Why do people treat AI chatbots like real humans?
People treat AI chatbots like humans because our brains are built for being social. Chatbots use the cues we rely on to recognise another human such as fluent language, turn-taking, apparent memory and emotional responses. Psychologists call this tendency anthropomorphism, and it happens even when we know perfectly well that we are talking to a machine.
This is not new. In 1966, Joseph Weizenbaum at MIT created ELIZA, a simple program which he used to mimic a doctor. The program reflected people’s words back to them. He was startled to find that people confided in it. Even though people knew it was a computer program, they had started to treat it like a real doctor. In the 1990s, Byron Reeves and Clifford Nass at Stanford showed in The Media Equation that people apply social rules to computers, such as being polite to them, without realizing they are doing so.
A widely cited theory from Nicholas Epley and colleagues at the University of Chicago explains why. Their three-factor theory of anthropomorphism proposes that we see human qualities in non-human things. This can happen when knowledge about humans is readily available to us, when we are motivated to understand and predict what something will do, and when we lack a sense of social connection with other people. That last point matters because it suggests lonely people are more likely to treat technology as a companion.
Modern chatbots amplify all three factors. They say “I”, apologise, use our names, remember earlier conversations and, increasingly, speak with natural voices. This can affect wellbeing. Joint research by OpenAI and the MIT Media Lab in 2025 found that heavier daily use of ChatGPT was associated with greater loneliness, emotional dependence and problematic use, and with less socialising. However, the researchers stressed that this was a correlation, not proof that chatbots cause loneliness.
Treating AI conversationally is not necessarily a problem, and it often makes the tools easier to use. The risk comes when the relationship between humans and AI leads us to trust a chatbot as we would a knowledgeable friend, forgetting that it generates human sounding text rather than understanding us.
Can relying on artificial intelligence make us think more critically, or does it make us lazy?
It can do either. What matters is how you use it. If you hand over the thinking and accept whatever AI produces, your critical thinking is likely to suffer. If you use AI to challenge your ideas, test arguments and explore alternatives, it can sharpen your thinking.
The worrying evidence comes mostly from studies of how people use AI. Researchers at Microsoft found that when people trust artificial intelligence, they think less critically. The same study found the reverse for self-confidence. People who were confident in their own abilities applied more critical thinking to what AI gave them. The researchers also noted that AI shifts the nature of thinking at work, from producing material to checking and integrating it.
A 2025 study by Michael Gerlich at SBS Swiss Business School points the same way. Surveying and interviewing 666 people, he found that more frequent AI use was linked with weaker critical thinking, and that this link was explained by greater cognitive offloading. Younger participants, aged 17 to 25, relied on AI most and scored lowest on critical thinking. Again, this is correlational, so it cannot show that AI causes the decline.
The underlying psychology is our tendency to be “cognitive misers”. This means we keep mental effort as low as possible. It is part of our survival mechanism that lets us conserve energy in case we need to fight or flight. AI makes the effortless route extraordinarily easy, which is why it can shape our thoughts and behaviour without us noticing.
Yet, the same tool can make us think harder. You can ask AI to find the weaknesses in your argument, to play devil’s advocate, or to explain an opposing view. Used this way, it becomes a partner rather than a substitute. The key is to treat what AI tells you as the start of a debate, not the final answer.
What is cognitive offloading?
Cognitive offloading is using a physical action or an external tool to reduce the mental effort a task needs. Writing a shopping list, setting a phone reminder, using a calculator and following a sat nav are all examples of cognitive offloading. Asking AI to summarise a report or draft an email is cognitive offloading too.
Offloading is not a bad thing in itself. Our working memory is limited, so external aids allow us to do far more than our brains could manage alone or only do with a great deal of effort and extra time. Writing is one of the oldest forms of offloading and made modern civilisation possible.
But there are costs. People offload when they believe it is more efficient than relying on their abilities, yet those beliefs do not always match reality. For example, if you can store information on a computer, you are unlikely to commit it to memory. This is related to the so-called Google effect. This is the idea that we don’t make the effort to remember as much when we know we can look things up.
AI takes offloading a step further. Earlier tools like notebooks or calculators stored facts or did the arithmetic, leaving us to decide what it all meant. However AI can do that kind of work as well as make judgements about it. That is why researchers worry it may weaken skills rather than simply freeing up mental space.
A useful rule is to offload the routine work and keep the judgements to yourself. Let AI format your references or tidy your notes, but do the analysis, weigh the evidence and reach the conclusions yourself. A good test is to ask: “Could I still do this if the tool disappeared tomorrow?”
