Back in 1995, Claude Steele published a study that showed that negative stereotypes could have a detrimental effect on students' academic performance. But the big surprise was that he could make that effect disappear with just a few simple changes in language. In this podcast, Radiolab revisits the topic of stereotype threat in light of the roil of replications and self-examination in the field of social psychology. Radioab speaks to Michael Inzlicht about his own experiences with the topic, including why he now has doubts about the robustness of the phenomenon.
When researchers asked people whether they’d rather get an empathetic response from a human or from AI, human responses won by a wide margin. When the same researchers actually showed them the responses, AI won by a wider one. People who said they preferred human responses rated the AI replies as more empathetic, more validating, and even better at making them feel heard.
The most recent evidence comes from a Penn State and University of Toronto team led by Joshua Wenger, published in January in Communications Psychology. Across four studies, participants were given the choice between receiving an empathetic response from a human or from ChatGPT. They picked humans 57% to 62% of the time.
Read More