Constructed utilizing transformer models, which empower large language models (LLMs) such as ChatGPT, the tool known as life2vec is trained on a dataset sourced from the entire population of Denmark. Life2vec exhibits the capability to forecast the future, encompassing lifespans.
The researchers leveraged this data to generate extensive sequences of recurring life events, which were then inputted into their model. They adopted the transformer model methodology commonly employed in training LLMs for language and applied it to represent human life as a sequential series of events.
The adaptable architecture of the model allows for straightforward adjustments and refinement, enabling predictions across various unexplored facets of human life. This renders life2vec a promising tool for future prognostication.
Lehmann notes that medical professionals have already reached out to him seeking assistance in developing health-oriented versions of life2vec. This includes a tool aimed at shedding light on population-level risk factors for rare diseases, among other applications.
He intends to utilize the tool to reveal latent societal biases, such as unforeseen correlations between professional advancement and factors like age or country of origin. Additionally, he aims to delve into the influence of relationships on quality of life and income, while also uncovering hidden societal biases.
In New Delhi, researchers have engineered an artificial intelligence (AI) tool capable of utilizing sequences of life events — comprising health history, educational background, employment status, and income — to forecast a wide spectrum of outcomes, ranging from an individual’s personality traits to their life expectancy.
