One in five Americans wrongly believes AI is conscious

A study conducted with 3,500 representative respondents revealed that one in five Americans is convinced that artificial intelligence possesses consciousness. The detailed results of the study were published on the arXiv website, a platform for publishing scientific articles before peer review.

The study was conducted by Jacy Reese Anthis from the Sentience Institute in New York, alongside a team of scientists. They surveyed a representative sample of 3,500 people in the USA, inquiring about their perceptions of AI and its consciousness. The survey, executed in three waves between 2021 and 2023, included questions like "Do you think any currently existing robots/AI are conscious?" and "Will it ever be possible for this technology to achieve consciousness?"

The study's authors explained that their goal was to gather information about the public's knowledge of artificial intelligence to understand better how public opinion might influence the future development of AI.

In 2021, about 18% of respondents expressed the belief that existing AI systems or robots were already conscious. This number rose to 20% in 2023, when the subsequent waves of the study were conducted. Additionally, 30% of Americans believe that there already exists artificial intelligence capable of performing any task typically done by a human. Surprisingly, one in ten people asked in 2023 believed that ChatGPT, launched at the end of 2022, is conscious.

Scientists emphasize that the technologies currently referred to as "artificial intelligence" are quite simple and it cannot be said that AI possesses consciousness. The fact that society believes otherwise is troubling because it might lead people to trust "opinions issued by AI" more than those from humans.

Jacy Reese Anthis expresses concern that society has too easily fallen for tech companies' marketing strategies, which often exaggerate artificial intelligence's capabilities.

"Big tech companies are selling their products as having greater abilities than the underlying technology suggests. I think people have focused too much on the term ‘artificial intelligence’," said Anthis, quoted by New Scientist. The scientist reminded that the term "artificial intelligence" was coined in the 1950s.

"People are often impressed by how AI models perform on IQ tests or standardized exams for humans. But that’s often a very bad way of thinking about these models because AI repeats answers found in its huge database, on which it was trained, instead of actually ‘knowing’ anything," Anthis stated.

Scientists also point out the media's role in shaping public perception of AI.

"The type of media coverage we've seen around large language models does not help with this, with overly excited and panic-stricken reports about existential threats from superintelligence," said Kate Devlin from King’s College London, also quoted by New Scientist.