Can something as simple as a college application predict a student’s GPA four years later? Using someone’s past to predict their future was the thrilling plot of Marvel’s “Captain America: The Winter Soldier,” with the evil group HYDRA utilizing artificial intelligence to figure out who could be a threat to the bad guys in the future — and eliminating them. Now, in the real world, scientists say AI really can examine how someone writes an essay to predict what their college career will look like years later.
Researchers Jonah Berger and Olivier Toubia explain that the way you express your thoughts and ideas could foreshadow your eventual grades in school. Published in PNAS Nexus, the researchers found that students who made connections between very different concepts or ideas in their admissions essays tended to get higher grades in college later on. It wasn’t about using big fancy words but about how they structured their thoughts and blended different ideas together.
Imagine each idea or concept is like a city on a map. Some students’ essays stayed within a small neighborhood, only visiting nearby “cities” or related concepts. However, the students who went on to be high achievers took a journey all across the map, visiting lots of far-flung cities and blending those distantly related ideas together in their essays.
Berger and Toubia called this spanning a large “semantic volume” – just like exploring a wide geographic area. Those students with essays covering more intellectual terrain ended up with better college GPAs.
The AI study found it wasn’t just about conceptual breadth. The researchers also looked at “semantic speed” – how smoothly the writers moved between different idea neighborhoods as they wrote their essays. The top students didn’t jump randomly between completely unrelated areas. Instead, they meandered fittingly, taking a logical path and moving coherently between adjacent idea clusters before venturing elsewhere.
It was like a road trip planning an efficient route between cities rather than teleporting erratically all over the map. This unified, structured exploration of different concepts is said to demonstrate creative thinking skills alongside strong analytical reasoning abilities.
So, did scientists just build their own Project Insight?
Obviously, Berger and Toubia haven’t built an AI crystal ball that hunts down the superheroes of the future, but the new program does look at a lot of the same information!
“The 21st century is a digital book,” Marvel character Jasper Sitwell famously said in the 2014 blockbuster. “Your bank records, medical histories, voting patterns, emails, phone calls, your damn SAT scores! Zola’s algorithm evaluates people’s past to predict their future.”
Just like Project Insight, this AI program also factors in a person’s SAT score, their parents’ education, their gender, ethnicity, college major, essay topics, and essay length to predict where that student will go in life. However, the study found that these factors didn’t play as much of a role in the final outcome as you might think.
AI was able to automatically analyze the “semantic geography” of the admissions essays using cutting-edge natural language processing. Just by modeling the concepts covered and paths taken through idea space, the AI could predict which students would get higher grades years into the future – even better than just looking at things like test scores or family income levels.
So, rather than multiple-choice tests, future AI programs could use these conceptual signatures in our writing as a neutral way to identify raw intellectual talents. It’s like giving the AI a digital map of our thoughts and letting it evaluate the scope, complexity, and logical connectivity of our idea explorations.
Source: https://studyfinds.org/project-insight-ai-predict-future/