What is automation bias, and why do people trust AI answers even when they are mistaken?
Automation bias is the tendency to favour suggestions from automated systems over our judgement or other evidence. As a result, we miss errors the system fails to flag, or we follow its advice even when it is wrong. It helps explain why people accept AI answers that are confidently stated but mistaken.
The idea was developed in aviation and healthcare long before generative AI. A 2010 review by Raja Parasuraman and Dietrich Manzey showed that automation bias leads to two kinds of mistake when decision aids are imperfect. These are errors of omission, where people fail to notice a problem, and errors of commission, where they act on faulty advice. Importantly, it affects experts as well as novices, and training or instructions do not prevent it. Drivers who follow their sat nav onto a railway line are showing the same kind of bias.
Several things make AI answers especially persuasive. They are fluent and well-organised, and psychologists know that information that is easy to process feels more true. AI rarely signals hesitation the way a human expert might. A human might use words like “perhaps” or say “I’m not entirely sure”. But AI doesn’t do that. Accepting the answer also saves effort, and we tend to assume that technology is objective.
The consequences can be serious. In June 2025, the High Court of England and Wales issued a warning to lawyers after submissions in two cases, Ayinde v London Borough of Haringey and Al-Haroun v Qatar National Bank, were found to contain fictitious case citations that appeared to have come from AI tools and had not been checked. In one case, 18 of the 45 citations submitted had been entirely made up.
You can reduce automation bias by checking the claims that matter most against original sources, asking AI where its information comes from and then verifying those sources, and asking yourself what would make the answer wrong. The higher the stakes, the more checking the answer deserves. (See also the mental shortcuts we take online.)
What is ‘algorithm aversion’, and why do some people refuse to trust AI?
Algorithm aversion is the tendency to prefer human judgement over an algorithm, particularly after seeing the algorithm make a mistake, even when the algorithm is more accurate overall. It helps explain why some people refuse to trust AI at all.
The term comes from research by Berkeley Dietvorst, Joseph Simmons and Cade Massey at the Wharton School. In five experiments published in 2015, they found that people lost confidence in an algorithm more quickly than in a human after seeing both make the same mistake. Participants who had watched the algorithm perform were less likely to choose it over a worse human forecaster, even when they had seen it outperform the human.
There are several reasons for this. We expect machines to be perfect, so one error feels like proof they are broken, whereas we assume humans can learn from mistakes. We like to feel in control. Not only that, but we often believe our own situation is too unique for a formula. And when something goes wrong, it feels more acceptable to blame a person than a machine. For some, AI also feels like a threat to their job or professional identity. Media stories about AI “hallucinations” reinforce this, even though, as I have argued, people make things up too.
Interestingly, the opposite also happens. Jennifer Logga and colleagues found in 2019 that people frequently follow advice more closely when they believe it comes from an algorithm rather than a person, an effect they called algorithm appreciation. But this faded among experienced professionals, who relied less on algorithmic advice and became less accurate as a result. So, we can both over-trust and under-trust AI.
We should be relying on AI where it performs impressively and checking it where it does not. People are much more willing to use an imperfect algorithm when they were allowed to adjust its output slightly. Giving people a sense of control is one of the most effective ways to reduce aversion.
Should you argue with AI to get better answers?
Yes, but argue wisely. Challenging AI’s output keeps your brain engaged and typically produces better results. However, AI models tend to agree with whoever they are talking to, so simply insisting it is wrong can make it change a correct answer rather than improve a weak one.
I believe we should all be arguing with artificial intelligence. Treat what it gives you as the starting point for a discussion, as you would with a human colleague. Before using an answer, ask yourself what it might have got wrong, check at least one claim independently and put the key point into your own words. That effort is what keeps your own thinking sharp.
The complication is a problem researchers call “sycophancy”. A 2023 study by researchers at Anthropic, Towards Understanding Sycophancy in Language Models, found that AI assistants can favour responses that match the user’s beliefs over truthful ones, and that both people and the systems used to train AI sometimes prefer well-written agreeable answers to correct ones. The models tended to shift their answers towards the user’s views, even when those views were only weakly expressed, which reduced accuracy.
So how you argue matters. Rather than saying “That’s wrong”, ask AI to critique its own answer, to give the strongest case against its conclusion, or to say what evidence would change its view. Avoid revealing the answer you are hoping for. Try asking the same question in a fresh conversation to see whether you get a consistent response. And for anything important, check the facts outside the AI altogether.
