Apple is facing mounting criticism over its artificial intelligence (AI)-driven news summary feature that has produced inaccurate and misleading headlines on its latest iPhones.
The BBC first flagged the issue in December, when a news alert summary that looked to come from the BBC app wrongly reported that Luigi Mangione, charged with alleged killing of UnitedHealthcare’s CEO, had shot himself.
But the BBC said Apple did not respond until Jan. 6, saying that it was working to clarify that summaries came from the AI. However, Apple repeated the offense on Jan. 3. This time, Apple’s news summary said Luke Littler had won the PDC World Darts Championship before the contest even began. There have also been other instances.
“These AI summarizations by Apple do not reflect — and in some cases completely contradict — the original BBC content,” the BBC said on Jan. 6. “It is critical that Apple urgently addresses these issues as the accuracy of our news is essential in maintaining trust.”
PYMNTS reached out to Apple and the BBC for comment but did not get replies in time for publication.
Other publications have also experienced similar hallucinations from Apple’s AI. For example, in November, it sent an alert purportedly from The New York Times that said Israel Prime Minister Benjamin Netanyahu had been arrested, according to a post on Bluesky by ProPublica journalist Ken Schwencke.
Apple AI notification summaries continue to be so so so bad
— Ken Schwencke (@schwanksta.com) November 21, 2024 at 2:22 PM
Vincent Berthier, head of the technology and journalism desk at Reporters Without Borders (RSF in French), called on Apple to remove this feature from its phones.
“The automated production of false information attributed to a media outlet is a blow to the outlet’s credibility and a danger to the public’s right to reliable information on current affairs,” he said on Dec. 18 on the group’s website.
RSF said such incidents prove that “generative AI services are still too immature to produce reliable information for the public and should not be allowed on the market for such uses.”
The hallucinations are not relegated to just news, either. Apple’s AI also summarizes texts and chats, with sometimes hilarious results. For example, Andrew Schmidt’s mother said that a hike had almost killed her. The AI summarized it as an attempted suicide by Pamela J. Schmidt, according to an Oct. 25 post on X.
My mom: That hike almost killed me!
Apple’s AI summary: pic.twitter.com/fyN5UopIdH
— Schmidt (@AndrewSchmidtFC) October 25, 2024
Apple isn’t the only tech giant grappling with challenges in GenAI tools.
Google has faced criticism for its AI-driven search summaries called AI Overviews, which also produced erratic and inaccurate responses last year. For example, it recommended eating at least one small rock per day because rocks are a vital source of minerals and vitamins, according to a May 23 post on X.
I couldn’t believe it before I tried it. Google needs to fix this asap.. pic.twitter.com/r3FyOfxiTK
— Kris Kashtanova (@icreatelife) May 23, 2024
Meanwhile, OpenAI’s Whisper transcription software would invent texts.In addition, Meta’s Galactica LLM for scientific research made up fake papers, among other hallucinations, according to MIT Technology Review.
The propensity of AI systems to be inaccurate underscores a pervasive challenge to businesses looking to adopt GenAI. Consider the 2022 case of Air Canada, whose chatbot told a traveler that he could buy a full-priced ticket to attend his grandmother’s funeral and then apply for the discounted bereavement fare later, according to the BBC. However, the chatbot made up the policy.
When the traveler, Jake Moffatt, tried to get his discount, the airline told him the chatbot was wrong and refused him a partial refund. Air Canada said the chatbot was its own legal entity and responsible for its own actions. He took the airline to a tribunal, and Air Canada was ordered to pay around $800.
But despite the hazards of hallucinations, GenAI is likely to stay, given the productivity gains it brings. AI foundation model developers, which are mainly Big Tech companies, are working on ways to mitigate hallucinations, including using high quality data, limiting responses, adding a human in the loop and using techniques like Retrieval Augmented Generation (RAG), among others. OpenAI also developed a method that rewards each correct step of reasoning in a model’s change-of-thought process when solving math problems, instead of just the final answer.
The post Apple Under Fire for Fake AI-Generated News Summaries appeared first on PYMNTS.com.
