A new AI system trained on life events can predict personal outcomes like personality traits and even time of death, sparking both excitement and ethical concern. (CREDIT: Canva)

Researchers from Denmark and the United States have developed a powerful new artificial intelligence system that can predict events in people’s lives—including estimating when someone might die. This system, called life2vec, analyzes a person’s life as a series of events—similar to how AI models like ChatGPT process language—and uses this data to make detailed predictions.

Key Takeaways

• A new AI model called life2vec can predict life events, personality traits, and even time of death with impressive accuracy.
• The model uses life-event sequences from millions of people to identify patterns in education, health, work, and income.
• Ethical concerns around privacy, bias, and the use of personal data must be addressed before applying this technology more broadly.

The project is a collaboration between the Technical University of Denmark, University of Copenhagen, IT University of Copenhagen, and Northeastern University in the United States. The project demonstrates how advanced transformer-based AI models can analyze millions of life events and learn patterns that forecast personal outcomes with striking precision.

The research team trained life2vec on a unique dataset from Denmark. It includes daily life-event records for six million people over a decade. These records contain everything from health and education to income, housing, work hours, and more. The system then encodes this information into vectors, allowing the model to «understand» the complex structure of human life in the same way it processes words in a sentence.

A schematic individual-level data representation for the life2vec model. (CREDIT: Nature Computational Science)

Life2vec and the science behind prediction

At the heart of the project is a machine learning technique that transforms real-life data into a mathematical language. Events like a person’s job changes, education level, or health history are placed into a structured system. This allows the AI to view a life the way it might read a story—one event after another, all influencing what happens next.

This method draws inspiration from transformer models that power natural language tools. Just as language models recognize how words relate to one another in a sentence, life2vec recognizes how life events relate across time.

“What’s exciting is to consider human life as a long sequence of events,” said Sune Lehmann, a professor at DTU and the study’s lead author. “Usually, transformer models are used for analyzing sentences. But here, we use them to analyze life sequences—events that have happened in a person’s life.”

Related Stories

• Are AI doctors good at holding medical conversations with patients?
• Artificial intelligence can predict the weather and human health
• New AI tool accurately predicts how cancer patients will respond to treatment

When fully trained, life2vec goes beyond predicting basic events. It can estimate personal traits and even answer questions like, “Will this person die within the next four years?” With shocking accuracy, the model’s answers often match real-world outcomes. For example, individuals with higher incomes or leadership roles tend to live longer, while those with certain mental health conditions or low skill levels face higher risks of early death.

Social patterns and personal forecasts

The model’s results reflect well-known trends in social science, offering both validation and new insights. It identifies patterns where factors like gender, income, and mental health influence life outcomes. However, it doesn’t just confirm what researchers already know—it pushes the boundaries of how precisely those outcomes can be predicted.

“Scientifically, what’s exciting for us is not so much the prediction itself,” said Lehmann, “but the aspects of data that enable the model to provide such precise answers.”

Two-dimensional projection of the concept space (using PaCMAP). (CREDIT: Nature Computational Science)

By creating a shared “embedding space”—a concept borrowed from language processing—the researchers can map events like a heart attack, job promotion, or move from a city to the countryside into a single structure. This allows for deep analysis of how various factors relate and affect future possibilities.

Their approach significantly outperforms previous AI models. It predicts outcomes like early death or personality nuances with greater accuracy than any current system. These findings suggest that even highly personal outcomes may be far more predictable than once thought.

The age of personal AI forecasting

Life2vec’s achievements mark a turning point in how AI might shape daily life. Across industries, algorithms already predict weather patterns, disease outbreaks, and consumer behavior. But now, predictions are becoming more personal.

By capturing the full complexity of someone’s life in a model, researchers are entering a new phase of AI that goes beyond recommendations or social media trends. These systems can now foresee life-changing events, offering both promise and risk.

Performance of models on the mortality prediction task quantified with the mean C-MCC with 95% confidence interval. (CREDIT: Nature Computational Science)

Yet the ability to predict the future so precisely raises major concerns. Privacy, fairness, and control over personal data are at the center of the debate. Models like life2vec depend on detailed life histories. If this information falls into the wrong hands or is used without proper oversight, it could lead to serious harm.

“We’re entering a moment where we need to think carefully,” Lehmann warned. “Technologies that predict human behavior already exist inside tech companies. They track our actions, profile us, and try to influence our choices. These systems should be part of the democratic conversation.”

Where the science leads next

The next step for life2vec is expansion. The team hopes to include new types of data—like written text, images, or social network patterns. This broader view would allow AI to make even more accurate and detailed predictions. By adding richer context, researchers hope to link personal behavior, social ties, and health trends in ways never seen before.

This new approach may also reshape how governments and scientists address major challenges. Life2vec could help public health agencies understand who is most at risk for disease, uncover new links between stress and mortality, or target early interventions to people before life events spiral into crises.

Representation of life-sequences conditioned on mortality predictions. (CREDIT: Nature Computational Science)

But experts warn that caution is necessary. If deployed without transparency, these tools could reinforce biases, mislabel individuals, or create new forms of surveillance.

Even so, the research opens a new frontier in understanding what shapes people’s lives. If used ethically, these models could become powerful tools for supporting well-being and planning for the future.

The fusion of language models with life data may be the start of a new kind of science—one that turns the messy, unpredictable story of a life into a map of possibilities.

Research findings are available online in the journal Nature Computational Science.