Artificial intelligence can analyze registry data on people’s residence, education, income, health and working conditions and, with high accuracy, predict life events.
Researchers have used a smart computer system, kind of like ChatGPT, to guess what might happen in people’s lives. They collected lots of information about different people and trained the system to understand it. Surprisingly, this AI could predict events in someone’s life, even estimating when important things might happen, like the time of death. It’s a pretty cool way that technology is learning to understand and anticipate life events!
In a recent article, ‘Using Sequences of Life-events to Predict Human Lives,’ published in Nature Computational Science, researchers examined health and job data for 6 million Danes using a model called life2vec. This insightful analysis delves into how certain life events can help predict various aspects of human lives.
Once the model undergoes its initial training phase, where it learns patterns from the data, it has proven to surpass other sophisticated neural networks in predicting outcomes such as personality traits and the timing of life events with remarkable accuracy.
“We employed the model to tackle a fundamental question: How much can we forecast future events based on your past conditions and experiences? What excites us scientifically is not just the predictions but understanding the specific data aspects that empower the model to offer such accurate insights,” explains Sune Lehmann, professor at DTU and the primary author of the article.
Estimates on Time of Death Predictions
Life2vec makes predictions for general questions like ‘death within four years.’ When researchers scrutinize the model’s responses, they align with established social science findings. For instance, individuals in leadership roles or with higher incomes are more likely to survive, while factors like being male, skilled, or having a mental diagnosis are linked to a higher risk of death.
In Life2vec, data is encoded into a system of vectors, a mathematical structure organizing various information. The model determines the placement of data related to birth, schooling, education, salary, housing, and health in this structured system.
Sune Lehmann shares, “The exciting part is treating human life like a continuous sequence of events, much like a sentence in a language made up of a series of words. Typically, transformer models in AI are applied to such tasks, but in our experiments, we use them to scrutinize what we term life sequences—meaning, the events that occur in a person’s life.”
Sparkling Ethical Concerns
The researchers highlight ethical concerns related to the life2vec model, including safeguarding sensitive data, privacy issues, and addressing potential biases in the data.
These challenges need comprehensive understanding before the model can be responsibly utilized, such as in assessing an individual’s risk of developing a disease or other preventable life events.
“The model introduces both significant positive and negative aspects that warrant political discussion and attention. Comparable technologies for forecasting life events and human behavior are already in use within tech companies, tracking our actions on social networks, creating highly precise profiles, and using them to predict and influence our behavior. This conversation must be integrated into democratic discourse, allowing us to deliberate on the direction technology is heading and whether this is a path we endorse,” emphasizes Sune Lehmann.
The researchers propose the next step to involve incorporating additional forms of information, such as text, images, or details about our social connections. This approach creates a novel intersection between social and health sciences, offering a broader understanding of the complex interplay between various data sources in predicting life events.
The research project
The research project, titled ‘Using Sequences of Life-events to Predict Human Lives,’ relies on data from the labor market, along with information sourced from the National Patient Registry (LPR) and Statistics Denmark.
The dataset encompasses all 6 million Danes and encompasses details such as income, salary, stipend, job type, industry, and social benefits. In the health dataset, information includes records of visits to healthcare professionals or hospitals, diagnosis, patient type, and urgency level.
While the dataset spans from 2008 to 2020, specific analyses conducted by researchers concentrate on the 2008-2016 period and a subset of individuals restricted by age.
Transformer model
A transformer model is like a smart computer system that learns language and does various tasks. It can be taught to understand and create language, making it useful for many language-related jobs. This model is designed to be quicker and more efficient than older ones, often used to teach big language models on huge amounts of data.
Neural networks
A neural network is a computer model inspired by the brain and nervous system of humans and animals. Various types of neural networks exist, such as transformer models. Similar to the brain, a neural network is composed of artificial neurons connected to send signals to each other.
Each neuron receives input from others, calculates an output, and then passes it on to more neurons. By training on large amounts of data, a neural network can learn to solve tasks. It relies on this training data to improve its accuracy over time.
Once these learning algorithms are finely tuned for accuracy, neural networks become potent tools in computer science and artificial intelligence. They enable us to classify and group data rapidly. One of the most well-known neural networks is Google’s search algorithm.
