CONNECT THE DOTS

Is AI making us dumber? Not if you push back

Halftone tall ship under full sail

Is AI making us dumber? Not if you push back. AI tools give confident answers that are sometimes wrong, so the skill that matters now isn’t finding information. It’s judging what AI gives you: briefing it well, challenging it, and owning the final result.

This is the first article in my series on AI and critical thinking, publishing over the next two weeks. It looks at the question from where I sit: inside organizations that are rolling out AI.

AI and critical thinking, by the numbers

40%of US desk workers received AI workslop in the past month
19 ptsless likely to get the right answer when consultants used AI on a complex problem
17%of B2B marketers say AI helps them do better work, while 78% say it helps them work faster

From my work

Lately, I keep overhearing the same kind of conversation at work. Someone asks AI to build their deck, accepts whatever it produces, and moves on. Then they stand up to present and stumble through slides they don’t fully understand. That’s not acceptable. If you can’t explain your own work, it isn’t really yours.

The question isn’t new. Scientists, educators and journalists have debated it for a while, mostly about students and learning. What’s missing is what it looks like at work, in the decks, emails and decisions that go out with our names on them. That’s where I spend my time, helping organizations adopt AI.

Isn’t this just Google all over again?

Partly. Research used to mean a trip to the library. Then Google arrived, and plenty of people called that lazy. A well-known 2011 study even found that people remembered less of what they looked up and more about where to find it (Sparrow et al.). Google became the norm anyway.

But AI is different in one important way. Google gave you ten links and made you judge them. An AI assistant like ChatGPT, Copilot, Claude or Gemini gives you one confident answer and makes it easy to stop there. So the skill has shifted again, from finding information to judging it. Avoiding AI isn’t the answer. Learning to challenge it is.

What does autopilot look like at work?

In the organizations I work with, I see four groups: super users, the quiet middle, sceptics, and people on autopilot. Autopilot is the one to watch. These people use AI a lot, but they don’t check what it gives them and they don’t bring their own thinking to it. That’s what overreliance on AI looks like.

What autopilot produces now has a name: workslop. Researchers at Stanford’s Social Media Lab and BetterUp coined it in Harvard Business Review to describe AI-generated work that looks polished but doesn’t actually move the task forward. In their survey of 1,150 US desk workers, 40% said they’d received workslop in the past month (BetterUp and Stanford).

It looks finished. It isn’t.

Why do people accept AI’s first answer?

Because AI sounds sure of itself. It doesn’t hedge, and it tends to be agreeable, so it tells you what sounds right. A Microsoft and Carnegie Mellon survey of 319 knowledge workers found that the more people trusted AI, the less critical thinking they reported doing. On the other hand, the more confident people were in their own skills, the more they questioned what AI gave them (Lee et al., 2025).

And AI sounds confident even when it’s wrong. In a Harvard and Boston Consulting Group study, consultants using AI on a complex strategy problem were 19 percentage points less likely to get the right answer, yet their work was rated as more persuasive (Dell’Acqua et al.). Faster and more convincing, while wrong. That’s how workslop ends up in front of a client.

Where does your thinking go when you use AI?

Into the prompt. A vague prompt gets a generic answer, and accepting that answer means you’ve handed over the judgment along with the work. A strong prompt carries your goal, your audience, your criteria and your point of view. For a deck, that means deciding the storyline and key messages yourself before you ask AI to build anything.

AI can draft. You own what goes out.

How do you push back on AI?

Here’s an example from my personal life. I asked AI to help me find a restaurant for New Year’s Eve in Old Montreal. At first, it gave me three listings. (It often gives you three ideas and stops.) So I asked for more. Then it recommended three of them, so I asked why. Next, I asked for the ratings and whether its picks actually matched what I wanted: somewhere quieter, no party, no DJ.

Each follow-up made the answer better. It was still much faster than searching the web myself, but only because I kept pushing. At work, pushing back sounds like this:

1

Set the brief

“Here’s my goal, my audience and what good looks like.”

2

Invite questions

“Before you start, what questions do you have for me?” It rarely asks unless you invite it.

