Home › Why this matters

Why this matters

AI should extend human capability, not replace human thinking.

Risk 1

Emotional and mental dependence

Some people turn to AI for companionship, reassurance or mental health support. AI is always available, so it can feel easier than talking to a person — and it may reinforce unhelpful thinking instead of challenging it.

Why it matters: AI is not a trained counsellor and cannot replace human relationships or professional care.

What the research shows

  • In a four-week randomised study by MIT Media Lab and OpenAI, completed by 981 participants, people who used the chatbot more each day tended to report more loneliness, emotional dependence and problematic use, and less socialising with people. The authors note this pattern cannot show cause and effect. Fang et al., 2025 (preprint)
  • In a 2025 survey of 1,060 US teens aged 13–17, 72% had used an AI companion and 52% used one at least a few times a month. Common Sense Media, 2025

What we do not know yet: this research is new. It shows patterns and associations, not proven harm, and effects differ from person to person.

If you are struggling, please talk to someone you trust or a qualified professional.

Risk 2

Loss of critical thinking

AI answers sound confident whether or not they are right. When we accept them without checking, mistakes pass unnoticed — and we stop practising the thinking AI is doing for us.

Why it matters: judging, checking and deciding are skills. They stay strong when we use them.

What the research shows

  • In a field experiment with high school maths students, those who practised with a standard GPT-4 chatbot scored 17% lower on a later exam taken without AI than a control group who never had access. A version designed with safeguards for learning largely avoided this drop. Bastani et al., PNAS 2025
  • In a survey of 319 knowledge workers, higher confidence in AI was associated with less critical thinking, and higher self-confidence with more. Lee et al., Microsoft Research and Carnegie Mellon, CHI 2025
  • In a survey and interview study of 666 people, more frequent AI use was linked to lower critical-thinking scores and to greater “cognitive offloading”. Gerlich, Societies, 2025

What we do not know yet: the experiment above is the strongest evidence; the other two studies are surveys that show associations, not causes. Studies so far are short-term, so long-term effects are not yet known.

Risk 3

Unstructured AI use

Many people use AI by default, without deciding what it is for, where it does not belong or how to check what it produces.

Why it matters: without a clear way of working, unchecked answers get relied on and mistakes slip through.

What the research shows

  • A 2025 survey of more than 48,000 people in 47 countries found that 66% use AI regularly. The study reports that many rely on AI output without evaluating its accuracy (66%) and are making mistakes in their work because of AI (56%). University of Melbourne and KPMG, 2025

What we do not know yet: these are self-reported survey answers. They show how people say they use AI, not the full consequences.

Risk 4

Economic exclusion

The risk that individuals and communities lack the awareness, skills, resources, or access needed to benefit from AI-driven opportunities and adapt to changing work.

Why it matters: AI is changing what many jobs involve. People who learn to use it well can gain; people left without the skills, tools or support risk falling further behind.

What the research shows

  • The International Monetary Fund estimates that almost 40% of jobs worldwide are exposed to AI, about 60% in advanced economies. It warns that lower-income countries are less prepared to benefit, and that in most scenarios AI is likely to worsen overall inequality. Cazzaniga et al., IMF, 2024
  • In a 2024 survey of 31,000 knowledge workers in 31 markets, 75% used generative AI at work, but only 39% of those using AI had received AI training from their employer. Microsoft and LinkedIn Work Trend Index, 2024
  • In a study of 5,179 customer-support agents, an AI assistant raised productivity by 14% on average and by 34% for novice and lower-skilled workers. Used well, AI can help the people with the least experience most. Brynjolfsson, Li and Raymond, NBER
  • A review of studies covering more than 100,000 people across many countries found that women use generative AI less than men across regions, sectors and occupations, and the gap persisted even when access was equal. Otis et al., Harvard, 2024

What we do not know yet: being exposed to AI is not the same as losing a job, and productivity studies so far cover specific roles. Who benefits will depend on access to skills, tools and training.

What safe AI use means

Five things a capable AI user can do

  1. Know how to use it Give AI real context and frame the task clearly.
  2. Know when to use it Reach for AI where it genuinely helps, not out of habit.
  3. Know when not to Keep sensitive information, important decisions and relationships human.
  4. Check what it produces Look for invented facts, errors, missing context and bias.
  5. Keep decision authority You stay accountable for what you accept and act on.