Why We Need to Humanise AI

Ultra-realistic human portrait representing the humanisation of artificial intelligence

As artificial intelligence becomes more powerful, making it more recognisable, relatable and human may be essential to helping people feel comfortable living alongside it.

Artificial intelligence is becoming extraordinarily capable. It can write software, analyse medical information, generate video, translate languages, hold complex conversations and solve problems that only a few years ago would have seemed beyond the reach of a machine. Stanford University's 2026 AI Index describes a technology reaching mass adoption faster than the personal computer or the internet, with generative AI reaching nearly 53% population-level adoption within three years.

But something else is accelerating alongside AI intelligence: anxiety about it. Stanford's 2026 public opinion analysis found that 59% of people globally said AI products and services offered more benefits than drawbacks, yet 52% also said AI made them nervous. Those figures are not contradictory. They capture perhaps the defining relationship we currently have with artificial intelligence. We are increasingly impressed by what it can do while becoming increasingly uncertain about what it might become.

In the United States, Pew Research Center found in 2026 that 52% of adults were more concerned than excited about the growing use of AI in everyday life, compared with just 9% who were more excited than concerned. In Great Britain, the Office for National Statistics found that only 13% of adults thought AI offered more benefits than risks, while 38% believed the risks outweighed the benefits. A UK government survey found that 74% of adults worried about AI producing inaccurate information, 69% were concerned about privacy and security, and 63% were concerned about AI-related job losses. AI clearly has an image problem, but perhaps it also has a human problem.

Intelligence Without Humanity Can Feel Intimidating

For most of human history, intelligence has been something we associate with another person. Intelligence came with a face, a voice, facial expressions, hesitation, emotion, humour, mistakes, body language and personality. Artificial intelligence separates those things, leaving us able to communicate with something that may have access to more information than any individual person while having no physical presence, no expression and no recognisable emotional state.

There is an important psychological difference between knowing that a machine is useful and feeling comfortable with that machine. Pew Research Center found in 2025 that 73% of Americans were prepared to allow AI to assist them at least a little with everyday activities, yet only 13% were willing to let it assist them a lot. Some 57% said they had little or no control over whether AI was being used in their lives, and 61% wanted greater control. Adoption does not automatically equal trust.

AI developers have spent enormous amounts of money making artificial intelligence smarter. The next challenge is not simply another increase in benchmark performance. It is making increasingly powerful artificial intelligence something humans can comfortably understand, communicate with and live alongside. That means humanising AI.

At the Moment, AI Has No Identity

There is another part of the problem that receives far less attention. To most people, artificial intelligence is still an abstract concept. They might use ChatGPT. Another person might use Claude. Someone else might interact with Gemini or an AI system built into a retailer, bank or airline. The underlying technologies, companies and models may be very different, but psychologically they are frequently grouped together under one enormous label: AI.

And AI itself has no face, no single personality, no recognisable character and often no obvious identity. It is simply an intelligence somewhere behind a screen. For somebody who does not follow the technology closely, the distinction between one large language model and another may mean very little. They are interacting with an unknown technological power whose capabilities appear to be increasing extremely quickly, and that anonymity matters.

Humans are accustomed to associating information, actions and decisions with identifiable people or organisations. If a colleague gives us advice, we know who gave us the advice. If a television presenter tells us something, we recognise the presenter. If we walk into a hotel and ask the concierge a question, we are interacting with an identifiable individual. There is a point of association. AI often removes that point of association completely. You ask a question and an answer appears, but psychologically there is no familiar person or character attached to it.

Of course, there is a company and a technological system behind it, and responsibility ultimately has to remain with the organisation operating the AI. Giving an artificial intelligence a face or a name should never be used to disguise that responsibility. But there is a difference between responsibility and identity. A consistent AI character can provide identity.

Give an AI a name, a recognisable voice, a face, a personality and consistent behavioural characteristics and suddenly the interaction has an anchor. Instead of speaking to an invisible and largely undefined intelligence, we are speaking to a recognisable character. We know who we are addressing, and humans build relationships through recognition and familiarity. Over time, we learn someone's mannerisms, voice, expressions and personality. An anonymous intelligence offers none of that, while a recognisable AI character can.

This does not mean users should believe the character is a real person. Quite the opposite. It should be completely transparent that the character is powered by artificial intelligence. But there is an enormous psychological difference between saying, "I am talking to AI," and saying, "I am talking to this particular AI character." The first describes a technology. The second creates a point of recognition and association.

Imagine arriving at a shopping centre and speaking to an anonymous voice generated by a computer system. Now compare that with approaching a recognisable AI guest ambassador who has a name, face, voice and consistent personality. Both may be powered by exactly the same underlying artificial intelligence. The technological capability may be identical, but the human perception of the interaction is not.

Humanising AI Does Not Mean Pretending It Is Human

There is an obvious danger in this argument. Humanising AI should never mean deceiving somebody into believing that artificial intelligence is actually a person. The objective should not be to hide the machine behind a human face. People should understand when they are interacting with artificial intelligence, particularly when decisions, advice or personal information are involved. Humanisation should be about how AI communicates and behaves, not about concealing what it is.

