Apple’s latest innovation in artificial intelligence is causing considerable controversy. Recently launched in the United Kingdom, the Apple Intelligence feature, which promises to simplify the management of notifications on iOS 18.2, finds itself at the heart of a controversy concerning the dissemination of false information. When it comes to sensitive news, trying too hard to shorten the information, Apple Intelligence ends up completely inventing it.
An interesting feature that goes wrong
Apple Intelligence represents a significant advancement in notification management on iPhone. Available on recent models (iPhone 16, 15 Pro and 15 Pro Max) equipped with iOS 18.2, this technology aims to reduce interruptions by intelligently grouping and summarizing notifications. However, the implementation turns out to be more complex than expected. Voices had already been raised regarding the wacky hallucinations of Apple Intelligence in the notification summaries, but the controversy has taken a more serious turn with recent news.
The most notable incident concerns the BBCwhere Apple Intelligence generated a fake headline about Luigi Mangione, suspect in the murder of UnitedHealthcare CEO. The feature falsely stated that Mangione had committed suicide, a completely incorrect information since the man is currently in custody. This error pushed the BBC, known for its journalistic rigor, to contact Apple directly for corrections.
The problem is not limited to an isolated case. The New York Times was also the victim of a similar distortion when the feature transformed information about an arrest warrant for Benjamin Netanyahu into a false announcement of his arrest. These incidents raise crucial questions about the reliability of AI systems in news processing.
Apple’s technical and ethical challenges
The phenomenon causing these errors is known as “hallucinations” in the field of AI. These deviations occur when the AI model generates inaccurate content, even in the face of perfectly clear source data. Tim Cook, CEO of Apple, had anticipated these difficulties when launching Apple Intelligence, admitting that the system would not be infallible while promising “very high quality”. This is the whole problem with generative AI: it is technologically incapable of admitting that it does not know or does not understand, so it invents.
The situation is all the more complex as Apple favors local processing of data on devices, in accordance with its strict confidentiality policy. This approach, while laudable from a privacy perspective, makes it difficult to collect feedback and improve the system in real time. In addition, the language and understanding model used by Apple is probably not up to par, which does not help the situation.
The similar case of the NY TimesAs we know, to try to prevent these problems, Apple had integrated specific instructions into its system, including the explicit directive “Do not hallucinate. Do not invent factual information.” Despite these precautions, incidents persist, demonstrating the complexity of the technical challenge. And yes, just because we instruct an AI not to hallucinate doesn’t mean it doesn’t do so. That would be a little too easy.
The current controversy raises fundamental questions about the balance between technological innovation and editorial responsibility. In a context where disinformation represents a major issue, the ability of an AI to unintentionally alter current information poses a real ethical challenge. For Apple, which has always emphasized the quality and reliability of its services, this situation represents a crucial test of its ability to maintain these standards in the field of AI.
It was a safe bet that controversies would emerge after the public availability of Apple Intelligence, and it is not about to stop, whether it concerns writing tools or image creation , Cupertino will be closely scrutinized, undoubtedly a little more than all the competitors. Apple, as a champion of reliability, security and careful user experience and much more observed than a company like Google which, objectively, has much more to be ashamed of from this point of view with a Gemini which could sometimes generate anything, including in Google Search results. It’s unfair but that’s how it is.
What is the solution for Apple? At this point, we see two. The first would be to simply improve the language model to really stop him from hallucinating — ideally before April and the release of Apple Intelligence in Europe. The second, more radical, consists of deactivating notification summaries on sensitive apps such as news or health. What do you think?
