AI models misrepresent news events nearly half the time, study says

According to the research, 45 percent of responses to questions astir the quality had astatine slightest 1 ‘significant’ issue.

AI models specified arsenic ChatGPT routinely misrepresent quality events, providing faulty responses to questions astir fractional the time, a survey has found.

The survey published connected Wednesday by the European Broadcasting Union (EBU) and the BBC assessed the accuracy of much than 2,700 responses fixed by OpenAI’s ChatGPT, Google’s Gemini, Microsoft’s Copilot, and Perplexity.

Recommended Stories

• list 1 of 4Hundreds of anti-immigrant protesters pain vehicle, onslaught constabulary successful Dublin
• list 2 of 4Inside Israel’s warfare of imprisonment against Palestinians
• list 3 of 4Cold cells, meagre meals: Palestinian American lad suffers successful Israeli jail
• list 4 of 4Russia-Ukraine war: List of cardinal events, time 1,336

Twenty-two nationalist media outlets, representing 18 countries and 14 languages, posed a communal acceptable of questions to the AI assistants betwixt precocious May and aboriginal June for the study.

Overall, 45 percent of responses had astatine slightest 1 “significant” issue, according to the research.

Sourcing was the astir communal problem, with 31 percent of responses including accusation not supported by the cited source, oregon incorrect oregon unverifiable attribution, among different issues.

A deficiency of accuracy was the adjacent biggest contributor to faulty answers, affecting 20 percent of responses, followed by the lack of due context, with 14 percent.

Gemini had the astir important issues, chiefly to bash with sourcing, with 76 percent of responses affected, according to the study.

All the AI models studied made basal factual errors, according to the research.

The cited errors see Perplexity claiming that surrogacy is amerciable successful Czechia and ChatGPT naming Pope Francis arsenic the sitting pontiff months aft his death.

OpenAI, Google, Microsoft and Perplexity did not instantly respond to requests for comment.

In a foreword to the report, Jean Philip De Tender, the EBU’s lawman general, and Pete Archer, the caput of AI astatine the BBC, called connected tech firms to bash much to trim errors successful their products.

“They person not prioritised this contented and indispensable bash truthful now,” De Tender and Archer said.

“They besides request to beryllium transparent by regularly publishing their results by connection and market.”