Artificial intelligence devices, such as ChatGPT, Gemini, and Perplexity, are gradually establishing themselves as preferred sources of information. The appeal of these technologies lies in their ability to provide quick answers. A recent study conducted by a consortium of public radio and television stations highlights an alarming precariousness of the information they transmit. The results reveal frequent major errors, thus compromising the accuracy of the content provided. Experience shows that, in a world saturated with information, the reliability of AI assistants remains highly questionable, posing a considerable challenge in the current digital age.
Precariousness of information provided by AI assistants
A recent study, published on October 22, sheds light on the shortcomings of smart assistants such as ChatGPT, Google’s Gemini, Microsoft’s Copilot, and Perplexity regarding their ability to provide reliable information. Conducted by European public radio and television stations, this assessment reveals alarming errors in nearly 45% of the responses provided by these AIs. Outdated information, major mistakes, and fabricated details often mar their responses.
Results of a collaborative study
The test involved twenty-two public media outlets from eighteen European countries, which submitted the same current affairs questions to the AI assistants. Of approximately 3,000 analyzed responses, a significant number presented “major issues.” The study, coordinated by the European Broadcasting Union (EBU) and inspired by previous work from the BBC, highlights a worrying trend: one in five responses contained major errors.
Analysis of the performance of different assistants
According to the study, Google’s Gemini stands out for its particularly poor results. Indeed, 76% of its responses contained significant issues, far exceeding the performance of its competitors. Investigators attribute this weakness to a failing ability to correctly cite its sources. Such shortcomings call into question the integrity of the information disseminated through these technologies.
Examples of notable errors
When asked about the pope, ChatGPT, Copilot, and Gemini all responded that Francis was still in office, although he has already passed away. This lack of precision raises serious questions about the reliability of historical data provided by these systems. Similarly, an incident during Donald Trump’s inauguration revealed comical errors, with Gemini interpreting a satirical column literally. This absurd response, claiming that Elon Musk had “an erection of the right arm,” reflects a serious lack of discernment.
Growing use of AIs for information
AI assistants, however, are experiencing rapid growth, particularly among younger generations. A recent study from the Reuters Institute indicates that around 15% of users under the age of 25 consult these tools weekly to stay informed about the news. Despite the disappointing results, this trend underscores a growing reliance on new technologies for consuming information.
Perspectives for improvement
Experts in the field, such as Jean Philip De Tender and Pete Archer, warn of the urgent need for reform. AI assistants require significant evolution to ensure quality information and avoid the spread of inaccuracies. Organizations like the EBU and the BBC are considering creating a practical guide to assist both developers and users in navigating the complex landscape of AI-generated information.
For further reading on this topic, please consult the following article on the rise of AI, which discusses the implications of these technologies on our access to information.
Frequently asked questions
What are the main errors made by AIs like ChatGPT and Gemini regarding current affairs?
The main errors include outdated information, incorrect responses, and fabricated details. A study revealed that 45% of the AIs’ responses contained at least one major issue.
Why did AI assistants like Gemini score the worst in the current affairs study?
Gemini exhibited “major issues” in 76% of its responses, primarily due to its poor performance in citing sources and the accuracy of the information.
How can users ensure the reliability of information provided by AI assistants?
It is advised to always verify information obtained from AIs against reliable and recognized sources, rather than relying solely on them. Cross-checking information is essential to assess its accuracy.
Do young people often use AI assistants to obtain information? If so, what is the statistic?
Yes, according to a report, 15% of young people under 25 use these assistants weekly to summarize current affairs, highlighting their growing popularity despite their questionable reliability.
What types of content do AI assistants tend to confuse?
AI assistants may confuse factual information with humorous columns, as evidenced by the example where Gemini misinterpreted a satirical column, taking it literally.
What advice can be given for securely using AI assistants during information searches?
Use AI assistants as a preliminary research tool, but always verify with human and reliable sources before considering the information as established.
Do the results of the study on AI assistants affect their use in media?
Certainly, the study highlights the precariousness of the information provided by these tools, prompting media outlets to be cautious about their use for informing or relaying news.
