is AI making us more stupid?
It depends, and the “it depends” is the whole point. What the emerging research says about AI, thinking and judgement, and how to use it without eroding the skills it relies on.
Sebastián Rodríguez Pérez is a European campaign strategist with direct experience running and advising campaigns for some of Europe's largest pro-European organisations. He founded the European Campaign Playbook to codify practitioner knowledge across the EU's 27 political environments and make it accessible to campaign professionals, advocacy teams, and civic organisations working at the European level. His work spans electoral strategy, EU public affairs, digital campaigning, and the integration of AI tools into political communication workflows.
Disclosure
Sebastián Rodríguez Pérez is the founder of the European Campaign Playbook and Campaign Intelligence Library. He has worked with pro-European organisations across multiple EU member states. He maintains editorial non-partisanship in all published content; where articles relate to campaigns or organisations he has directly worked with, this is disclosed within the article. He holds no current positions in active electoral campaigns.
What it covers
What the latest research (Bocconi with OpenAI, Microsoft Research with Carnegie Mellon, the OECD, NBER and the book Messy Jobs) says about how generative AI affects our thinking, learning and judgement, and how to use it without eroding the skills it depends on.
Who it is for
Public affairs, policy and communications professionals using AI in daily knowledge work, and the managers responsible for developing junior talent.
When to use it
When you are deciding how to bring AI into your team’s workflow, or worried that leaning on it is quietly weakening the skills it was meant to support.
Key takeaway
Use AI to support the hard thinking, but keep enough of the hard thinking to keep building your own judgement. The friction AI removes is often where professional judgement is built.
In brief
- AI makes knowledge work faster, but producing better work is not the same as becoming better at the work.
- The most useful research distinction is between how you perform with AI and what you can still do when the tool is removed.
- AI tends to improve the answer; your own thinking is what improves the thinker.
- Critical thinking is moving rather than disappearing: from finding information to judging it.
- AI narrows education-based performance gaps while it is available, but the gap can return once it is gone.
- AI is strong at clean, well-specified tasks; public affairs value lives in the messy ones.
- The clean work AI now handles is often how juniors learn the messy work later, an apprenticeship risk to manage.
so, is AI making us more stupid?
I’m preparing a new round of workshops on AI for small public affairs and creative teams in Europe, and the whole thing falls apart if the answer to that question isn’t at least a hopeful version of “it depends”.
We already know AI can make us faster. Research takes less time. First drafts appear sooner. Documents can be compared in seconds. Long reports become short summaries. A Commission proposal can become a briefing before you’ve finished your coffee.
The productivity case is getting easier to make.
The more interesting question is what happens to our own abilities along the way.
better work doesn’t always mean better skills
AI can remove a lot of friction from knowledge work. And some of that friction may have been doing something useful.
Struggling with a blank page helps you learn to write. Reading an 80-page consultation document teaches you where important details tend to hide. Developing your own arguments before hearing everyone else’s helps build judgement.
AI can now do a lot of that work for us.
That creates an important distinction: producing better work isn’t necessarily the same as becoming better at the work.
And that distinction is starting to show up in the research.
get ready for conflicting AI headlines
We’re going to see much more research on the cognitive effects of AI. And it won’t give us one neat answer.
Some studies will look at students, others at knowledge workers. Some will measure productivity, others memory, learning, creativity, attention or critical thinking. Some will rely on surveys. Others will use controlled experiments.
And one of the most useful distinctions will be between how people perform with AI and what they can still do when the tool disappears.
This means we need to become better consumers of AI research too. A survey of thousands of people can give us a useful picture of behaviour. A randomised controlled trial can provide stronger evidence about cause and effect, while often looking at a much narrower task. Large reviews can help us see patterns across different studies.
The methodology matters. And a pattern is starting to appear.
AI can improve the answer. thinking improves the thinker.
A recent experiment from Bocconi University, in collaboration with OpenAI Economic Research, involved 1,053 students.
Students were assigned to four groups: ChatGPT access, causal-reasoning training, both, or neither. The results are interesting.
Students using ChatGPT produced higher-quality, more coherent work that looked closer to expert output. Causal-reasoning training did something different: it led students towards more diverse ideas. The combination gave students both benefits.
For public affairs, that distinction matters. Imagine you have a new Commission proposal in front of you. One approach is:
Tell me what our position should be and give me five arguments.
You’ll probably get something useful. Another approach is:
I’ve read the proposal. These are the three provisions I think matter most. Show me what I’m missing. Challenge my assumptions. Tell me how a consumer group, a sceptical MEP and the Commission might respond.
Same AI. Very different intellectual exercise.
critical thinking may be moving, rather than disappearing
Research from Microsoft Research and Carnegie Mellon offers another useful clue. The study surveyed 319 knowledge workers about their use of generative AI at work.
People who reported greater confidence in AI also reported using less critical-thinking effort. People with greater confidence in their own abilities tended to question, verify and intervene more.
The methodology matters here: this was based on self-reported examples rather than a controlled experiment. Still, one idea from the study feels particularly relevant for our work. AI may be changing where critical thinking happens.
Maybe you no longer spend two hours manually researching what 20 MEPs have said about a file. AI gives you a first analysis in ten minutes. Great.
Your value then shifts towards being able to look at that analysis and say:
- That was their position before the election.
- You’re treating the political-group position as the individual MEP’s position.
