The AI that can write sentences or compose news articles can also accurately predict the unfolding of individual human lives. A new tool called life2vec can predict outcomes, including early death, by leveraging similarities between how sequences of events progress in human lives and sequences of words progress in language, according to a recent study in Nature.
SFI External Professor Tina Eliassi-Rad (Northeastern University) and coauthors tested the tool by tapping into the records of 6 million people gathered by the government of Denmark. That data—which includes detailed day-to-day life events related to health, education, occupation, income, address, and working hours—was used to train life2vec, which is based on transformer models like those that run large language models like ChatGPT.
Tools like life2vec shouldn’t be used to make predictions about real people, explained Eliassi-Rad, but can be used to scan for how policies, rule and regulations are affecting people.
Read the paper “Using sequences of life-events to predict human lives” in Nature (December 18, 2023). DOI: 10.1038/s43588-023-00573-5