Stanford, Meta Find Chatbots Are Ready For People – But Are We Ready For GAI?
A study by the Deliberative Democracy Lab (DDL) at Stanford University and Meta suggests that the more people talk about generative artificial intelligence (GAI) and chatbots, the more receptive they are, and the more they accept the technology.
The research was conducted in a forum-type setting with participants age 18 and older from the United States, Spain, Brazil, and Germany who discussed the principles that responses from generative AI (GAI) chatbots often reflect.
The findings, which were released this week, suggest the advertising industry and companies involved in this technology could benefit by teaching consumers about the technology using public-service announcements and ad campaigns.
The groups discussed several statements. Responses were analyzed before and after discussions and then were tallied based on responses from the four countries.
The statement “chatbots replicate biases that exist in data they were trained on” gained the highest approval rating after the group participating in the study discussed the question. Spain and the United States each gave the statement an 11% approval rating, whereas Brazil gave it 2.9%, with Germany at 3.9%.
The statement that lost the most approval as a result of a discussion — “the increased use of AI chatbots will lead to students losing their ability to think critically” — decreased among participants in all countries, from -0.7% for the United States to -7.1% for Brazil.
Participants were less likely to agree that AI chatbots would lead to students losing their ability to think critically after deliberations.
The statement that received the least approval rating — “people will feel less lonely with AI chatbots” — received an approval rating of less than 50% across all four countries, except for Spain. Germany had the lowest approval before deliberations, at 31.3%, and 38.9% after.
Increases in support for the statement from all four countries ranged from 0.1% for Brazil to 10.6% for Spain. Spain was the only country with majority support from its participants after deliberations.
The DDL and Meta conducted the deliberation study with 1,545 participants from Brazil, Germany, Spain, and the United States. About 1,108 people also participated in a control group, where they did not deliberate in any discussions.
The control group completed two surveys. The main purpose of the control group was to demonstrate that any changes occur after deliberation are a result of the discussions. YouGov was commissioned to recruit nationally representative samples of the general public 18 years of age and older.
A majority of participants from each country had used ChatGPT or similar chatbots. All showed statistically significant increases in participants who used ChatGPT or similar chatbots as a result of the deliberations.
Most participants in the study thought AI chatbot’s increase efficiency through the automation of tasks saves companies time and resources.
More than 70% received an approval rating from participants in Germany, Spain, and the United States, but only around 50% from Brazil.
(7)