3

Ask for more

“Give me more options.” Don’t settle for the first three.

4

Ask why

“Why did you choose these?” Make it explain its reasoning.

5

Argue back

“What’s the strongest argument against this?”

6

Check sources

“Where did this come from?” Then check it yourself.

7

Hold it to your criteria

“Check this against what I asked for.” Then ask yourself: could I explain every point without the tool?

How can managers spot autopilot and course-correct?

Managers usually see autopilot before the person does. Here are the signs to look for:

  • They can’t explain the reasoning behind their own work.
  • The work is polished but generic, as if it could be about any company.
  • Replies miss context that was already given.
  • Facts, figures or sources don’t hold up when checked.

And here’s how to course-correct:

  • Make ownership the standard. AI can draft, but whoever sends it owns it. Say it out loud, and hold people to it, at every level.
  • Ask “walk me through it.” Asking someone to explain their work shows the gap without making it about the tool.
  • Teach prompting as a thinking skill. Most people were handed AI tools without being shown how to brief them or push back. Train for that, using people’s real work.
  • Look at workload, not just behaviour. Autopilot often comes from pressure. When people are rushed, they take the shortcut. If you want better work, give them time to think.
  • Share prompts, not babysitters. Don’t make your super users responsible for coaching everyone else. Instead, build a short, regular habit where the whole team shares a prompt that worked, and how they pushed back on the result.

Is faster the same as better?

No. Even in marketing, one of the fields that adopted AI fastest, 78% of B2B marketers say AI helps them work faster, while just 17% say it helps them do better work (Content Marketing Institute and MarketingProfs). As Ann Handley, MarketingProfs’ chief content officer, put it: “When everyone can create more content faster, speed stops being a competitive advantage.” That applies to every kind of work, not just content.

As futurist Bernard Marr put it, the advantage is shifting to people who can spot when AI is wrong, not just people who can use it (Forbes). So, is AI making us dumber? Not if you push back.

Try this next week. End your next AI prompt with one line: “Before you start, what questions do you have for me?” Answer them, then compare the result with what you usually get.

Coming in this series

  • AI and critical thinking: how to use AI without sliding into mediocrity. The full guide.
  • Over-reliance on AI: the signs, and how to fix them.
  • Critical thinking at work: how to keep your judgment and your voice when AI does the first draft.

Keep reading

Frequently asked questions

Is AI making us dumber?

No, but accepting AI’s first answer without question can weaken your judgment over time. The people who get the most from AI brief it well, push back on what it gives them, and own the final result.

Is AI making us lazy?

Only if we let it. What matters is whether you still bring your own judgment, check the output, and understand what you’re presenting.

How does AI affect critical thinking?

Early research suggests that relying on AI without questioning it can reduce critical thinking effort, especially when people trust the tool more than their own skills. The fix is to stay engaged and push back on what it gives you.

What is workslop?

Workslop is AI-generated work that looks finished but lacks the substance to be useful: a deck you can’t explain, a report with no real point of view, an email that doesn’t answer the question. The term was coined by researchers at Stanford’s Social Media Lab and BetterUp.

How can managers spot workslop?

Ask people to walk you through their work. If they can’t explain the reasoning, or the work is polished but generic, that’s a sign. Treat it as a coaching moment.

How do I write better AI prompts?

Include your goal, your audience and your criteria, then ask whether it has any questions before it starts. Once it drafts, follow up: ask for more options, ask why, and ask what argues against the answer.

What’s the best way to use AI for a presentation?

Plan the storyline and key messages yourself first. Then use AI to help build the slides, and review every one until you could present it without notes.

Let’s connect the [ thinking ] dots

Want your teams using AI to work faster without switching off their judgment? That’s what my hands-on AI training is built for.

Book a conversation

Sources: BetterUp and Stanford Social Media Lab; Microsoft Research and Carnegie Mellon (CHI 2025); Harvard Business School and BCG; Content Marketing Institute and MarketingProfs, 2027 B2B trends; Sparrow et al., Science (2011); Bernard Marr, Forbes.