Predictive text blunders: Smartphone users report AI-driven errors
Translated from English, summarized and contextualized by DistantNews.
At a glance
- Smartphone users report a significant decline in the accuracy and usefulness of autocorrect and predictive text features.
- Experts suggest that recent AI integration in operating system updates may be responsible for the perceived drop in performance, despite increased sophistication.
- The issue is widespread, with users on forums for major phone companies detailing errors ranging from nonsensical word suggestions to incorrect capitalization.
Autocorrect and predictive text on smartphones seem to have lost their way, leaving users frustrated with nonsensical suggestions and persistent errors. From "bus" becoming "butks" to "no longer sensible" transforming into "mi longer shiny sync," the experience can be baffling.
There's been a demonstrable decline in 'old-fashioned' predictive text, as well as several deliberate patches/updates from companies to fix them.
Online forums for major brands like Samsung and Apple are filled with complaints. Users report autocorrect changing correct words to wrong ones, repeating mistakes even after retyping, failing to capitalize "I," and inserting random words. Some describe the software as inserting "tondel" instead of "to" or consistently changing "the" to "Theresa."
Experts point to the integration of artificial intelligence, particularly transformer models similar to those used in ChatGPT, starting around 2023. PhD candidate Leon Furze notes a "demonstrable decline" in older predictive text models, which were replaced or augmented by AI. He states that companies like Apple and Samsung followed similar patterns, adding more AI into their systems.
Whether there has been an objective deterioration is difficult to establish, as these systems have become substantially more sophisticated over time.
While objectively measuring the deterioration is difficult due to the systems' increasing complexity, senior lecturer Morteza Namvar acknowledges "considerable anecdotal evidence and growing public discussion" about the declining quality. Users report encountering more "unexpected corrections and unusual suggestions" than in earlier versions of the technology.
Users did appear to encounter more 'unexpected corrections and unusual suggestions' than they did with earlier predictive text technologies.
Originally published by ABC Australia in English. Translated, summarized, and contextualized by our editorial team with added local perspective. Read our editorial standards.