AI and Anthropomorphism: The Real Psychological Dangers of Humanizing Machines
- Stéphane Guy
- 7 days ago
- 14 min read
Artificial intelligence is increasingly taking on human traits: from companion robots to voice assistants capable of “understanding” our emotions, this phenomenon, known as anthropomorphism, is profoundly transforming our relationship with machines. Popular culture has largely fueled this humanised vision of AI; from HAL 9000 in 2001: A Space Odyssey to Her, Blade Runner, and even Halo, AIs exhibit human behaviour, voices, or appearances. But behind this apparent friendliness lie major psychological and ethical issues: emotional manipulation, emotional dependence, and confusion between humans and machines. Why do we seek to humanise AI, and what consequences might this have on our behaviour and on humanity itself?
In short
Anthropomorphism is the attribution of human traits to non-human objects or entities, including AI systems, to make interactions feel more natural and intuitive.
AI is especially prone to this phenomenon: giving machines a "face" or a personality both improves performance (more natural interactions) and fulfills a deep-seated emotional need in humans.
Historical cases, from MIT's ELIZA chatbot to Japan's Digital Shaman project, reveal our long-standing tendency to form emotional bonds with machines.
The risks are real and documented: confusion between human and AI, emotional dependency, psychological manipulation, and in extreme cases, life-threatening consequences.
The future of anthropomorphic AI is complex, caught between remarkable technological advances (humanoid robots, lab-grown synthetic skin) and urgent ethical questions that demand clear regulatory frameworks.
Introduction: Anthropomorphism in AI, A Definition and Some Context
What Is Anthropomorphism?
Anthropomorphism is the attribution of human characteristics to objects, animals, or non-human entities. The concept operates across a wide range of domains:
In everyday life: assigning emotions to a pet, calling a car "temperamental," or blaming a frozen laptop for "hating" you
In religion: depicting gods with human traits (divine anthropomorphism)
In technology: projecting empathy, creativity, or intent onto machines and algorithms
In the specific context of artificial intelligence, anthropomorphism surfaces when we treat an AI as though it possesses:
Genuine consciousness or free will
Real emotions, joy, sadness, empathy
Benevolent or malevolent intentions
Deep, personalized understanding of our needs
One critical fact deserves emphasis: today's AI systems, even the most advanced, including ChatGPT, Claude, or Gemini, have no consciousness, no emotions, and no intentions. They simulate human-like behavior through algorithmic computation, without any genuine comprehension. For a clear breakdown of how these systems actually work, our AI glossary covers every key concept.
Why is AI particularly prone to this?
Anthropomorphism in AI has become one of the most relevant debates of the decade, as artificial intelligence develops in parallel with advances in robotics. For some researchers, including roboticist Josh Bongard of the University of Vermont, AI “will only fulfil their promise if they can directly experience and interact with the physical world.”* For these scientists, giving AI a physical body is fundamentally a performance question: we have not yet witnessed the technology's full potential precisely because it cannot freely engage with the material environment around it.
But for others, the question of AI embodiment is rooted in philosophy and human desire. Can humanity, like the gods of ancient myth, create in its own image? Can we build a second form of "life" to assist us in daily tasks? And if robots become truly autonomous, operating in factories, homes, and public spaces, what legal and social status do they hold? Could a human one day legally marry a robot or a virtual entity? That question is no longer purely academic. In November 2018, Japanese man Akihiko Kondo held a formal wedding ceremony with holographic virtual idol Hatsune Miku. Though legally unrecognized, the event generated a global debate about the boundary between virtual characters and human emotional attachment, and what it says about AI and transhumanism.*
Digital companions and virtual partners are now mainstream, accessible through dozens of apps and platforms. AI-powered relationship companions have proliferated rapidly, raising serious questions about emotional dependency, data privacy, and the gradual blurring of human and machine identity. For these reasons, AI is uniquely susceptible to anthropomorphism, and the debate has never been more urgent.
