US study: Less educated people are more likely to use AI

OpenAI launched ChatGPT on the market around a year and a half ago. According to a recent US study, language models have since found their way into the professional world. Around a quarter of professional communication is now at least carried out with the help of artificial intelligence (AI). The level of education plays a role here.

For the study, which is available as a preprint on the document server ArXiv, a team from Stanford University, the University of Washington and Emory University examined the use of large language models (LLM) in various industries from January 2022 to September 2024. The study analyzed 687,241 complaints filed with the US Consumer Financial Protection Bureau (CFPB), 304.3 million job postings, 537,413 corporate press releases and 15,919 United Nations (UN) press releases using a statistical recognition system.

The study shows a new reality in which companies, consumers and even international organizations largely relied on generative AI for communication, the researchers write: AI assisted in the formulation of about 18 percent of complaints submitted to the CFB, 15 percent of job advertisements, 24 percent of companies and 14 percent of UN press releases.

Initially, urban and rural areas were on a par

According to the study, AI was used more in urban areas than in rural areas (18.2 percent compared to 10.9 percent) - but not right from the start: At the beginning of 2023, there was little difference between urban and rural areas. The discrepancy only emerged in the middle of the year.

It is astonishing that AI is adapted differently than new technical developments: These are generally adopted first by more highly educated population groups, he said. According to the study, they are used more frequently in regions with a lower level of education than in those with a higher level of education (19.9 percent compared to 17.4 percent). In urban areas, the difference between communities with lower and higher levels of education was even greater (21.4 percent versus 17.8 percent. The state median bachelor's degree attainment rate served as the benchmark.

"In the area of consumer complaints, the geographic and demographic patterns of LLM use represent a startling departure from historical trends in technology diffusion, where technology adoption has generally been concentrated in urban areas, higher-income groups, and populations with higher levels of education," the study states. According to the study, the use of AI could benefit people with a lower level of education. AI tools could therefore create a balance.

Proof is difficult

Proving whether AI has been used in writing is not so easy. The systems used for this are not always reliable. However, if the database is large enough, patterns can be identified that indicate the use of AI. According to the researchers, recognition is more difficult if the AI-generated text has been edited again. In addition, the language models are becoming increasingly better, which also makes recognition more difficult. AI use could therefore be more widespread than the study suggests.

The authors warn of the negative effects of excessive use of AI tools: "In sensitive areas, blind trust in AI can result in messages that do not address concerns or provide less credible information overall. Blind trust in AI can also lead to public distrust of corporate communications."

(wpl)