Are We Making AI Sound Too Human?

Artificial intelligence has become a huge part of everyday life. We hear people say that AI can “think,” “learn,” “read,” “write,” and even “create.” But according to a new study from Carnegie Mellon University, the words we use to describe AI might actually be changing the way we understand it. Historian Christopher Phillips and University of Pittsburgh researcher Alison Langmead recently published a paper examining how people have talked about computers and artificial intelligence since the 1950s. They argue that humans have been using language that makes computers sound more human for decades. The researchers call this “strategic ambiguity,” meaning that a word can have a specific technical meaning to scientists while giving the general public a much bigger idea. For example, when scientists say a computer “learns,” they might mean that it follows certain rules that help it produce better results. To most people, however, “learning” sounds like what a student does when they understand something new. Phillips says this difference is important because it can make computers seem much more similar to humans than they actually are.

Interestingly, this isn't a new problem created by ChatGPT or the rise of generative AI. Phillips and Langmead looked back at conversations about computers during the 1950s and 1960s, when computers were still relatively new. They found that scientists were already debating how human-like computers should be described. Computer scientist Norbert Wiener, for example, used the word “learning” to describe computers that followed rules and became more successful at certain tasks. To researchers, the word had a technical meaning, but to everyone else, it could sound like a computer was actually learning in the same way a person does. Today, we see the same thing happening with AI. We say that ChatGPT “writes” an essay or that an image generator “creates” a picture. Those descriptions aren't necessarily completely wrong, but they can make the process sound much more human than it really is. A computer doesn't have personal experiences, emotions, or memories in the same way people do. It processes information and produces an output based on how it was designed and trained.

The researchers aren't arguing that modern AI isn't impressive. In fact, Phillips says the opposite. Today's large language models and image generators can accomplish things that would have seemed almost impossible just a few decades ago. If someone asks an image generator to make a picture of a dog riding a pony in the middle of a New York Mets parade, for example, and the computer produces something that looks close to

what they imagined, that's an incredible technological achievement. But Phillips questions whether we need to call that “creativity.” He argues that describing exactly what the technology does could actually make its accomplishments more interesting because we wouldn't be comparing it to something humans do. This also applies to AI tests. Programs such as Massive Multitask Language Understanding, or MMLU, are used to test how well AI systems answer questions across different subjects. Another newer test is called “Humanity's Last Exam.” While strong performance on these tests can show that an AI system is very capable, Phillips and Langmead argue that the results don't necessarily prove that a computer has human-like knowledge or understanding. Getting the correct answer is not automatically the same thing as understanding why the answer is correct.

The debate becomes even more complicated when AI enters areas that people usually consider deeply human, like writing, art, and creativity. An AI can produce a poem, write a story, or generate an image in just a few seconds. But when humans read a poem, they often care about more than just the words on the page. They might wonder about the writer's emotions, experiences, or reason for creating it. A human writer has a personal history behind their work, while an AI system doesn't have those same lived experiences. That doesn't mean AI-generated writing or art has no value, but it does mean the process behind it is different. Phillips believes that recognizing that difference is important because if we constantly describe computers using human qualities, we may start seeing them as replacements for people instead of tools created by people. This could become especially important as AI becomes more common in schools, workplaces, and everyday communication.

Ultimately, Phillips and Langmead aren't asking people to stop using words like “learning” or “thinking” completely. Instead, they want people to be more careful about what those words actually mean when we're talking about technology. They also point out that conversations about computers used to involve people from many different fields, including historians, psychologists, engineers, literary scholars, and computer scientists. Today, AI affects almost every part of society, from education and jobs to art and communication, so understanding it probably shouldn't be left to technology experts alone. The bigger question isn't necessarily whether AI is becoming human. It's what AI can actually do, how it does it, and what makes those abilities different from our own. As AI continues to develop, being precise about those differences could help people understand the technology without getting caught up in either the hype or the fear surrounding it.