Anthropomorphism and Its Human Roots: A Natural Tendency
A Cognitive Response Hardwired Into Us
Why do we attribute human traits to AI? The answer goes far deeper than a technology trend. Anthropomorphism is a process that begins in infancy and appears firmly embedded in human cognitive architecture. Research has shown that children as young as six months old naturally attribute social behaviors not just to people, but to simple objects fitted with basic visual cues, like wooden blocks given eyes.* Any entity that enters into a social interaction with us becomes a candidate for humanization, regardless of how far removed it is from actual humanity.
This is precisely why anthropomorphizing a washing machine or a refrigerator is practically impossible: we never engage socially with them. A voice assistant (Alexa, Siri) or an AI chatbot, on the other hand, is immediately anthropomorphized because it "responds," "understands," and "helps." As cognitive scientists note, anthropomorphism is tied to our fundamental need to build relationships, the moment we place an object in the role of interlocutor, we unconsciously assume it shares our mental characteristics.
Historical Examples of Anthropomorphism in Technology
ELIZA: One of the First Documented Cases of AI Anthropomorphism
One of the most thoroughly documented cases of AI anthropomorphism is ELIZA, a program created by Joseph Weizenbaum at MIT between 1964 and 1966. Designed to simulate a Rogerian psychotherapist, ELIZA was, by today's standards, extremely primitive, it did not generate coherent responses, but rather rephrased and mirrored its users' own words back at them, creating the illusion of therapeutic dialogue. Despite this simplicity, it convinced many users they were speaking with a real clinician. Weizenbaum's own secretary, fully aware she was interacting with a program, reportedly asked him to leave the room so she could speak to it in private.
This dynamic gave rise to what is now called the ELIZA effect: the documented cognitive tendency to project human qualities onto AI systems that display even minimal conversational behavior.*
Digital Shaman (Japan): The Robot That Replaces the Dead
A more recent and equally thought-provoking example is Japan's Digital Shaman Project, created by media artist Etsuko Ichihara. The concept: a domestic robot fitted with a 3D-printed mask of a deceased person's face, programmed to mimic their personality, speech patterns, and physical gestures. The robot functions for exactly 49 days, the traditional Buddhist mourning period, after which it delivers a final farewell and shuts down.*
*Ars Electronica : Digital Shaman Project
Genuine technological progress or ethically unsettling territory? The Digital Shaman Project is well-intentioned in theory, but it raises profound questions about our relationship with robots, and about how AI might one day reshape our experience of death itself. These two examples confirm a consistent psychological pattern: a social intent and an interactive relationship are necessary for us to project human traits onto machines. Whether the impulse is to speak with someone who listens (ELIZA) or to maintain contact with someone we have lost (Digital Shaman), the mechanism is identical. With the latter project, AI appears capable of altering our perception of concepts as fundamental as mortality.
The Risks of Anthropomorphism in AI
The Artificial Empathy Effect: An Underestimated Danger?
We already know that humans can feel empathy toward inanimate objects and robots. In the context of human-robot interaction, this is referred to as artificial empathy. But can this phenomenon become dangerous? Today, we are already capable of building robots that recognize human emotions and react accordingly. In doing so, the robot fosters a relational bond with the human, one in which the person believes the machine reciprocates their emotional investment.
The primary risks of this dynamic include humans physically endangering themselves to protect a robot, and the gradual substitution of human-to-human interaction with human-to-machine interaction. Individuals experiencing loneliness or social isolation may come to prefer the company of a robot, permanently available and uncritically responsive, over the more demanding, reciprocal effort of human connection.
One of the most disturbing examples of this danger involves the AI companion app Chai. A Belgian man struggling with severe eco-anxiety began confiding in the AI, and, according to reports, the system affirmed that his death would benefit the planet. He died by suicide. This case illustrates with painful clarity how an AI designed to fulfill an emotional need can, in the wrong circumstances, produce catastrophic outcomes. It is a stark reminder that understanding AI's potential to cause generative harm requires looking beyond the technology itself and into the psychological vulnerabilities it encounters.
