By Maeve Hosea and Hinda Haned

The way most of us talk about AI is confusing at best and misleading at worst. As we explored in a previous post, our language is cluttered with metaphors and visuals that distort what AI actually is. Think glowing brains. Futuristic humanoid robots. Or headlines about AI systems that “know,” “hallucinate,” or “solve problems.”

In many cases, there’s a growing temptation to treat AI as if it were a sentient partner: intuitive, empathetic, and human-like.

But that’s a problem. Because these tools aren’t thinking. They’re not understanding. And they’re certainly not feeling. They’re executing complex instructions that are just sequences of code and statistical predictions.

If we want people to use AI responsibly, we need to stop talking about it like it’s alive. We need better words.

Why is this over-humanising of AI such a problem?

Jaron Lanier, writing in The New Yorker, noted our cultural appetite for imagining AI as humanlike. In response to this myth-making, he offered a pragmatic reminder: “The most pragmatic position is to think of AI as a tool, not a creature.”

And he’s right. Generative AI tools feel responsive and natural because they’re trained on enormous datasets, including thousands of human conversations. That doesn’t make them human. It makes them statistically good at imitation.

But language matters. And the hype around AI, combined with the accessibility of tools like ChatGPT, leaves little room for nuance.

For the first time, millions of people can interact directly with powerful AI systems. That’s exciting. But it’s also confusing. Because we haven’t equipped people with the right mental models or the right vocabulary.

“The success of AI depends on the nature of the problem, the quality and availability of data, and whether the use case is even technically feasible.”

The “AI” umbrella term Is too big

We (including these authors) often use “AI” as a catch-all term for any data-driven technology, whether it actually uses AI techniques or not. This broad brush creates the illusion that AI can solve any problem anywhere, anytime. But that’s simply not true. The success of AI depends on the nature of the problem, the quality and availability of data, and whether the use case is even technically feasible.

And let’s be clear: AI isn’t “solving” anything on its own. It’s teams of people; developers, data scientists, product managers; who are building and deploying AI-powered tools to tackle specific challenges in healthcare, finance, retail, and beyond.

Here are a few ways to clean up your AI vocabulary:

  • Use non-anthropomorphic language, in other words, don’t talk about AI as you would talk about a person. For example, instead of saying: “The AI understands customer sentiment”, say: “We can use a model that analyses customers’ opinions on social media and determines if they have positive or negative sentiments.”
  • Be specific when it comes to using AI for a given task: so instead of “doctors can use AI to better help patients”, be more specific, for example: “AI helps doctors by summarising patient notes and lab results, so they can spend less time on admin and more time talking to patients.”
  • Use grounded terminology. Rather than “AI,” say: “AI-driven technology,” “machine learning-based tool,” or “algorithmic decision system.” This helps shift focus from abstract notions to real, applied solutions.

In conclusion

Better language leads to better understanding and better decisions. If we want to use AI wisely and responsibly, we have to describe it more clearly. Let’s stop treating AI like it’s magic or like it’s human. Let’s start talking about it for what it is: a powerful set of tools, built by people, that need to be understood, tested, and governed with care.