Patients in Rotherham, South Yorkshire, report that an AI receptionist system is unable to understand broad regional accents [1].
The failure of the technology creates a barrier to healthcare access. When patients cannot navigate the automated system to book appointments, they may delay necessary medical treatment.
GP practices in the area have implemented the EMMA AI system to handle incoming calls. However, the software has struggled to process the specific phonetic patterns of the Yorkshire dialect [1], [2]. This technical gap has led to significant frustration among local residents who find the system unresponsive to their speech [3].
Some patients have opted to abandon the booking process entirely due to these difficulties. One patient said, "I ended up just hanging up and not bothering to try and book an appointment" [4].
Healthwatch Rotherham said the issue highlights the disconnect between the AI's training and the reality of the local population's speech [1]. The system was not adequately trained on the region's broad accents, which resulted in the misrecognition of speech [1], [5].
While AI is intended to streamline administrative tasks for doctors, the implementation in Rotherham shows a lack of linguistic diversity in the software's development. Patients continue to encounter errors that prevent them from reaching human staff, or securing time slots with their physicians [3].
“"I ended up just hanging up and not bothering to try and book an appointment."”
This incident highlights a critical flaw in the deployment of healthcare AI: algorithmic bias in speech recognition. When AI systems are trained on standardized or limited datasets, they can inadvertently marginalize populations with strong regional dialects. In a medical context, this is not merely a technical glitch but a public health risk, as it creates an uneven distribution of access to care based on how a patient speaks.
