In the fascinating realm of artificial intelligence, a groundbreaking tool has emerged—life2vec. Crafted using transformer models, the very engines that propel large language models like ChatGPT. This app delves into the sequences of life events unlocking the ability to predict everything. This blog will guide you through the journey of life2vec, a powerful tool crafted using transformer models, the same tech behind language giants like ChatGPT.
Life2vec analyzes sequences of life events to forecast everything from personality traits to the complex topic of mortality. This AI tool uses a fancy system of vectors to organize information about a person’s demographics, status and health. Think of it like sorting out the details of someone’s life, just as words make up a sentence in language. This clever use of transformer models gives researchers a new way to look at what they call “life sequences“.
Imagine gathering info from a whole country, for example Denmark has 6 million people! The Danish government shared this data with researchers, and that’s what life2vec learned from. But here’s the thing: even though it’s awesome at guessing, the researchers say it’s not a crystal ball for real people.
According to Tina Eliassi-Rad, a luminary in computer science at Northeastern University:
“Even though we’re using prediction to evaluate how good these models are, the tool shouldn’t be used for prediction on real people. It is a prediction model based on a specific dataset of a specific population.”Tina Eliassi-Rad
Tina Eliassi-Rad, states these computer tools let us peek into how our society works—like checking out rules and policies. It’s kind of like scanning what’s happening in our world. To keep it real, the researchers joined forces with social scientists. Thereby making sure the computer magic doesn’t forget about actual humans in its super-big dataset.
As reported by Sune Lehmann:
“This tool shows us more about how real people live than many others. It’s like a mirror reflecting our world”Sune Lehmann
The guesses from Life2vec are like answers to questions such as ‘might this person pass away in the next four years’?
When the researchers looked at what the model said, it matched up with things we already know from social science. For example, if everything else is the same, someone in charge or with more money is more likely to stay alive. On the flip side, being male, skilled, or having a mental health issue might mean a higher chance of not making it.
Life2vec sorts the data into a big system of vectors, which is like a math way of organizing different info. It decides where to put data about birth, school, education, salary, home, and health.
“What’s interesting is thinking about life like a big chain of events, kind of like how a sentence in a language is made up of lots of words. This is usually what computers in AI do, but we’re using them to look at life events—things that happened to people”Sune lehmann
Visualizing the prediction space, Lehmann paints a vivid picture
“It looks like a long cylinder that takes you from low probability of death to high probability of death.”Sune lehmann
The model accurately reflects instances of mortality. Where high probabilities align with actual deaths and low probabilities may lead to unforeseen causes like car accidents.
The people who did the study talk about some tough questions that come with the life2vec model. Like, how do we keep personal info safe? What about privacy? And what if the data is a bit one-sided?
Before we can use the model to figure out if someone might get sick or something bad might happen, we need to figure out these hard questions says Sune Lehmann.
“The model gives us good and not-so-good things to think about. Other tech companies are already using similar things to guess what we might do and nudge us in certain ways. We need to talk about this as a group, so we know where technology is going and if that’s where we want it to go”Sune Lehmann
Wrapping it up, Tina and Lehmann see this computer tool as a starting point, not the final answer. They believe big tech companies are likely using similar tools behind closed doors.
So, next time you hear about computers predicting things, remember it’s like a society telescope—cool. But we should also be aware of its limits and potential threats.