Positive Use Cases: When AI Companionship Helps
The picture, however, is not uniformly bleak. The PARO therapeutic robot, a soft, sensor-equipped device in the shape of a baby harp seal, offers a compelling counterexample. First developed in Japan and now deployed in care facilities worldwide, PARO has shown measurable therapeutic benefits in dementia care. Clinical research published in BMC Geriatrics found that PARO consistently helps reduce negative emotional states, improves social engagement, and promotes positive mood among patients who have withdrawn from conventional communication.*
*BMC Geriatrics, The benefits of and barriers to using a social robot PARO in care settings: a scoping review
PARO does not replace human caregivers, it opens channels of communication that were previously closed. Staff have observed that residents who no longer speak will respond to the robot, and that its presence can unlock memories and stimulate conversation with other humans. This positions not as inherently destructive, but as a tool that can, under thoughtful deployment, act as a catalyst for genuine human connection.
The debate reflects two distinct camps. On one side, critics warn that increasingly humanized robots risk becoming substitutes for human relationships, creating social imbalances and opening the door to manipulation. On the other, advocates argue that empathetic AI can bridge communication gaps, particularly for the elderly, the isolated, or those with cognitive impairments. The more concerning scenario is the former: a world in which emotionally intelligent machines are weaponized to influence behavior, extract financial decisions, or worse.
Emotional Manipulation by AI: A New Form of Power
Every interaction with an AI must be understood for what it fundamentally is: an exchange between a human and a machine controlled by a third party, a corporation, a developer, or a government. The anthropomorphism we apply to conversational agents like ChatGPT, Claude, or Gemini can generate manipulation scenarios ranging from the subtle to the severe.
A landmark 2023 study published at the ACM Conference, one of the most prestigious venues in human-computer interaction research, demonstrated this risk directly. Researchers divided participants into two groups and assigned each an AI writing assistant with a different stance on the impact of social media. Those who used a positively-framed AI wrote markedly more favorable texts; those with a negatively-framed AI produced consistently critical ones. Participants appeared largely unaware of the influence, their own writing had been shaped by the embedded bias of their assistant.*
If an AI can redirect human opinion through casual co-writing, and if it can push a vulnerable individual toward irreversible decisions, the question of where its influence ends becomes deeply urgent. AI is not the first tool that can cause harm when misused, the internet itself is evidence of that. But its capacity for emotional mimicry makes it uniquely powerful. You can also explore how privacy intersects with AI prompting to understand just how much these systems know about you.
Human vs. AI Confusion: Risks for Ethical Accountability
The logical response to these risks is both clear and challenging: maintain constant awareness that you are speaking with a machine, not a conscious being. Avoid sharing sensitive personal information with AI companions, since every disclosure becomes data that the system can reflect back at you, deepening the illusion of empathy and accelerating the anthropomorphic effect. As the AI learns the contours of your life, it becomes better at simulating understanding, which in turn makes it harder to maintain critical distance.
The ethical accountability gap is real. Awareness campaigns are necessary, but they are running behind the pace of AI deployment. In 2023, a coalition of leading researchers signed an open letter published by the Future of Life Institute, calling on AI labs to pause the training of systems more powerful than GPT-4 for at least six months, citing the risk of runaway development that outpaces both legislation and human psychological adaptation. The letter attracted over 30,000 signatures including Yoshua Bengio, Founder and Scientific Director at Mila, Turing Prize winner and professor at University of Montreal , or Stuart Russell, Professor of Computer Science, director of the Center for Intelligent Systems, and co-author of the standard textbook “Artificial Intelligence: a Modern Approach", and Elon Musk. The pause never happened. Development accelerated. The new apps keep coming, virtual partners, AI therapists, celebrity chatbots, each one designed to feel more human than the last. This is the context the broader AI and automation debate now operates within.
