In a pioneering research collaboration involving DTU, the University of Copenhagen, ITU, and Northeastern University in the US, scientists have unveiled a groundbreaking artificial intelligence (AI) model named life2vec. This innovative model utilizes extensive personal data to predict various life events, even estimating the time of death, marking a significant leap in the realm of predictive analytics.
The research project, detailed in the recent Nature Computational Science article titled ‘Using Sequences of Life-events to Predict Human Lives,’ introduces life2vec, a model based on transformer architecture similar to OpenAI’s ChatGPT. Trained on vast health and labor market data from 6 million individuals, life2vec displayed superior predictive capabilities after an initial learning phase, outperforming other advanced neural networks. Beyond forecasting outcomes like personality traits, the model accurately estimated the time of death.
Professor Sune Lehmann, the first author of the article and a researcher at DTU, explained the model’s focus on understanding life as a sequence of events. This is usually the type of task for which transformer models in AI are used, but in our experiments, we use them to analyze what we call life sequences, i.e., events that have happened in human life, stated Professor Lehmann.
Life2vec organizes data into a complex system of vectors, incorporating information related to birth, education, salary, housing, and health. The predictions generated include intriguing insights, such as the likelihood of death within a specified timeframe. Notably, the model’s outcomes align with existing social science findings, revealing correlations between factors like leadership roles, higher income, and increased chances of survival.
Despite its potential, the life2vec model raises ethical concerns, including data privacy, sensitivity, and potential biases. Professor Lehmann stressed the importance of addressing these challenges before widespread application, particularly in assessing risks or preventing life events. As a next step, researchers plan to integrate additional information types, such as text, images, and social connections data, to further enhance the model’s predictive capabilities.
This groundbreaking research opens avenues for collaboration between social and health sciences, potentially revolutionizing our understanding of human life events. The study, based on labor market data and health records from 2008 to 2020, offers a glimpse into the future of AI applications in predicting and comprehending the complexities of individual lives. As the model evolves, researchers advocate for a careful consideration of ethical implications in the broader discourse surrounding AI predictions and their real-world applications.
