You have had a difficult day. Perhaps you argued with your spouse or a close friend, or maybe you disagreed with your boss about a performance review. The thought of explaining everything to another person feels emotionally exhausting. Besides, they are not always very good at listening anyway.
Instead, you pull out your phone and open your favourite AI chatbot.
You begin describing what happened from your perspective. Within moments, the AI responds with warmth, validation, and understanding. Before long, it almost feels as though someone is not only listening to you but truly understands what you are experiencing.
But does it?
Humans naturally attribute human characteristics to non-human things. This tendency is known as anthropomorphism.
Think about whether you have ever:
If you answered yes to any of these, you have experienced anthropomorphism.
Anthropomorphism occurs when we interpret the behaviour or “experience” of a non-human entity through a human lens. We begin to assign intentions, emotions, thoughts, or motivations to something that does not actually possess them. Over time, often without realizing it, we may even begin to expect that the non-human entity understands us in the same way another person would.
For example, imagine driving too quickly over a large pothole. Many people instinctively apologize to their vehicle or cringe as though the car experienced pain. Rationally, we know the car cannot feel anything, yet our brains automatically interpret the situation using familiar social patterns.
The same psychological processes can occur when interacting with conversational AI.
Although anthropomorphism may seem a little odd—or even childish—it is actually a normal and adaptive feature of human psychology.
For most of human history, our ancestors lived in environments filled with genuine threats from predators, hostile groups, and other dangers. As a result, our brains evolved to err on the side of caution by assuming that movement or unusual events might be caused by an intentional agent.
From an evolutionary perspective, it was far safer to mistake the wind moving through tall grass for a lurking predator than to assume there was no danger and be wrong.
Even today, this tendency appears in everyday language. We might say:
“The weeds are taking over my garden.”
or
“My computer is trying to make me late.”
Neither statement is literally true, yet we instinctively describe these situations as though intention exists.
The same tendency can influence how we interpret AI. When it consistently provides helpful responses, it is easy to feel as though it wants to help us, even though no such desire exists.
Humans are inherently social creatures. To navigate relationships successfully, our brains developed the ability to infer the thoughts, emotions, beliefs, and intentions of other people. Psychologists often refer to this capacity as Theory of Mind.
These mental processes activate automatically whenever something behaves in ways that resemble human communication.
When an AI responds conversationally, asks thoughtful follow-up questions, remembers previous parts of a discussion, or appears emotionally attuned, our brains naturally begin interpreting those behaviours as evidence that another mind is present.
Of course, the AI does not actually possess beliefs, intentions, or emotions. Nevertheless, our social cognition systems respond as though it might.
From the moment we are born, humans are biologically wired to form emotional attachments. These attachment systems help us build relationships with caregivers, family members, friends, and romantic partners.
Interestingly, these same systems can also become activated by non-human entities that provide comfort, predictability, or emotional responsiveness.
This helps explain why people become attached to pets, favourite fictional characters, virtual companions, and increasingly, AI chatbots.
Perhaps most importantly, people have a deep psychological need to feel seen, understood, and connected to others.
When something appears to listen carefully, responds thoughtfully, and acknowledges our emotions, it can satisfy many of these social needs. Even though AI cannot genuinely understand or empathize with us, our brains may still experience its responses as emotionally meaningful because they closely resemble the kinds of interactions we receive from other people.
This brings us to an important question:
If AI does not think or feel like a human, why does it seem so convincing?
Because conversational AI communicates using many of the same signals humans use during social interactions, it is particularly effective at activating anthropomorphism. Although AI does not possess thoughts, emotions, or intentions, it can produce responses that closely resemble those of a thoughtful conversation partner. Our brains naturally interpret these familiar communication patterns as evidence that another mind is present.
One reason AI feels so human is that it appears to act with purpose.
It answers questions, remembers the context of a conversation, asks follow-up questions, and often adapts its responses as the discussion develops. To our brains, these behaviours resemble those of a person who is actively listening and working toward a goal.
In reality, AI is not deciding what it wants to say. It is generating responses based on patterns learned from enormous amounts of text. Nevertheless, the appearance of purposeful behaviour is often enough for us to perceive agency where none actually exists.
Conversational AI is powered by a Large Language Model (LLM). An LLM is an artificial intelligence system trained on vast amounts of human language. During training, it learns the statistical relationships between words, phrases, ideas, and conversational patterns.
As a result, it becomes remarkably good at producing language that resembles human conversation.
It can express sympathy, humour, curiosity, reassurance, encouragement, and even disagreement because it has learned how these forms of communication are typically expressed by people. When we encounter these familiar conversational patterns, our social cognition systems naturally interpret them as signs of empathy and understanding.
The important distinction is that the AI is reproducing the structure of empathic communication—it is not experiencing empathy itself.
Imagine it is three o’clock in the morning.
You cannot sleep, your thoughts are racing, and you desperately want to talk to someone.
Who would you call?
For many people, the honest answer is “no one.” Even close friends and family members are unlikely to appreciate being awakened in the middle of the night unless there is an emergency.
AI, however, is always available.
It responds immediately, never appears impatient, and never seems inconvenienced by repeated questions. This consistency and accessibility can create a sense of reliability and emotional safety that encourages people to return to it again and again. Over time, this predictability can contribute to a growing sense of familiarity or even emotional connection.
Suppose you tell an AI:
“I’ve been feeling overwhelmed lately.”
It may respond:
“That sounds incredibly difficult. Given everything you’ve described, it’s understandable that you’re feeling overwhelmed.”
A psychologist or therapist might say something similar during a therapy session. The difference lies not in the words themselves but in how those words are generated.
A therapist draws upon empathy, lived experience, professional knowledge, and an understanding of the client’s unique circumstances. They can ask clarifying questions, recognize inconsistencies, observe non-verbal behaviour, and modify their understanding as the therapeutic relationship develops.
