0
0
Researchers at Northeastern University have pioneered an innovative AI tool called Life2vec, capable of forecasting human life events, personalities, and even mortality using sequences of life occurrences such as health history, education, employment, and income.
This groundbreaking tool, powered by transformer models akin to those behind large language models like ChatGPT, was trained on data encompassing the entire population of Denmark—approximately 6 million individuals. The dataset, made accessible by the Danish government, enabled the development of a predictive model with unparalleled accuracy, surpassing current state-of-the-art models.
However, while Life2vec exhibits exceptional predictive capabilities, the team emphasizes its use as a foundation for further research rather than a definitive tool for real-life predictions. Professor Tina Eliassi-Rad stresses that the model, based on specific data from a particular population, should not be employed for predicting individuals.
The AI tool delves into societal observations, offering insights into societal norms and policies. Its creation involved collaborations with social scientists, aiming to infuse a human-centered approach into AI development, acknowledging the importance of human factors amidst the vast dataset.
Sune Lehmann, co-author of the study published in Nature Computational Science and a professor at Technical University of Denmark, highlights the tool’s ability to depict a comprehensive sequence of life events, transforming them into vector representations that categorize and link occurrences like income, education, and health factors. The model’s predictive power extends to aspects like mortality probability and personality traits.
Despite the accuracy, the researchers emphasize the limitations associated with cultural biases and societal contexts present within the dataset, cautioning against overreliance on the tool for individual predictions. They envision this work as a starting point for public discussions on the usage and implications of such predictive AI in various societal domains.
While the tool opens avenues for potential applications in healthcare, Eliassi-Rad stresses the importance of accountability and ethical considerations, urging cautious implementation. The aim is not to foresee every facet of an individual’s future but to explore societal trends and policies in a more informed manner.
Lehmann underscores the necessity for public discourse around these predictive tools, aiming to guide their implementation in democratic societies effectively. The researchers envision leveraging Life2vec’s capabilities to enhance healthcare, offering insights to professionals rather than directly impacting individuals’ lives based on predictions.
Amidst the technological advancements, Eliassi-Rad emphasizes the need to preserve human empathy and avoid reducing individuals to mathematical representations, emphasizing the tool’s role in societal exploration rather than deterministic predictions.
