Imagining artificial intelligence that can predict when a person will die is no longer just the stuff of science fiction. Life2vec, an emerging model from research at the Technical University of Denmark, is taking a significant leap in analyzing complex data to forecast crucial human life events, such as early mortality.
Recently published in Nature Computational Science, this model is emerging as a revolutionary tool with potential applications in public health, social planning and understanding sociodemographic patterns.
The researchers fed Life2vec data collected between 2008 and 2016, challenging it to predict future circumstances, including the probability of death in the next four years. The single approach included data pairs in which one person in the studied pair would die, achieving a remarkable accuracy of 78%. Although the model could not predict deaths due to accidental causes, this advance represents a milestone in the application of artificial intelligence to understand and predict crucial aspects of human life.
Life2vec is based on advanced deep learning and data modeling techniques, using a complex neural network architecture, specifically transformer models. These models are efficient in processing data sequences and recognizing patterns in large sets of information. This approach allows Life2vec to analyze and learn from sequences of events in people’s lives, transforming these events into numerical representations.
The Life2vec development process includes rigorous data preprocessing that normalizes and categorizes key variables such as education, health, income and occupation. The structure of the data is organized in a way that makes it easy to identify relevant patterns and correlations. The model is trained in several phases, from a massive data set to fine-tuning to improve accuracy in predicting specific outcomes, such as early mortality.
The predictive capacity of Life2vec is evaluated with an impressive 78% accuracy, representing a significant improvement of 11% compared to standard models. Concept space analysis reveals how the model organizes and understands human life events, highlighting critical patterns and relationships.
It is essential to highlight that the application of artificial intelligence to predict events such as early mortality raises fundamental ethical considerations. Although life2vec promises to positively influence areas such as public health and social planning, responsible development and use of these technologies must be advocated to avoid potential discrimination or harmful decisions.
