Unveiling the potential of AI in healthcare, a recent study delves into the ability of artificial intelligence to predict health risks based on individual life details. Read on for a comprehensive look into this intriguing development.
A research team dedicated their efforts in training an AI model on data from six million Danish individuals. Named ‘life2vec', this AI model uses techniques akin to those used in Large Language Models (LLMs). By creating embedding spaces, ‘life2vec' manages to establish connections between various factors that influence mortality rates. Impressively, it has shown promising results in outperforming existing methods when it comes to predicting the timing and causes of death.
The research has been published in Nature Computational Science.
Furthermore, ‘life2vec' is not just limited to predicting mortality. The AI model has also shown capability in predicting elements of personality, such as the level of extroversion in individuals. One could say the applications of this AI model extend beyond medical predictions and into the realm of personality analysis.
Notwithstanding its promising capabilities, the study does come with certain limitations. For one, its applicability remains exclusive to Danish people. There is also the presence of sociodemographic bias in the existing dataset used to train the AI. Before ‘life2vec' can be applied in real world scenarios, researchers have made it clear that issues regarding privacy and personal data need to be addressed.
Further emphasizing the caution necessary in the application of ‘life2vec', researchers have advised against using this model on real people for prediction purposes just yet. They reiterate that while capable of predicting, ‘life2vec' remains based on a specific dataset from a specific population, hence limiting its real-world applicability at this time.