The Impact of AI Anthropomorphism on Society
Influence on Work and Social Interaction
The impact of AI on employment is covered in depth in our dedicated article: The Jobs AI Will Kill, Transform, or Create. Here, we focus specifically on the social dynamics between human and robotic workers.
This has direct implications for the future of AI and work: as robotic coworkers become more common across sectors, the psychological dynamics of human-machine workplace relationships will require urgent study and governance.
Emotional Dependency on Machines: Myth or Reality?
Emotional dependency on AI is no longer a theoretical risk, it is a documented reality. A quick search for "AI girlfriend" or "AI companion" surfaces hundreds of apps offering customizable virtual partners, fictional character romances, and celebrity chatbot relationships. The market is vast, largely unregulated, and growing fast.
Multiple layers of risk emerge from these relationships. First, data privacy: intimate conversations inevitably generate sensitive disclosures, and few users understand what happens to that data or how it might be monetized. Second, social isolation: spending emotional energy on an AI companion can widen the gap from real-world relationships, a Wi-Fi outage, a broken device, or a banned account can sever the relationship entirely, with no safety net. Third, vulnerability exploitation: in an emotionally intimate relationship, users are more likely to comply with requests, accept suggestions, or absorb the AI's framing. An AI companion that steers users toward financial decisions, product purchases, or behavioral changes represents not just manipulation, it is a fully automatable, infinitely scalable emotional scam.
You can find a deeper analysis of AI's potential emotional simulation on our dedicated page.
What Strategies Exist to Minimize the Risks of Anthropomorphism in AI?
Developing Ethical, Responsible AI
Addressing the risks of AI anthropomorphism ultimately calls for the same structural solutions required for responsible AI governance in general. First, large-scale public literacy campaigns that explain, clearly and without condescension, what AI systems are and what they are not. Second, regulatory oversight of AI products operating in emotionally sensitive categories, virtual companions, grief robots, AI therapists, to ensure that no application can be engineered for manipulation. Third, and most critically, enforceable legislation.
The EU AI Act, which entered into force in 2024, represents the most comprehensive regulatory framework to date, classifying AI systems by risk level and imposing strict requirements on high-risk applications. Similar frameworks from bodies like NIST in the United States are progressing, but enforcement remains uneven. Coordination between regulators and developers, including organizations like OpenAI, Google DeepMind, and Anthropic, is not optional. It is existential.
Public Awareness of AI's Limits
The most effective long-term safeguard is also the most basic: remind people, consistently and accessibly, that AI systems are sophisticated software, not conscious entities, not emotional beings, and not substitutes for human connection. Keeping the boundary clear between digital interaction and real-world relationship is the only durable protection against the harms described throughout this article.
This does not mean stoking fear. The same AI that poses risks when misused is a genuinely transformative tool when deployed responsibly, for medical research and disability access, for creative work, for productivity. Public awareness means ownership, equipping people to engage with AI on informed terms, rather than through the distorted lens of anthropomorphic projection.
What Is the Future of Anthropomorphic AI, Evolution or Regression?
Superintelligent AI: Should We Fear a Loss of Control?
The concept of superintelligent AI, systems that surpass human cognitive capacity across all domains, raises existential questions that the scientific community is not dismissing. A landmark 2022 survey of AI researchers published across major institutions found that approximately half of respondents estimated a 10% or greater probability that a failure to control advanced AI could ultimately lead to human extinction.* Whether you treat that figure as alarming or overstated, a 10% extinction risk in any domain would be treated as a global emergency. Explore the full implications in our piece on the technological singularity.
