Critical Curiosity

Curiosity
before answers.

In an age of instant answers, we believe in the enduring value of a good question.

Read our manifesto

Rethinking education in the age of AI.

The world has more answers.
It needs better questions.

We believe education is an active practice, not the passive reception of information. Learning to ask good questions is the first step towards becoming a critical thinker. Imagination and creativity make us uniquely human, and we need to keep nurturing them.

AI makes answers easier to obtain. Our concern is what happens to thinking, creativity and imagination along the way. We want to turn AI into a tool that trains critical thinking.

Curiosity before answers.

We begin with the awareness that our understanding is unfinished. A good question opens a path to discovery, and the freedom to change our minds.

The scientific method.

Stating our assumptions, forming a hypothesis, testing it, assessing the evidence and looping back: this cycle is the backbone of understanding in any subject. It is a universal tool that is too often dismissed.

Imagination and creativity.

They are the traits that helped our species adapt, invent and thrive, and they make us who we are. In an age of machine-made answers, they need to be protected and nurtured.

The thinking behind this

“Liberating education consists in acts of cognition, not transferals of information.”

Paulo Freire, Pedagogy of the Oppressed.

Critical curiosity Labs

Where our name comes from.

The Brazilian educator Paulo Freire gave us the idea at the heart of our name. In Pedagogy of Freedom, he describes how the spontaneous curiosity we are all born with can grow into critical curiosity: a disciplined habit of questioning, investigating and testing what we think we know. Nurturing that curiosity is one of the central tasks of education.

Today, most school systems do the opposite. Curiosity and imagination are pushed aside in favour of transferring knowledge, as if students were empty vessels waiting to be filled. Lessons reward the right answer more than the good question and leave little time to wonder, guess and be wrong. In the age of AI, when any fact or piece of knowledge is available at any moment, this approach is not just useless but counterproductive. We need to learn again how to nurture curiosity and imagination in the new generations.

By answering instantly and confidently, AI is taking away not only the students’ chance to explore and discover on their own, but also their confidence in their own ability to find answers. We want to reverse this, and use AI to coach learners to ask better questions and solve problems for themselves.

Secondary education · Observational study

Completed work can conceal a learning gap.

A working paper following 26,811 pupils in China reports an 18% rise in homework scores after AI adoption, alongside a 20% fall in monthly exam scores within six months. The losses were concentrated among pupils whose behaviour was consistent with outsourcing their homework.

Study context and source

Strömberg, Lei & Wu (2026) analyse 30 months of data using a difference-in-differences design. Pupils chose when to adopt AI, so possible confounding remains a concern. This is a working paper, not yet peer-reviewed, about general-purpose AI use rather than a purpose-built tutor.

The Generative AI Learning Penalty · CEPR DP 21577

Higher education · Randomised experiment

AI can support lasting understanding.

Undergraduates with AI access scored higher on unaided knowledge tests, with gains persisting a week later. Improvements in essay quality were larger among those who used AI to explain concepts rather than generate text.

Study context and source

Contractor & Reyes (2026) randomly assigned AI access in proctored, in-person learning sessions. Immediate knowledge-test scores rose by 0.27 standard deviations. The study followed students for one week; it does not establish long-term effects, and how students chose to use AI was not itself randomly assigned.

Experimental Evidence on the Learning Impact of Generative AI · IZA DP 18792

Higher education · Randomised crossover trial

A well-designed AI tutor can double learning gains.

At Harvard, physics undergraduates learned more than twice as much from an AI tutor built on teaching best practices as from an in-class active learning lesson, in less time. They also reported feeling more engaged and more motivated.

Study context and source

Kestin, Miller, Klales, Milbourne & Ponti (2025) studied fewer than 200 students in an introductory physics course. Every student experienced both conditions, with the order randomly assigned, and the comparison was a lesson already using active learning rather than a traditional lecture. The study covered two topics and measured learning shortly afterwards, so it does not show long-term effects. The tutor was designed to guide students step by step rather than give them answers.

AI tutoring outperforms in-class active learning · Scientific Reports

We believe that AI has the potential to bring real benefits to students, including a deeper and more structured understanding of complex subjects.

Coaching learners to ask better questions.

We are developing AI tools that coach scientific reasoning, critical thinking, curiosity and imagination and are inspired on the scientific method and on the work of educators and scientists. Our tools are designed to trigger learners to ask questions and guide them to break complex problems into manageable steps, without giving them easy answers.

Our first product is "Fermi", an AI tutor for numeracy, maths and physics. Our ambition is to see Fermi helping the next generation of students and pupils into becoming confident and skilled workers who are comfortable working alongside AI, and that are secure of their own contribution and value to the world.

For educators, learners and anyone who wants to explore the potential of AI in education.

How we work with schools, colleges and universities