The possibility of pre-planning someone’s entire life is both exciting and frightening. A new artificial intelligence trained on the personal data of everyone in Denmark can do just that.
Today’s deep learning-based AI systems are predictive machines. They work by ingesting vast amounts of data and using it to extract statistical patterns that can be used to make informed inferences about previously unseen data.
Although AI chatbots have amazing linguistic fluency, they operate in much the same way. They learn from vast amounts of text data and try to predict which word will come next in a text string.
Enabling the breakthrough advances in functionality we’ve seen over the past few years are new deep learning architectures known as transformers, which can now be trained on much more data than previous algorithms. . If you can train a model on almost the entire internet, you’ll find that its predictions become very sophisticated.
Now, researchers have shown that similar techniques can be used to train a model on a huge database of health, social, and economic information collected by the Danish government. The resulting AI can now make very accurate predictions about people’s lives, including their likelihood of dying within a certain time frame and their personality traits.
“This model opens up important positive and negative perspectives to politically discuss and address,” Sune Lehmann from the Technical University of Denmark, who led the study, said in a statement. . “Similar technologies for predicting life events and human behavior are already in use within technology companies today, for example by tracking our behavior on social networks and predicting our behavior with great accuracy. They can create profiles and use these profiles to predict our behavior and influence us.”
The dataset the researchers used spans from 2008 to 2020 and includes all 6 million Danes. Contains information about income, work, social security, health care provider visits, disease diagnosis, and more.
However, some work was required to convert the data into a format that the transformer could understand. They reconstructed all the information in the database into a so-called “life sequence” in which all the events related to each individual are arranged in chronological order. This allows AI chatbots to predict the next event in much the same way that they predict the next word.
By training many of these life sequences, the model can find patterns that connect different events in someone’s life and help predict their future. Researchers say that between 2008 and 2016, they trained a model based on the life course of people from ages 25 to 70 and used it to make predictions for the next four years. Ta.
When we asked them to estimate the likelihood that someone would die during this period, they outperformed the current state-of-the-art by 11%. They also obtained a model for predicting how people scored on personality tests, and the results were better than models trained specifically for that task.
While performance on these two tasks is impressive, the paper describing the study natural computational science, the team points out that what’s really interesting about this model is the fact that it can potentially be used to make all sorts of predictions about people’s lives. Until now, AI has typically been trained to answer specific questions about people’s health and social trajectories.
Clearly, this type of research raises some troubling questions about privacy and human agency. But the researchers point out that understanding what this type of technology can do is useful, since private companies are almost certainly doing similar things with their own data. .
And given that AI capabilities are rapidly advancing, it will be important to have a public discussion about what kinds of AI-powered predictions are allowed in both the private and public realms. Probably, Lehman says.
“I have no such answer,” he said in a press release. “But now is the time to start a dialogue, because what we do know is that detailed predictions about human life have already been made, and right now there is no conversation. That’s what’s happening behind closed doors.”
Image credit: Nat / Unsplash