This distinction matters because there is already evidence that humans respond differently to exactly the same information when they believe a person rather than an AI generated it. A 2025 study in Nature Human Behaviour involving 6,282 participants presented people with empathetic responses generated by AI but described those responses as coming either from humans or from AI. When participants believed the responses came from another person, they rated them as more empathetic and supportive, despite the underlying responses being the same.

Another study of more than 1,400 participants found an equally interesting effect. AI-generated healthcare responses were rated as more empathetic than physician responses overall, yet responses believed to have been written by physicians received an additional boost in empathy, trust and liking regardless of who had actually written them. The technology may therefore already be capable of communicating empathetically, while the psychological hurdle remains that people still value the feeling of humanity behind that communication.

Humans Are Social Creatures. AI Needs to Understand That.

Human communication is extraordinarily complicated. The words we speak are only one part of it. We communicate through tone, rhythm, pauses, facial expressions, gestures, gaze, posture, interruption, silence and emotional context. A slight hesitation can suggest uncertainty. A smile can change the meaning of a sentence. A pause can signal that somebody is thinking. Looking away, leaning forward or changing tone can completely alter how another person interprets what we say.

AI conversation has traditionally stripped away most of those signals. Ask a conventional AI system a complicated question and it may immediately produce a perfectly organised answer. From an engineering perspective, that is impressive. From a human perspective, it can also feel unnatural. Research into human conversation has found that gaps between conversational turns are often only around 200 milliseconds, while longer pauses carry meaning of their own. Timing itself is part of human communication.

This does not mean AI should simply insert artificial delays before every answer. Sometimes the right response should be immediate. Sometimes a fraction of hesitation makes sense. Sometimes an acknowledgement should come before the answer. Sometimes an AI should recognise that somebody has not finished speaking, and sometimes the socially intelligent thing to do is remain silent. The goal should not always be the fastest possible response. It should be the most natural one.

We Should Not Have to Learn How to Talk to Machines

There is another reason humanisation matters. At the moment, humans frequently change their own behaviour to accommodate technology. Research published in Scientific Reports found that both adults and children altered the way they spoke when addressing a smart speaker compared with speaking to another person. They produced more effortful speech, including longer duration and higher pitch, and children changed their speech even more after the device deliberately misunderstood them.

Think about how strange that is. We have created incredibly sophisticated machines and then taught humans to modify their natural behaviour so those machines can understand them. The better direction is the opposite. Technology should increasingly adapt to us. People should be able to hesitate, change their mind halfway through a sentence, interrupt, laugh, mumble, ask an ambiguous question, point at something or refer back to something they said five minutes earlier. AI should learn the rhythms of human conversation rather than forcing people into the rigid rhythms of machines.

The Evidence Suggests Human-Like AI Changes How We Respond

There is now considerable research into what psychologists call anthropomorphism, the tendency to attribute human characteristics to non-human things. A major 2025 meta-analysis examined 800 effect sizes from 199 datasets across 142 papers, representing 41,642 participants. It found an overall positive effect from introducing human-like social cues into text-based conversational agents, with improvements across areas including perception, rapport, trust and positive emotion.

The effect was not enormous, and that is important. Simply giving a chatbot a friendly name or making it use conversational language will not suddenly eliminate public concerns about artificial intelligence. Humanisation is much deeper than adding a photograph beside a chat window, but the direction of the evidence is significant.

Another study involving more than 12,000 open-ended responses from a representative sample of Americans tracked the metaphors people naturally used to describe AI. Researchers found that perceptions of AI as warm and human-like increased after the arrival of ChatGPT and that perceptions of warmth, competence and human-likeness were associated with greater trust and willingness to adopt AI. Once a machine communicates with us, social psychology begins to matter.

Humanising AI Is More Than Giving It a Face

This is where the debate around humanoid robots becomes particularly interesting. Humans have a strange instinct when designing robots. We often give them heads, faces, eyes, arms and hands even when none of those features is technically necessary for the machine to complete its task. A warehouse robot does not require a human face to move a box. A cleaning robot does not need legs. A manufacturing robot does not need to smile.

Yet when machines are expected to communicate socially with humans, anthropomorphic characteristics often begin appearing. There is a reason for that. The human body is an interface we have spent our entire lives learning to understand. Research into human-robot interaction broadly supports the value of anthropomorphism, but it also offers an important warning: benefits are not universal, and making a machine look more human does not automatically make it safer, more empathetic or better at its task.

Trying to make a machine too human can create the opposite reaction. The phenomenon is commonly known as the uncanny valley. As robots or digital characters move closer to human appearance, small imperfections can become more noticeable rather than less. An almost-human face with slightly unnatural eyes, movement or expression can be considerably more unsettling than something that clearly looks artificial. This is why humanising AI cannot simply be a race towards creating an artificial person who is visually indistinguishable from a real one. A machine does not have to fool us to connect with us.

Warmth May Matter as Much as Intelligence

For decades, computer interfaces have been designed around efficiency. Faster is better. Fewer clicks are better. More information is better. Social interaction does not work quite like that. When humans interact with one another, competence is only one dimension of the relationship. We also make almost instantaneous judgements about warmth, intention, attentiveness, personality and trustworthiness.

