Friday, April 12, 2024

New AI Tool Predicts Human Lifespan and Personalities Based on Life Events


A team of researchers has developed a new artificial intelligence (AI) tool called life2vec that can predict various aspects of human lives, including personality traits and mortality, based on sequences of life events. The tool utilizes transformer models, similar to the ones powering large language models like ChatGPT. It was trained on a dataset of 6 million individuals from Denmark, which was made available to the researchers by the Danish government.

The accuracy of life2vec surpasses that of existing state-of-the-art models in predicting future outcomes. However, the researchers emphasize that the tool should not be used for predicting real individuals, as it is specifically trained on a specific population and dataset. Instead, they see it as a foundation for further research and development.

The team behind this project includes experts in AI ethics who aim to bring a human-centered approach to AI development. By involving social scientists in the process, they hope to ensure that the tool considers the well-being of individuals within the context of the vast dataset it was trained on.

One of the key components of life2vec is the massive dataset from Statistics Denmark, which contains detailed information about every Danish citizen. The researchers used this data to create patterns of recurring life events that were fed into the model. The model then categorizes and establishes connections between different life events, such as income, education, and health factors.

Life2vec’s predictions are based on vector representations in embedding spaces that the model learns from observing millions of life event sequences. For example, the model can predict a person’s probability of mortality and individual responses to personality questionnaires.

However, the researchers acknowledge that the model’s predictions are based on correlations and specific cultural and societal contexts. They caution against generalizing these findings to other societies and highlight the biases inherent in any dataset.

Overall, the researchers consider their predictive model as a means to understand society better rather than an end product. They hope that their work will contribute to a more open and public understanding of predictive algorithms and their appropriate use.

Read more

Local News