Mild cognitive impairment, also known as Alzheimer's disease, typically develops rapidly and quietly in 15% to 20% of individuals aged 65 and above. Early detection is crucial as these individuals are at a high risk of developing dementia.
Researchers from Apple Inc. and the University of Tübingen propose that the usage patterns of iOS applications can predict cognitive impairment. They have developed a machine learning approach to uncover pattern differences between users with and without cognitive impairment. The model achieved an area under the receiver operating characteristic curve of 0.79, indicating that it can correctly identify symptomatic individuals approximately 80% of the time.
"In many people's lives, smartphones are ubiquitous, making them a rich source of information about individuals' psychological and cognitive states," the research team wrote in their report. "Here, we investigate to what extent [application usage patterns] contribute to people's cognitive health."
The scientists' unsupervised model automatically discovers the representations required for data classification, identifying different types of interactions from sequences of application usage. These representations are then passed to a separate cognitive health prediction model. The entire framework is carefully designed and interpretable, making the relationship between applications, such as messaging, and cognitive health strength evident. The diagnosis heavily relies on the structure of user behavior.
The system first segments usage during interaction sessions to group together applications opened from the time the phone is unlocked to when it is locked. To encode the similarity between these applications, researchers trained an algorithm to predict which application a person is likely to launch based on the preceding three and subsequent three applications. The applications are represented using embeddings (i.e., numbers), which are averaged together to obtain a single representation for each session. The representations are then clustered to identify different session types, and a series of data points are indexed and summarized based on session time and type to represent the user's application usage. This serves as input for cognitive health prediction indicators.
The research team conducted a study involving 113 elderly participants, of which 31 were diagnosed with cognitive impairment and 82 were not. These participants contributed 12 weeks of phone usage data. The researchers analyzed the four session types that contributed the most to the model's decision and visualized the differences between applications within each session, as well as the overall distribution of the 15 most common applications.
Interestingly, the researchers found that the session types most closely associated with high symptom scores were primarily determined by the following applications: calls and phone, messages and mail, mail and Safari, followed by clock and calendar. On the other hand, the types corresponding to low scores were dominated by messages, Safari, mail, and Facebook. Overall, sessions with high usage rates of messages and mail or mail and Safari significantly improved the model's symptom prediction score, while sessions with a high number of messages or Safari sessions or sessions with mail and Facebook decreased the score.
The researchers noted that for participants with higher symptom scores, applications such as phone, calendar, and clock made significant contributions, in contrast to lower-scoring applications such as messages, Instagram, and camera. However, the impact of applications like messages or mail seemed to depend on the surrounding applications within the session. For example, when messages shared a session with mail or Safari, it greatly increased the prediction score, whereas when messages existed alone or in a session with Facebook or Instagram, it decreased the prediction score.
The researchers acknowledge several potential limitations in their work, including a small sample size and the inclusion of participants who were already diagnosed with symptoms upon entering the study. Nevertheless, they state that the differences observed between healthy and symptomatic participants based solely on application usage demonstrate the potential for further research by incorporating the order of applications within each session, time of day, physical activity status, and other contextual factors.
It is worth noting that this is not Apple's first attempt to utilize artificial intelligence for predicting user health conditions. Apple engineers conducted a study involving over 2,500 individuals to train an algorithm for detecting falls, which was then applied to Apple Watch Series 4 and Series 5.
Alphabet Q1 paid subscribers reached 350M, fueled by YouTube & Google One. YouTube ad revenue slightly missed at $9.88B amid Premium growth. Total Q1 revenue hit $109.9B, beating forecasts.
Anthropic releases Claude connectors for Adobe, Blender, and Ableton, enabling AI workflow automation and app control. It also sponsors the Blender Development Foundation.
Apple introduces an App Store model splitting 12-month plans into monthly installments. On iOS 26.4+, it lowers upfront costs without reducing total price, adding billing tracking.
Google launches Aadhaar digital IDs in Wallet for India, enabling secure QR/NFC verification. A phased partner rollout targets key sectors, while passport-based credentials expand globally.
YouTube tests "Ask YouTube," an AI tool blending text and video for complex queries. Initially for US Premium users, it supports interactive follow-ups, expanding Google's AI search.
Google grants US DoD classified AI access, joining OpenAI and xAI in defense deals amid Anthropic's dispute. Contractual safeguards against surveillance and lethal weapons remain unenforceable.
Mistral AI launches Workflows, a Temporal-powered orchestration platform for secure, Python-driven enterprise AI automation featuring human-in-the-loop approvals.