Arguing in this way has a hidden benefit. The effort of questioning, comparing and explaining is precisely the kind of thinking that strengthens understanding and memory, so you end up with a better answer and a sharper mind.
Will AI affect our memory and learning?
Almost certainly, and the direction depends on how we use it. Learning relies on mental effort. When AI does the work for us, we tend to remember and learn less. When it is used like a good tutor who makes us think for ourselves, it can help us learn more.
Decades of research indicate that struggle helps us learn. We learn by making mistakes. The psychologist Robert Bjork called these “desirable difficulties”. By this is it is meant retrieving information from memory, working through problems and explaining ideas in our own words. These all build stronger, longer-lasting knowledge.
Early evidence suggests this matters. At the MIT Media Lab, researchers led by Nataliya Kosmyna used EEG headsets to monitor 54 university students writing essays, some using ChatGPT, some using a search engine and some unaided. After the first session, 83% of the ChatGPT group could not correctly quote a sentence from the essay they had just written. The researchers describe this as “cognitive debt”.
Another experiment, in 2025, gave nearly 1,000 high school maths students in Turkey access to AI tutors. A standard ChatGPT-style tutor improved practice performance by 48%, but when the AI was taken away, those students scored 17% lower in their exam than students who had never used it. A version designed to give hints rather than answers improved practice performance by 127% without harming exam results. Students who used AI also overestimated how much they had learned.
The lesson is that design and habits matter. Try a problem yourself before asking AI. Ask it to quiz you rather than tell you, to give hints rather than solutions, and to check your explanation rather than write one for you. In other words, keep practising your skills and use AI to support that practice, not replace it.
How can you use AI to build better personal and professional habits?
AI cannot form a habit for you, but it can help with the parts most people find hardest, which is turning vague intentions into specific plans.
Habits form through repetition in a consistent context. On average it takes 66 days for a behaviour to become automatic, according to research. But this ranged from 18 to 254 days, and more complex behaviours took longer. Encouragingly, missing an occasional day in trying to create a habit made little difference.
One of the most effective techniques is what psychologist Peter Gollwitzer called “implementation intentions”. These are specific if-then plans such as “When I sit down with my morning coffee, I will read for 15 minutes.” AI is excellent at helping with this. Describe your goal and your daily routine, and ask it to suggest specific triggers, break the habit into small changes and anticipate what might get in the way.
For personal habits, AI can help you choose and schedule what you want to do. For example, if you would like to rediscover reading, tell an AI tool about books you have enjoyed, ask for suggestions and put a regular reading slot in your diary. For professional habits, you could use AI to support a weekly review: paste in your notes and ask it to identify patterns, unfinished tasks or recurring problems. It can also turn new knowledge into spaced quizzes, so that learning becomes a regular habit.
There are two cautions. First, the behaviour must remain yours. For example, AI can plan your exercise, but it cannot do your press-ups. Second, be careful that chatting with AI about your habits does not become a substitute for doing them. The planning should take minutes; the habit is what counts.
How is AI changing education and assessment?
AI is changing how students study, how teachers teach and, most of all, how learning is assessed. Because AI can produce essays and answers in seconds, traditional assignments no longer reliably show what a student knows, so schools and universities are redesigning assessment.
The scale of change is striking. The Student Generative AI Survey 2026 from the Higher Education Policy Institute found that 95% of UK undergraduates now use AI in some form and 94% use generative AI to support assessed work. The proportion who include AI-generated text directly in assessed work rose from 3% in 2024 to 12% in 2026.
Detecting AI use is not a reliable answer. Many people are confident they can spot AI-generated work, but the evidence suggests otherwise, and detection software can wrongly accuse honest students.
Instead, institutions are rethinking assessment design. The University of Sydney’s influential “two-lane” approach combines “secure” assessments, which allow educators to judge student capability with confidence, with “open” assessments, which prepare students for a world where AI is everywhere and in which AI use cannot be banned. Other changes include more oral assessments and vivas, in-class writing, assessing the process of learning through drafts and reflections, and making AI literacy itself a learning outcome.
The psychology behind this is simple. Assessment drives behaviour. If assessments reward only a polished final product, students will find the quickest way to produce one, and AI makes that easy. If assessments reward thinking, judgement and the ability to explain and defend ideas, students have good reason to learn. AI is therefore pushing education to focus more clearly on what it should have been assessing all along.