Future Forecasting with AI: Unveiling Life's Secrets through Life2vec

Dive into the groundbreaking world of Life2vec, an AI model that decodes your life story and predicts everything from personality traits to the inevitable question: 'death within four years'? Unveil the ethical dilemmas and future possibilities in this riveting exploration of AI's crystal ball on human lives.

Scientists from DTU, the University of Copenhagen, ITU, and Northeastern University in the US have conducted a research project exploring the use of artificial intelligence (AI) to predict events in people's lives. By employing large datasets related to individuals and training transformer models (similar to ChatGPT) to process language, the researchers found that these models can systematically organize data to forecast various aspects of a person's life, including predicting personality traits and estimating the time of death.

Once the model is trained in its initial phase, where it learns patterns from the data, it has proven to outperform other advanced neural networks, providing highly accurate predictions. Sune Lehmann, a professor at DTU and the lead author of the study, emphasizes that the excitement lies not only in the predictions but in understanding the underlying data aspects enabling the model to provide precise answers.

The Life2vec model, developed in the project, encodes data into a mathematical structure of vectors, organizing information about an individual's time of birth, schooling, education, salary, housing, and health. Analyzing the model's responses, the researchers found correlations with existing social science findings, such as individuals in leadership positions or with higher incomes being more likely to survive, while being male, skilled, or having a mental diagnosis is associated with a higher risk of mortality.

However, the researchers acknowledge ethical concerns surrounding the Life2vec model, including issues of data protection, privacy, and bias in the data. They stress the importance of addressing these challenges before the model can be used to assess an individual's risk of various life events.

Lehmann emphasizes the need for a democratic conversation about the implications of such technology, particularly in light of similar technologies already used by tech companies to predict and influence human behavior on social networks. The researchers suggest that incorporating additional types of information, such as text and images or details about social connections, could enhance the model's capabilities.

The research project, titled 'Using Sequences of Life-events to Predict Human Lives,' utilizes labor market data, the National Patient Registry (LPR), and Statistics Denmark. The dataset covers all 6 million Danes and includes information on income, salary, job type, and health records from 2008 to 2020.

In technical terms, the transformer model, a type of AI architecture, is designed for language learning and other tasks. It is faster and more efficient than previous models, often used for training large language models on extensive datasets. Neural networks, inspired by the human brain, are fundamental to these models, consisting of artificial neurons that learn and improve accuracy over time through training data. The researchers believe that integrating this data-driven approach could open new possibilities for interaction between social and health sciences.