Stop Talking About AI as if It’s Human. It’s Not
In the race to make AI models ever more impressive, tech companies have taken a theatrical approach to language. They continue to talk about AI as if it were a person. Not just about AI “thinking” or “planning” – those words are already loaded – but now they are discussing the “soul” of an AI model and how models “confess,” “want,” “plan,” or “feel uncertain.”
This is not a trivial marketing approach. Anthropomorphic AI is misleading, irresponsible, and ultimately corrosive to the public’s understanding of a technology that already struggles with transparency, at a time when clarity matters most.
Research by major AI companies, intended to shed light on the behavior of generative AI, is often phrased in a way that obscures rather than illuminates. Take, for example, a recent article from OpenAI that details its work to get its models to “confess” to their mistakes or shortcuts. This is a valuable experiment that probes how a chatbot self-reports certain “bad behaviors,” like hallucinations and scheming. But OpenAI’s description of the process as a “confession” implies that there is a psychological element behind the results of a large language model.
Perhaps this comes from recognizing how difficult it is for an LLM to achieve true transparency. We have seen, for example, that AI models cannot reliably demonstrate their work in activities such as solve Sudoku puzzles.
There is a gap between What AI can generate and how it generates it, which is exactly why this human terminology is so dangerous. We could discuss the real limitations and dangers of this technology, but terms that label AI as conscious beings only downplay concerns or obscure risks.
Don’t miss any of our unbiased technical content and lab reviews. Add CNET as your preferred Google source.
AI has no soul
AI systems have no soul, motivations, feelings, or morals. They don’t “confess” because they feel compelled by honesty, any more than a calculator “apologizes” when you press the wrong key. These systems generate text models based on statistical relationships drawn from large data sets.
That’s it.
Everything that seems human is the projection of our inner life on a very sophisticated mirror.
Anthropomorphized AI gives people a false idea of what these systems actually are. And that has consequences. When we start attributing consciousness and emotional intelligence to an entity where none exists, we start trusting AI in ways it was never meant to trust.
Today, more and more people are turning to “Doctor ChatGPT” for medical advice rather than relying on licensed and qualified clinicians. Others are looking to AI-generated answers in areas such as finances, emotional health and interpersonal relationships. Some become addicted pseudo-friendships with chatbots and rely on them for advice, assuming that whatever an LLM spits out is “good enough” to inform their decisions and actions.
How should we talk about AI
When companies turn to anthropomorphic language, they blur the line between simulation and sentience. The terminology inflates expectations, incites fear, and distracts from the real issues that actually deserve our attention: bias in data sets, misuse by bad actors, security, reliability, and concentration of power. None of these topics require mystical metaphors.
Take for example Anthropic’s recent leak of its “soul document,” used to form the character, self-perception, and identity of Claude Opus 4.5. This wacky internal documentation was never intended to make a metaphysical claim – more like its engineers were riffing on a debugging guide. However, the language these companies use behind closed doors inevitably seeps into the way the general population talks about them. And once that language is retained, it shapes our thoughts about technology, as well as how we behave around it.
Or study OpenAI’s research Research into AI “plots”where a handful of rare but misleading answers led some researchers to conclude that the models were intentionally hiding certain abilities. Reviewing AI results is a good practice; it is not the case to suggest that chatbots may have motivations or strategies of their own. OpenAI’s report actually stated that these behaviors were the result of training data and certain incentive tendencies, not signs of deception. But because she used the word “plot,” the conversation turned to concerns that the AI was some sort of complicit agent.
There are better, more precise and more technical words. Instead of “soul,” talk about architecture or the formation of a pattern. Instead of “confession,” call it error reporting or internal consistency checks. Instead of talking about “blueprints” of a model, describe its optimization process. We should refer to AI using terms like trends, results, representations, optimizers, model updates, or training dynamics. They are not as dramatic as “soul” or “confession”, but they have the advantage of being grounded in reality.
To be honest, there are reasons why these LLM behaviors seem human: companies have trained them to imitate us.
As the authors of the 2021 paper “On the Dangers of Stochastic Parrots” pointed out, systems designed to replicate human language and communication will ultimately reflect this: our verbiage, our syntax, our tone, and our tenor. Resemblance does not imply true understanding. This means that the model is performing what it was optimized for. When a chatbot imitates as convincingly as chatbots now can, we end up reading the humanity in the machine, even if no such thing is present.
Language shapes public perception. When words are sloppy, magical, or intentionally anthropomorphic, the audience is left with a distorted image. This distortion benefits only one group: AI companies that profit from LLMs that appear more capable, useful, and human than they actually are.
If AI companies want to gain public trust, the first step is simple. Stop treating speech patterns as mystical beings with souls. They don’t have feelings – we do. Our words should reflect this, not obscure it.
Read also: In the age of AI, what does meaning look like?
