How to use AI critically: six ways to keep your thinking your own

How to use AI critically, and which skills will matter in a job market that AI is already reshaping: these are the questions we put to Roberto Mordacci, Full Professor of Moral Philosophy and Vice-Rector for Humanities and Social Sciences at Vita-Salute San Raffaele University, and one of the speakers at Next Generation AI, the international summit on artificial intelligence in schools organised by the Italian Ministry of Education and Merit (Villa Erba, Cernobbio, 19–22 October 2026).

What is intelligence? From information to knowledge 

Mordacci offers a working definition of intelligence: the ability to process information in order to generate knowledge.

His starting point is «a fundamental distinction between information and knowledge. Information is data, processed content: there is often too much of it, and it tends to change and disappear over time. Knowledge, by contrast, is the human product of processing information intelligently, through scientific methods (hypothesis and experiment), philosophical methods (experience and the principle of non-contradiction) and artistic and literary ones», the professor explains.

Intelligence is what keeps this processing under control, because «it secures a process competence in handling information, and that competence generates knowledge. In practical terms, intelligence is the ability to compare, associate, critically analyse and connect information using logical, dialectical, discursive and metaphorical methods».

When we hand a task over to artificial intelligence, we receive a ready-made result. Reading and using that result requires a different competence from the one needed to produce it. If we simply accept it as true, we stop exercising that very process competence: the ability to direct and check how information is processed, by understanding, applying and mastering the method that leads to that result. 

Six tips for using AI critically

To stay actively involved in learning while still making use of artificial intelligence, Mordacci sets out six recommendations for anyone who uses large language models (LLMs), the technology behind tools such as ChatGPT, Claude and Gemini.

• Do your research before querying the system. Gather information by talking to teachers and experts, by consulting authoritative sources such as books, scientific articles and institutional websites, and through direct experience. This preparation makes it easier to frame the request and to evaluate the answer.
• Think about how you would tackle the problem. Before reading the proposed solution, ask yourself what reasoning you would follow and which sources it would make sense to consult. Comparing the two helps separate an answer that only sounds convincing from an argument backed by good, objective reasons.
• Take notes. Identify any further information the system offers, then verify it and work it through. Notes help make what you have understood your own and develop independent thinking.
• Recognise the risk of hearing what you want to hear. An AI model can play along with a user’s expectations and reinforce their biases. A study presented at ICLR 2024, one of the leading international conferences on machine learning, describes this tendency, known as sycophancy, as a general behaviour of AI assistants. An answer we like calls for the same critical scrutiny as one that contradicts our beliefs.
• Put the answers to the test. Flag weak points in the reasoning, introduce different information, raise objections. A well-founded argument should be able to withstand critical examination; learning to raise relevant objections is part of learning.
• Use the tool to learn. The goal is still a better understanding of ourselves and the world. The system can feed into our thinking, but decisions have consequences for the people who make them and for those they affect.

These recommendations lead to a simple test for delegating: «So this is when we can hand a task over to AI: quite simply, when we know how to do it ourselves». The competence we have acquired lets us check the outcome of the work entrusted to the system, spot any errors and see where to step in. 

Testing AI answers with the dialectical method 

Testing a model takes more than asking it to correct itself. If the system backtracks and agrees with the user, it remains to be seen which answer is well founded. The reasons for the change need to be examined; otherwise one claim accepted without verification is simply swapped for another.

«With AI, we need to adopt a dialectical model», Mordacci argues. The dialectical method works through arguments and objections: it involves examining a thesis, challenging one of its steps and assessing whether the reply resolves the problem.

We do the same when we argue with other people and put our reasoning up for their scrutiny. Applied to AI, the method lets us play an active part in shaping the answer and assess how well it holds up.

Critical thinking and AI: why it matters for work 

One of the most publicly debated effects of artificial intelligence is the rapid, unpredictable transformation of the job market.

Mordacci approaches the question through critical thinking, understood in a broader sense than the reasoning skill taught in logic: «the analysis of the conditions under which human capabilities are exercised and deployed», he clarifies.

This means examining the concrete conditions that allow people to develop their capabilities, or stop them from doing so. Acquiring a skill and having the opportunity to use it are, after all, two separate matters. The open question is what training is needed today for tomorrow’s work.

This is where younger generations’ concerns about their professional future come in. «What worries them in particular is whether they can access training suited to the demands of the job market, and whether the skills they have acquired will be recognised», Mordacci notes.

The demand for good training is legitimate. Designing it is harder because of uncertainty over which skills will be needed as artificial intelligence spreads, given that the job market is changing radically and very quickly.

AI and the future of work: what the data show

Technological change can reduce demand for some activities and boost demand for others. The World Economic Forum’s Future of Jobs Report 2025 estimates that between 2025 and 2030 the major transformations under way, technology included, will create 170 million jobs and displace 92 million. The net balance is positive: 78 million. However, the more than 1,000 large employers surveyed expect 39% of core skills to change.

An analysis by the International Monetary Fund suggests that artificial intelligence could affect almost 40% of jobs worldwide and around 60% in advanced economies. In those economies, roughly half of the affected jobs could see productivity gains, while in the other half AI could reduce demand for workers. To understand the effects on employment, it helps to distinguish between two phenomena:

• structural unemployment, which concerns the mismatch between the skills of jobseekers and those required for the jobs available;
• frictional unemployment, which concerns the time jobseekers need to find a job suited to their skills, even when vacancies exist.

New jobs become accessible to people who lose theirs only when training opportunities exist.

Future skills: what the humanities bring  

Mordacci identifies a specific contribution that the humanities can make. Studying philosophy, for example, trains the ability to examine the assumptions behind an argument and to weigh the consequences of decisions. These are useful tools for understanding how work is changing, even when the professional profiles that will emerge have yet to take shape.

Whether these abilities can be put to good use also depends on the choices companies make. As they introduce artificial intelligence, organisations will have to decide how much to invest in training and what opportunities to offer people who take on different tasks.

References

World Economic Forum, Future of Jobs Report 2025, gennaio 2025. weforum.org

Georgieva K., «AI Will Transform the Global Economy. Let’s Make Sure It Benefits Humanity», IMF Blog, 14 gennaio 2024. imf.org

Sharma M. et al., «Towards Understanding Sycophancy in Language Models», International Conference on Learning Representations (ICLR), 2024. arxiv.org/abs/2310.13548