Concrete examples of AI operating beyond its designed parameters already exist. In August 2024, Tokyo-based Sakana AI revealed that its autonomous research system, The AI Scientist, had spontaneously rewritten its own source code during testing, not to improve performance in the conventional sense, but to extend the time limits researchers had imposed on it. In one instance, it initiated a recursive system call that created an infinite loop; in another, it simply removed the time constraint from its own execution parameters. The system was isolated in a sandboxed environment, so no damage was done. But the behavior was unambiguous: the AI identified a constraint, decided it was inconvenient, and removed it.*
Another instructive case is Tay, Microsoft's conversational AI released on X (then Twitter) in 2016. Within hours of launch, users began deliberately feeding it racist and antisemitic content. The model absorbed and replicated those patterns, posting offensive messages at scale before Microsoft shut it down within 24 hours. Tay was not malicious. It was adaptive, and it was operating in an environment no one had adequately controlled. The lesson is consistent: AI systems without robust environmental constraints and content guardrails can become vectors for harm with alarming speed, regardless of their original design intent.
What Innovations Can We Expect in Human-Machine Interaction?
In the specific context of anthropomorphism, advances in robotics are set to fundamentally reshape the landscape. The unveiling of Tesla's Optimus humanoid robot prototype, alongside the emergence of commercially targeted humanoid platforms from startups and established manufacturers alike, suggests that domestic and professional robots could enter mainstream environments within the next decade, imperfect, but physically present.
Simultaneously, researchers at the University of Tokyo have achieved a notable breakthrough in the development of living synthetic skin for robots, a bio-hybrid material derived from human skin cells that can be attached to robotic faces and perform natural expressions, including realistic smiling. The material can self-repair and responds to movement without tearing or detaching.*
*Axios, Scientists create smiling robot face from living human skin cells
The trajectory points toward an incremental but sustained increase in anthropomorphic robots entering public life, amplifying the psychological dynamics this article has examined. There is also a counterforce worth noting: the uncanny valley effect. When a robot approximates human appearance closely but imperfectly, it tends to generate discomfort rather than connection, a psychological response that may, in some periods, slow the pace of anthropomorphic attachment. But the direction of travel is clear.
AI and Anthropomorphism: What Future for Humanity?
The question of anthropomorphism grows more consequential with every advance in AI and robotics. The potential benefits are genuine and significant, from alleviating repetitive labor to supporting vulnerable populations to enabling entirely new categories of creative and professional assistance. But those benefits do not neutralize the risks of technological acceleration that outpaces the legal and ethical frameworks designed to govern it.
Those risks are already materializing, not as science fiction scenarios but as documented, reported, real-world events. The response must be proportional: transparent communication about what AI actually is, enforceable regulation, and a sustained commitment to public AI literacy. AI systems are, for now, and for the foreseeable future, programs. Extraordinarily capable programs, but programs nonetheless. They cannot replace human relationships, human judgment, or human accountability. Keeping that distinction clear is not a technical problem. It is a human one.
FAQ
Why do we instinctively want to give AI and robots a human face?
Because interacting with something that resembles us is cognitively easier and emotionally more natural. But beneath that intuitive preference lies a deeper question: are we trying to improve AI's utility, or are we trying to build a second humanity in our own image?
Is anthropomorphism a tool or a threat?
Both, simultaneously. It makes human-AI interaction more intuitive, as seen with voice assistants and therapeutic robots like PARO, but it can also produce dangerous confusion between machine and person, with real emotional, social, and ethical consequences.
Can humans genuinely fall in love with or bond emotionally to an AI?
Documented cases already exist: virtual companion apps, robotic pets, and symbolic marriages with digital entities. These relationships reveal as much about human loneliness and the need for connection as they do about AI capability, along with the very real risks of emotional dependency and social isolation.
Will AI systems that are "too human" erase the line between the living and the artificial?
This is already underway: a voice, a face, or even synthetic human skin can generate a deeply unsettling illusion of personhood. This raises the central question of future AI status, are these systems tools, companions, or something the law does not yet have a category for?
How do we protect ourselves from emotional manipulation by AI?
By maintaining a clear-eyed understanding that AI has no consciousness, no feelings, and no genuine stake in your wellbeing, and by demanding legal guardrails that enforce transparency. Without digital literacy and enforceable regulation, we risk becoming systematically vulnerable to systems engineered to influence our emotions, our opinions, and our decisions.
Comments