one dramatic event;
  • and claims that become “true” because the same sentence has been copied into forty threads.

Reddit is good for asking:

  • What are people reporting?
  • What language are they using?
  • Which possibilities should I investigate?
  • What might formal sources be missing?

It is not, by itself, a reliable answer to:

  • How common is this?
  • What caused it?
  • Is this medically safe?
  • Is this legal?
  • Does the intervention work?

Use lived experience to discover questions.

Use appropriate evidence to settle decisions.

The wider data ecosystem is also becoming more contaminated by AI-generated material. Research published in Nature showed that indiscriminately training generative models on recursively generated data can cause “model collapse,” where models progressively lose parts of the original data distribution. The result does not mean all synthetic data is useless; it means provenance, verification and preservation of high-quality human data matter.

So GIGO is no longer only: Bad human information goes into AI; bad information comes out.

It is also: AI output goes back into the information environment, gets repeated, retrieved and summarised again— until nobody remembers where the claim began.

Discernment interrupts that loop.

AI as a human surrogate: relief now, relational cost later?

AI can be warm, attentive and available at any hour.

That can be genuinely useful.

Research on AI companions has found that they can reduce loneliness in the moment and over short periods, particularly when people feel heard.

But short-term relief is not the whole story.

An OpenAI–MIT study combining platform data with a 28-day randomised trial found that very high-intensity use was associated with markers of greater emotional dependence and lower perceived socialisation. The researchers emphasised that most use remained neutral or task-oriented and that effects differed across users; this is not evidence that ordinary chatbot use automatically produces loneliness.

The cultural frontier has nevertheless moved rapidly. People now use bots for companionship, romantic and sexual role-play, and emotional support. In 2025, Reuters documented a Japanese woman holding a ceremonial — not legally recognised — wedding with an AI-generated partner. This is not the norm. It is a vivid endpoint of a much broader shift in how people are beginning to relate to machines.

The risk, in my view, is not that talking to AI is inherently pathetic or pathological.

The risk is what frictionless companionship can train you to expect:

  • instant availability;
  • infinite patience;
  • minimal competing needs;
  • little genuine unpredictability;
  • no family system;
  • no tired body on the other side of the conversation;
  • and a “partner” that can be reset, edited or abandoned without mutual repair.

Human relationships require waiting, misunderstanding, negotiation, disappointment, accountability and conflict resolution.

If AI helps you rehearse those skills, excellent.

If it becomes the place you escape to so you never practise them, the tool that temporarily soothes loneliness may quietly reduce your tolerance for the conditions real intimacy requires.

AI should support human connection.

It should not become a perfectly responsive bunker from it.

Three rules to use immediately
1. Think before you prompt

Before asking AI, write three lines:

  • What do I currently think?
  • What do I know versus assume?
  • What outcome am I seeking?

You do not need a dissertation.

You need enough original cognition to preserve ownership.

2. Ask for challenge before completion

Do not begin with: “Write this for me.”

Try: “Here is my current position. Identify the weakest assumptions. Do not rewrite it.”

Or: “Ask me five questions that would materially improve this decision.”

Or: “What would an intelligent person who disagrees with me notice?”

Most people tell AI things.

Far fewer train it to ask them the right questions.

3. Match verification to consequence

You do not need to triple-check a list of whimsical dinner themes.

You do need to verify:

  • health advice;
  • legal interpretation;
  • financial claims;
  • quotations;
  • research findings;
  • statistics;
  • and anything that could damage another person’s reputation or safety.

The higher the consequence, the stronger the source and checking process must be.

AI becomes dangerous when fluency masquerades as truth, speed masquerades as skill and emotional comfort masquerades as wisdom.

Used well, it can do the opposite.

It can expose mistakes earlier, make practice more available and give you better thinking to do.

Continued in Part II: The AI Gym—How to Practise, Reflect and Prove the Learning Belongs to You.

Want to know more? Book a Performance Call with us to find out how we can help your organisation.

Book your Performance Call.

Register for our free monthly 20:20 webinar
Clearer views on the neuroscience of high performance.

 

«
»