Pros & cons: Should Apple Intelligence summarize notifications?

Apple's summaries of notifications are sometimes wrong. Should a function that potentially provides incorrect information be offered at all?

Save to Pocket listen Print view
Two editor photos. Leo Becker on the left, Wolfgang Kreutz on the right

(Image: Mac & i)

4 min. read

Apple Intelligence optionally summarizes messages – so far only in English. According to user reports, there are sometimes serious errors that can completely distort the original content. Following complaints from the BBC about several incorrectly summarized headlines, Apple now has to make improvements and wants to make the AI summaries more clearly identifiable. But should such an error-prone function be offered at all?

Mehr Infos

This pro & con is from Mac & i issue 1/2025, which will be published on January 30. The new issue can be ordered now from the heise store – as a print magazine (free shipping for the time being) or as a PDF.

Leo Becker welcomes AI summaries, even if they are far from always correct.

High-frequency chat groups from family, friends, colleagues, neighbors, clubs, schools, etc. all too often cause a flood of messages on the iPhone. You can, of course, hide them, but you might miss something important. This is exactly where – comes in at the request of – Apple Intelligence: The AI model turns 57 short messages peppered with emojis into a single message. Ideally, it soberly informs you that Bernd urgently needs a garden hose or that there is another lice alarm in 1d. This does not always work perfectly: colloquial language and allusions that can only be understood by insiders can quickly lead to misunderstandings. This applies to AI models just as much as it does to us humans.

The concrete benefits make it bearable for me if Apple Intelligence sometimes gets it wrong when summarizing messages – which does happen. For really important chats or personal messages, it is advisable to read the entire history anyway. Considering the hundreds of millions of iPhone messages sent every day, the error rate so far seems manageable. Especially as the summaries on each iPhone are individual because Apple's language model does this locally on the device.

“Fake news” is undoubtedly a massive problem. However, they are not caused by sloppy AI summaries, but are deliberately and widely disseminated by malicious actors. Ultimately, there is no way around questioning information – regardless of whether it comes from the original source or Apple Intelligence provides a summary. And ultimately, as a user, I decide whether I want to activate the function at all. I can specifically exclude news. (lbe)

Wolfgang Kreutz believes that summarizing short texts using AI is basically a bad idea.

It certainly sounds tempting if I only receive a summary of a group chat instead of a firework display of messages. But it's the unsuccessful exceptions that give me a stomachache. Not every missed appointment may be dramatic, but an undisclosed or trivialized emergency might be. I also doubt that everyone involved will find it amusing if a flippant “That nearly killed me” becomes a “tragic accident”.

I am even more critical of the idea of summarizing news items that are already very abbreviated. Apple Intelligence has already proven several times that this can backfire: News items from the BBC and the New York Times have been twisted into false claims and displayed with their logos. If readers believe a false report to be true, this damages the reputation of reputable media and undermines trust. Every avoidable mistake is one too many.

Even the best language models currently have hardly any real text comprehension – in short, ticker messages lack any context anyway. False information is therefore inevitable. Since AIs are primarily optimized and trained for English, the risk even increases for other languages.

Sure, I don't have to activate the function. But if my environment takes a distorted AI message at face value, I ultimately have to deal with it. And the less often problems occur, the more people rely on AI.

I don't think Apple should offer such an automatic system – not even as an option. The well-intentioned but error-prone summaries may have the same effect as real fake news, even without the intention to deceive: they spread misinformation. (wre)

Who is right? Join the discussion!

Last at Pro & Contra: Pro & Contra: Do we need a Photoshop from Apple? (lbe)

Don't miss any news – follow us on Facebook, LinkedIn or Mastodon.

This article was originally published in German. It was translated with technical assistance and editorially reviewed before publication.