A 12-month federally-funded CCTV trial using AI technology to detect falls and screams in two South Australian aged care homes generated over 12,000 false alerts, overwhelming staff and resulting in at least one actual fall going unattended.
A federally-funded CCTV trial was conducted at two South Australian aged care homes (Mount Pleasant District Hospital and Northgate House) over 12 months beginning in March 2021. The $785,000 trial used artificial intelligence technology designed to detect falls and screams through cameras and audio devices installed in common areas and bedrooms (with consent from 41 of 57 residents). The AI system was programmed to capture video and audio and alert staff when excessive sounds or movements were detected. However, the system generated over 12,000 false alerts across both sites, with many triggered by staff crouching or bending down to care for residents. The high volume of false alerts overwhelmed already overworked staff, and in the final months of the trial, staff could not respond to every alert. This resulted in at least one instance where staff did not respond to an actual fall. Despite improvements over time, with the system detecting 22% of actual events in the final three months compared to less than 2% initially, an independent audit by PwC found the AI technology was 'not yet sufficiently accurate at detecting incidents in a residential aged care setting' and did not achieve acceptable accuracy levels for staff and management.
Domain classification, causal taxonomy, severity scores, and national security assessments were LLM-classified and may contain errors.
AI systems that fail to perform reliably or effectively under varying conditions, exposing them to errors and failures that can have significant consequences, especially in critical applications or areas that require moral reasoning.
AI system
Due to a decision or action made by an AI system
Unintentional
Due to an unexpected outcome from pursuing a goal
Post-deployment
Occurring after the AI model has been trained and deployed