This distinction will become particularly important as AI moves away from being something we occasionally type questions into and becomes something we encounter throughout physical life. Imagine an AI concierge in a hotel, an assistant inside a shopping centre, an information service at an airport, an educational guide in a museum or an AI representative inside a retail store. In each environment, raw intelligence is only part of the experience.

The AI needs to understand when somebody is confused. It needs to recognise when a person wants to interrupt. It needs appropriate facial reactions. Its tone should change depending on the situation. Its personality should remain consistent. Its body language should correspond with its words, and it should be able to acknowledge uncertainty instead of confidently pretending to know everything. That is when AI begins to feel less like an unknown intelligence and more like something humans intuitively know how to communicate with.

Human Presence Can Change the Relationship

Physical or visual embodiment may become especially important as artificial intelligence enters public spaces. Research comparing an embodied humanoid conversational agent with a conventional smart speaker in a work scenario found that the embodied agent was potentially more trusted and produced greater engagement and task performance. Its facial features, expressions and eye gaze provided social information that a disembodied speaker could not.

This does not mean every AI needs a body. It means there are environments in which presence matters. A text box is an excellent interface when somebody is sitting at a computer. It is less natural when somebody walks into a hotel lobby, an airport terminal or a shopping centre and needs immediate assistance. In those situations, giving AI a recognisable presence can make the technology easier to approach.

A name gives the intelligence identity. A voice gives it personality. A face adds expression. Eye contact creates attention. A body provides gesture and direction. Movement adds emotional information. Over repeated interactions, those characteristics also create familiarity. People are no longer repeatedly approaching an anonymous technological system. They are returning to a character they recognise.

A holographic AI avatar takes this idea further by giving artificial intelligence the impression of physical presence without requiring a mechanical robot to imitate a biological human. The important part, however, is not the hologram itself. It is what happens when identity, appearance, voice, movement, intelligence, personality and conversational behaviour work together.

There Is a Fine Line Between Humanisation and Manipulation

There is also a responsibility that comes with making AI more human. Modern conversational AI can already mimic human-like conversation, persuasion and empathy extremely effectively without possessing genuine empathy or social understanding. If humanisation increases trust, developers have a responsibility not to exploit that trust.

AI should not pretend to experience emotions that it does not experience. It should not deliberately create emotional dependency. It should not manipulate vulnerable users into believing that a commercial system genuinely loves, needs or cares for them. Nor should an AI character become a convenient shield that obscures who is actually responsible for the system behind it. A character can provide familiarity and identity, but accountability must always lead back to the organisation deploying the technology.

The objective is not to make somebody forget they are speaking to AI. The objective is to make speaking to AI feel comfortable. Those are very different goals. Transparency, controllability and accuracy still matter enormously, and a friendly face cannot fix an AI system that repeatedly gives the wrong answer.

The More Powerful AI Becomes, the More Important This May Become

The debate about artificial intelligence is often framed as a race towards greater intelligence. Can AI reason better? Can it code better? Can it see? Can it create video? Can it use computers autonomously? Can it eventually outperform humans across almost every intellectual task? Those questions matter, but they may not determine how comfortable society eventually becomes with artificial intelligence.

AI adoption is already happening remarkably quickly, yet public concern remains high. The Pew Research Center has documented widespread use alongside continuing scepticism about how AI is developed, governed and used. That should concern the AI industry because we are building technology faster than society is becoming comfortable with it.

Making AI even more intelligent will not necessarily solve that problem. In some circumstances, it may make it worse. An increasingly powerful intelligence that remains anonymous, cold, invisible, unpredictable or difficult to understand may feel more threatening, not less. Humanising AI provides a different path.

Give it an identity without pretending that identity is a biological person. Give it warmth without deception. Give it personality without pretending it is conscious. Give it natural conversational timing without theatrical fakery. Allow it to listen, pause, react, remember context and understand social signals. Where appropriate, give it a name, voice, expression, gaze and physical presence. Most importantly, design it around humans rather than expecting humans to redesign their behaviour around machines.

AI Does Not Need to Become Human. It Needs to Become More Human

Perhaps the most important technological challenge of the next decade will not be teaching machines how to think. It may be teaching them how to interact with the people they are thinking alongside. Artificial intelligence does not need to become human, and we should never pretend that it is, but as AI becomes more capable and more deeply embedded in everyday life, giving it recognisably human characteristics, including a stable identity, could play an important role in making that transition feel less intimidating and more natural.

At Miirage, this is one of the ideas behind bringing AI out of the traditional screen and giving it human presence in the physical world. A holographic avatar is not valuable simply because it resembles a person. Its real potential emerges when identity and appearance are combined with natural voice, conversational rhythm, expression, movement, personality and artificial intelligence.

Instead of interacting with an anonymous intelligence somewhere in the cloud, people can interact with a recognisable character standing in front of them. They can know its name, recognise its face, become familiar with its voice and understand its personality while still knowing that it is AI. The technology behind AI may be artificial, but the people interacting with it are not. If artificial intelligence is going to become part of everyday human life, perhaps it is time we started designing it accordingly.