When AI Speaks for Humans, Whose Voice Are We Really Hearing?

Artificial intelligence is becoming part of almost everything we do, from writing and searching for information to analyzing data. Because of this, I was interested in the Pew Research Center article “Can AI Stand in for Human Survey-Takers? Not Really.” What stood out to me most was not simply that AI made mistakes, but what those mistakes tell us about the difference between predicting people and actually listening to them.

Pew researchers asked an AI model to act like real survey participants using information about those participants. They then compared the AI-generated answers with responses from actual people. Across nearly 300 survey questions, the AI-generated results differed from the human results by an average of about 12 percentage points. The researchers also found that AI sometimes exaggerated characteristics associated with particular groups and did not represent the full diversity of opinions found among real people. Pew Research Center

One finding that especially interested me was how AI handled uncertainty. Human respondents were about four times as likely as the AI model to select a “not sure” response when that option was available. This makes me think about something that happens in education as well. Students sometimes believe that being knowledgeable means always having an answer. However, admitting that we are unsure can actually be an important part of learning. A student who says, “I don’t understand this yet,” can ask questions, investigate, and eventually develop a deeper understanding. If AI tends to give confident answers even when uncertainty would be more realistic, students need to learn to question that confidence.

This also connects to disciplinary literacy. Being literate in subjects such as statistics, social science, or research means more than simply reading the numbers presented to us. We need to ask: Where did these numbers come from? Who was included? How were the questions asked? What assumptions were made? A chart can look professional and convincing while still giving us an inaccurate picture of reality. In an age when AI can quickly generate statistics, explanations, and even simulated opinions, knowing how to examine methods and sources may become even more important than simply knowing how to find information.

I also wonder what happens to human experiences when AI tries to represent entire groups of people. No demographic category can perfectly predict someone's beliefs, interests, experiences, or opinions. Pew found examples where the AI treated attitudes associated with certain demographic or political groups as much more common than they actually were. To me, this demonstrates why real people still need opportunities to speak for themselves rather than having a machine decide what they probably think.

This leaves me with a question: As AI becomes better at imitating human language and behavior, how can we teach students to recognize the difference between information that represents real human voices and information that only predicts what those voices might say? Maybe one of the most important forms of literacy in the future will be learning not only how to use AI, but also when not to trust it.