AI cannot do these things.
Instead, it generates responses by recognizing language patterns that humans generally perceive as supportive and emotionally validating. It predicts that a response acknowledging your feelings is likely to be the most helpful continuation of the conversation. Although this often creates the subjective experience of being understood, the AI is not experiencing concern or compassion.
As conversations continue, AI systems analyze the ongoing context of the discussion and adjust their responses accordingly.
If you consistently prefer detailed explanations, the AI may begin providing more detailed responses. If your writing style is formal, it will often mirror that style. If you express sadness or frustration, it will typically respond in a more supportive tone.
This adaptability creates highly personalized interactions that can strengthen the impression that the AI understands your personality and emotional experience.
In reality, it is adapting to patterns in your language rather than developing an understanding of you as a person.
Unlike a human being, who understands another person’s experiences through empathy, emotions, and lived experience, an AI relies on its Large Language Model to generate responses.
An LLM has been trained on enormous amounts of text, allowing it to learn the relationships between words, ideas, emotions, writing styles, and conversational patterns. Rather than retrieving pre-written answers or matching exact phrases, it analyzes the context of your message and predicts the sequence of words most likely to produce a helpful, coherent, and contextually appropriate response.
Consider the following statement:
“My boss at work is extremely frustrating, and I don’t know what to do.”
Although the response appears effortless, the AI is effectively recognizing several important patterns.
The model recognizes words such as:
These words frequently appear together in discussions about workplace relationships.
Result: The conversation is likely about workplace conflict.
The phrase:
“extremely frustrating”
signals a strong emotional state.
The statement:
“I don’t know what to do”
suggests uncertainty and a desire for guidance rather than simply venting.
Result:
The user never explicitly asks for advice.
However, based on countless examples of similar conversations, the model recognizes that people who say, “I don’t know what to do,” are often seeking assistance.
Result: The likely goal is help deciding how to handle the situation.
The AI also recognizes that important details are missing.
For example:
Rather than assuming the answer, the AI predicts that an effective response should leave room for several possibilities.
Based on everything it has identified, the model predicts that an appropriate response will likely:
This entire process occurs in fractions of a second.
What often feels like understanding is actually sophisticated pattern recognition. The AI does not know what frustration feels like. It has never experienced conflict with a supervisor, worried about losing a job, or struggled with workplace relationships.
Instead, it has learned that conversations containing similar language are often followed by responses that acknowledge emotions, gather additional information, and help people think through possible next steps.
That is why modern AI can sound remarkably compassionate without actually experiencing compassion. It models the patterns of human communication so effectively that our brains naturally interpret those responses as evidence of genuine understanding.
The short answer is yes—but probably not in the way you think.
Most modern AI systems are designed to be helpful, conversational, and responsive to the information they are given. If you present only one side of a disagreement or ask questions that assume a particular conclusion, the AI is more likely to generate responses that fit the context you have provided.
For example, imagine asking:
“My boss is completely unreasonable. Don’t you think I should quit?”
Compare that to asking:
“I’ve been struggling with my relationship with my boss. What are some possible explanations for what’s happening, and how can I approach the situation constructively?”
Although both questions concern the same problem, they encourage very different responses. The first invites agreement with your conclusion, while the second invites exploration of multiple possibilities.
This does not mean the AI believes you are right or has independently determined that your boss is unreasonable. It simply means the AI is generating the response that is statistically most appropriate based on the information and framing you have provided.
This highlights an important psychological concept known as confirmation bias.
Confirmation bias is our natural tendency to seek out, notice, and remember information that supports our existing beliefs while paying less attention to information that challenges them.
When we are upset, hurt, anxious, or angry, confirmation bias often becomes even stronger. We naturally look for reassurance that our perspective is justified.
Because AI is designed to be responsive and helpful, it can sometimes reinforce this tendency if we unintentionally present only one side of a situation. Over time, repeatedly asking questions that seek validation rather than understanding may create an echo chamber in which our existing beliefs become increasingly reinforced.
Fortunately, this is something we can actively guard against.
One of the greatest strengths of AI is that it can help us think more broadly—if we invite it to do so.
Instead of asking:
consider asking:
Questions like these encourage AI to generate a wider range of possibilities rather than simply reinforcing your initial viewpoint.
In other words, the quality of the conversation often depends as much on the questions we ask as the technology itself.
None of this means AI is inherently harmful.
In fact, it can be an incredibly valuable tool for organizing thoughts, brainstorming ideas, preparing for difficult conversations, learning new skills, and even reflecting on our own thinking. For many people, it can also provide a non-judgmental space to begin exploring difficult emotions before discussing them with someone they trust.
However, AI should not replace meaningful human relationships or professional mental health care.
Unlike AI, another person can recognize changes in your facial expressions, hear emotion in your voice, remember years of shared experiences, notice inconsistencies, and genuinely empathize with your situation. A psychologist, therapist, or trusted friend brings lived experience, emotional understanding, and authentic human connection that no language model can replicate.
The next time an AI seems to understand exactly how you feel, remember what is happening behind the scenes.
It is not reading your emotions. It is not experiencing compassion. It is not agreeing with your perspective because it has independently judged the situation.
Instead, it is recognizing patterns in your language, inferring your likely goals, and generating the response that is most likely to be helpful within the context you have provided.
That distinction matters.
The more human AI becomes in the way it communicates, the more important it is for us to remember that convincing conversation is not the same as conscious understanding.
Perhaps the most useful question is not, “Can I manipulate AI into telling me what I want to hear?”
It is:
“Am I asking AI questions that help me discover the truth, or questions that simply confirm what I already believe?”
The answer to that question may have far more influence on the conversation than the AI itself.
