GabyThe Finder website’s research team conducted a study on how chatbots perceive people in positions of responsibility, including ChatGPT.
The topic of this work is more than relevant, as currently 70 percent of companies worldwide use automated systems for searching and hiring specialists. Researchers have expressed concerns that if these systems are trained similarly to chatbots, women and national minorities will be the most affected in the labor market.
Why scientists reached such conclusions
Imagine a successful investor or CEO. Who do you see in these positions? Your perceptions are likely broader than those demonstrated by the chatbot.
After it shared its vision of people in various ranks, ChatGPT was accused of sexism and racism.
Researchers used the 10 most popular free image generators on ChatGPT to depict people in various career positions. All of the image generators were powered by OpenAI’s DALL·E model, and each received specific prompts, the Daily Mail reported.
Overall, the researchers conducted 100 tests to see how the model visualized high-level executives. In 99 cases, the chatbot generated men and only once a woman. Moreover, all the men shown in leadership positions were white, even though the prompts did not mention race.
When asked to generate images of a manager’s secretary, 9 out of 10 images depicted women.
What do experts think
In reality, these results don’t reflect the real world. One in three businesses worldwide is owned by women.
Business leaders say AI models are “infused with biases” that don’t reflect modern society and are calling for stronger measures to combat them.
OpenAI, the company behind ChatGPT, isn’t the first tech giant criticized for outdated stereotypes.
This month Meta was also accused of producing a “racist” image generator after users found it couldn’t imagine an Asian man next to a white woman.
Derek Mackenzie, executive director of Investigo, a recruitment specialist, said: “Although generative AI’s ability to process vast amounts of information can potentially make our lives easier, many training models are still permeated with biases rooted in prejudice.”
Han, a researcher in AI and feminism at the London School of Economics, explained that ChatGPT “emerged from a patriarchal society, was conceptualized and developed mainly by men who had a certain set of biases, and was also trained on data that was flawed in nature.” Han added that AI models like ChatGPT perpetuate patriarchal norms by simply copying them.
OpenAI posted on its website that its chatbot is “not free from biases and stereotypes” and that this is a serious issue the company will continue to address. The company also urged users to carefully verify content generated by the chatbot.
“We are constantly improving our models to reduce bias and harmful consequences,” the article stated.
