Apple Says It Will Clarify That Its Bad Notification Summaries Are AI-Generated

22 hours ago 1

Apple will soon push an update to iOS that more clearly denotes when notifications have been summarized by its Apple Intelligence software. The company has received significant pushback over the summarization feature, especially from the BBC, which has complained that the AI technology inaccurately summarized its news notifications.

Back in December, not long after Apple rolled out some of the first major features of Apple Intelligence, the BBC wrote in an article that the AI had erroneously summarized a series of notifications from its app to say that Luigi Mangione, the alleged murderer of UnitedHealthcare CEO Brian Thompson, had shot himself when in fact he had not.

The BBC said that the error could hurt trust in a media industry that is already suffering from a lack of confidence from the general public.

After several weeks, Apple finally responded to the BBC to say that it would be addressing the issue.

“A software update in the coming weeks will further clarify when the text being displayed is summarization provided by Apple Intelligence,” the company told the BBC. “We encourage users to report a concern if they view an unexpected notification summary.”

Screenshot from an iPhone showing a notification from the BBC that is summarized incorrectly.A notification from the BBC app that includes an incorrect summary. Credit: BBC

It is unclear how Apple will clarify to users when a notification has been summarized. As it functions today, an app’s official icon is displayed next to the summaries. Perhaps when iOS does summarize a notification, it should replace the official icon—in this case for the BBC app— with the icon for Siri instead, to make the distinction clear. That may not even necessarily be enough, though. How often might a novice user read these summarized notifications and not understand what is going on?

The point of notification summaries is to rewrite multiple notifications from an app into one single alert so that users can quickly get up to speed and identify anything that needs attention. But besides blatant errors, AI summaries have also been the subject of mockery over sometimes awkward or useless summaries devoid of necessary context that still require reading the full notifications to understand. If users do not trust the summaries and have to review the complete notifications anyway, they are not saving any time. It is the same issue that befell Siri before it: The voice assistant failed at basical queries and users quickly lost faith they could rely on it for much of anything besides starting a timer.

ai summaries be like pic.twitter.com/KbmRecoU06

— Devin Lewtan (@DevinLewtan) January 4, 2025

Apple has been criticized for its debut of Apple Intelligence, with sentiment across social media suggesting that many people find the initial features largely useless or harmful to the iOS experience. In December, the company added support for ChatGPT in Siri, meaning some more queries are now kicked over to the OpenAI chatbot to answer. Apple is expected to beef up Siri itself with conversational AI abilities, but recent reports suggest that will take time. It also intends to add other AI features to iOS, including the ability for Siri to interface with apps directly allowing users to, for instance, ask the assistant to find a dinner reservation in their email.

The driver for Apple releasing Apple Intelligence is clearly out of a need to juice iPhone upgrades; the smartphone accounts for more than half of Apple’s revenue, and Apple Intelligence is only available on the iPhone 15 Pro and above. But most of the features released thus far have been underwhelming—like the ability to identify real-world objects by pointing the camera at them—and recent surveys suggest consumers are not buying new iPhones to get them.

It should be no surprise to anyone who has interacted with generative AI that these problems were going to occur. But it seems uncharacteristic of Apple to incorporate technology into its products like generative AI, which is notoriously hard to control and prone to hallucinating or producing outputs that are flat-out wrong. For a company that cares so much about polish, this is a mess so far.

Read Entire Article