- Technically, that’s the right committee. Politically, that isn’t the person deciding the outcome.
The hard work moves from finding information to judging it.
performance and learning are two different things
The OECD’s Digital Education Outlook 2026 takes a broader look at the evidence. One of its clearest messages is simple: successfully completing a task with AI doesn’t automatically mean you learned how to complete that task yourself.
People can produce stronger work while using general-purpose AI and then lose that advantage when AI is removed.
There’s a more promising side too. AI used with a clear learning purpose can support deeper thinking. Instead of simply providing answers, it can question, nudge, challenge and give feedback. That difference matters well beyond education.
think about the junior colleague
Picture someone working on their first complicated legislative file. They spend an afternoon understanding the proposal. They check the existing legislation. They work out who has competence. They read Parliament amendments. Delegated acts confuse them for a while. They write a fairly rough two-page note and get it covered in comments from a director.
Slow? Absolutely. Six months later, they understand the file.
Now AI can produce a polished briefing in 45 seconds. That’s a huge efficiency gain. The learning depends on what happens next.
If the junior colleague checks the sources, challenges the analysis and works through the reasoning, AI can help them learn faster. If the briefing simply gets forwarded, the organisation gets the output without necessarily developing the person.
AI can close gaps while we’re using it
An NBER randomised experiment gives us another piece of the picture. 1,174 adults completed a workplace-style business problem-solving task, with some participants receiving access to generative AI.
AI improved performance and particularly helped participants with lower levels of education, narrowing much of the performance gap while the tool was available. That’s significant. AI can give more people access to capabilities that previously required more experience, education or resources.
And the follow-up matters too. When AI was removed, a substantial part of the original education-based performance gap returned. Participants who combined intensive AI use with sustained personal effort performed better afterwards.
The lesson isn’t that heavy AI use is the problem. AI use and intellectual effort can work together.
public affairs is full of messy work
Luis Garicano, Jin Li and Yanhui Wu make a related argument in Messy Jobs. AI is particularly strong at clean tasks: work that can be clearly defined, specified and evaluated. Public affairs has plenty of those.
Knowing what a proposal says is relatively clean. Understanding why the Commission proposed it now is messy.
Producing a stakeholder list is clean. Knowing who actually matters is messy.
Summarising an MEP’s public statements is clean. Working out what might persuade them is messy.
Generating ten arguments is clean. Knowing which argument you should leave outside a particular meeting room is messy.
A lot of our value lives in those messy spaces: judgement, relationships, timing, trust, institutional culture and negotiation. AI can process the official record. Understanding that “we remain open to further discussion” can sometimes mean “this is going nowhere” requires something else.
the apprenticeship problem
Here’s the part organisations should pay particular attention to. A lot of the clean work AI can now handle is also how junior professionals learn to do the messy work later.
- You monitor the committee. Eventually, you understand the committee.
- You write the meeting notes. Eventually, you learn what matters in the room.
- You draft the first position paper. Eventually, you learn how arguments are built.
- You read the legislation. Eventually, you know where to look.
You make mistakes. Someone more experienced corrects them. It’s inefficient. It’s also apprenticeship.
Organisations could therefore become much more productive in the short term while weakening the process that develops their future experts. That’s a management challenge, rather than an argument against AI.
the question is what we choose to outsource
The evidence doesn’t support a simple claim that AI is making us less capable. It does give us a reason to pay attention to which parts of thinking we hand over.
There’s a meaningful difference between:
Read this proposal and tell me what to think.
And:
I’ve read this proposal. Here’s my interpretation. Show me where my reasoning could be stronger.
There’s a difference between asking AI to produce your position paper and asking it to challenge your position from the perspective of the Commission, Parliament, an NGO and your strongest opponent. There’s a difference between using AI to replace reading and using it to understand what you’ve read more deeply.
keep the useful friction
The rule I’m taking into the next round of workshops is simple:
Use AI to support the hard thinking. Keep enough of the hard thinking to build your own judgement.
Because some of the friction we’re trying to remove is also where professional judgement gets built.
I want AI to help me produce better work. And, more importantly, I want to use it in a way that helps me become better at the work itself.
Sources
- 1.Bocconi University with OpenAI Economic Research, “Training novices to think, or giving them LLMs?” — https://openai.com/index/what-students-gain-from-chatgpt-critical-thinking-training/
- 2.Microsoft Research and Carnegie Mellon, “The Impact of Generative AI on Critical Thinking” — https://www.microsoft.com/en-us/research/publication/the-impact-of-generative-ai-on-critical-thinking-self-reported-reductions-in-cognitive-effort-and-confidence-effects-from-a-survey-of-knowledge-workers/
- 3.OECD, “Digital Education Outlook 2026” — https://www.oecd.org/content/dam/oecd/en/publications/reports/2026/01/oecd-digital-education-outlook-2026_940e0dd8/062a7394-en.pdf
- 4.NBER Working Paper 34851, “Does Generative AI Narrow Education-Based Productivity Gaps?” — https://www.nber.org/papers/w34851
- 5.Luis Garicano, Jin Li and Yanhui Wu, “Messy Jobs: The Work That AI Cannot Reach” — https://messyjobs.ai/
- 6.UC Berkeley / Science, “Generative AI use and misuse call for assessment reform in higher education” — https://news.berkeley.edu/2026/05/21/the-largest-study-of-ai-use-by-undergrads-is-in-revealing-disparities-in-access-and-in-cheating/