Researchers have created an artificial intelligence (AI) program that predicts everything from a person’s personality to their longevity by using sequences of life events, such as health history, education, employment, and income.
The life2vec tool is based on transformer models, which are used to power large language models (LLMs) such as ChatGPT. It is trained using a dataset that includes all of Denmark’s citizens.
The researchers claimed that Life2vec can forecast the future with a level of accuracy that surpasses current state-of-the-art models, including individual lifespan predictions. Though predictive in nature, the research team stated that it is better utilized as a starting point for additional research rather than as a goal unto itself.
Professor Tina Eliassi-Rad of Northeastern University in the US states, “Even though we’re using prediction to evaluate how good these models are, the tool shouldn’t be used for prediction on real people.”
Eliassi-Rad explained, “It is a prediction model based on a specific data set of a specific population.”
The team expects that by including social scientists throughout the tool’s development, it will contribute to the development of artificial intelligence (AI) by keeping people in mind despite the enormous amount of data that the tool has been trained on.
According to Sune Lehmann, the study’s author and author of the journal Nature Computational Science publication, “this model offers a much more comprehensive reflection of the world as it is lived by human beings than many other models.”
The enormous data collection that the researchers utilized to train their model is the foundation of life2vec.
Using the transformer model technique, which is used to train LLMs on language, the researchers modified it for a human life portrayed as a series of events. They then used the data to generate extended patterns of recurrent life events to feed into their model.
Professor Lehmann of the Technical University of Denmark remarked, “The whole story of a human life, in a way, can also be thought of as a giant long sentence of the many things that can happen to a person.”
The model begins to categorize and make links between life events, such as wealth, education, or health factors, using the knowledge it learns by watching millions of life event sequences to create what are known as vector representations in embedding spaces.
According to the researchers, these embedding spaces form the basis for the predictions the model ultimately makes.
One of the life events that the researchers forecasted was an individual’s likelihood of dying.
“The model’s prediction space appears as a long cylinder that moves from a low probability of death to a high probability of death when we visualize it,” Lehmann stated.
“Thereby, we can demonstrate that in the scenarios with a high probability of death, many of those individuals did in fact pass away, and in the scenarios with a low probability of death, the causes of death are events that are beyond our control, such as automobile crashes,” the researcher continued.
Leave a Reply