A recent international study shows that many AI assistants are not reliable when it comes to sharing news. The research, coordinated by the European Broadcasting Union (EBU) and led by the BBC, looked at responses from 22 public service broadcasters across 18 countries and 14 languages. They examined over 3,000 answers from AI tools like Chat GPT, Copilot, Google’s Gemini, and Perplexity. The results are concerning: nearly half of the responses had at least one serious mistake.

AI Responses Often Contain Errors and Misinformation

The study found that 45% of all AI replies included serious errors. These could be wrong facts, outdated information, or fabricated details. About 31% of responses had sources that were either incomplete or misleading, making it hard to verify the information. Additionally, 20% of answers contained major factual errors, which could mislead people or spread false information. These errors aren’t just minor slip-ups—they can significantly distort the truth.

Google’s Gemini Performs the Worst

Among the AI tools tested, Google’s Gemini was the poorest performer. It had problems with 76% of its responses. The main issue was a lack of proper source attribution, making it difficult to trust the information it provided. This high error rate raises concerns about the reliability of AI assistants, especially as they become more common sources of news and information for the public.

The Broader Impact and What’s Being Done

The study’s findings suggest these issues are not isolated incidents. They happen across different countries, languages, and platforms. Jean Philip De Tender from the EBU emphasized that such systematic problems could undermine public trust in media and digital information. When people can’t be sure what’s accurate, they may stop trusting any news altogether, which can hurt democratic participation.

In response, the EBU and BBC launched a toolkit called “News Integrity in AI Assistants.” This guide aims to help developers and users improve AI responses and promote better media literacy. They also urged European and national authorities to enforce existing rules on information accuracy, digital services, and media fairness. The organizations called for ongoing independent reviews of AI tools to ensure they meet standards for trustworthy information.

So far, major tech companies like OpenAI, Microsoft, Google, and Perplexity AI have not responded publicly to the study’s findings. As AI continues to grow and influence how we get news, these issues highlight the need for better oversight and more reliable tools. It’s clear that while AI can be helpful, there’s still a long way to go before it can be trusted as a safe source of information.

Inspired by

• https://www.computerworld.com/article/4077344/ai-chatbots-are-wrong-about-news-45-of-the-time.html

Sources

• gmpg.org
• co.uk
• bbc.com
• podigee.com
• podigee.com