December 1, 2025 By: JK Tech
A new study from MIT brings attention to the often-overlooked concept of cognitive effort and what it reveals about how we think. We all know when something is mentally exhausting, whether it’s a tough puzzle, a complicated math problem, or a logic question that forces us to pause and rethink. According to this research, modern reasoning-focused AI models experience something surprisingly similar.
The team, led by Evelina Fedorenko at the McGovern Institute, compared how humans and AI handle the same set of problems. For people, the researchers measured how long it took to figure out each question. For AI models, they looked at how many “tokens” the system used while working through the answer, which is basically a rough measure of how much internal computation the model spent.
What stood out is how closely the two lined up. Whenever humans took longer to solve a certain type of task, the AI also needed more tokens to sort it out. This wasn’t just for math problems. Even with visual reasoning challenges, like those puzzles where you look at colored grids and try to guess the pattern, the effort patterns matched.
The interesting part is that nobody designed these AI models to imitate human brains. Yet both end up spending more mental or computational energy on the same kinds of difficult problems. Fedorenko mentioned that the similarity was much stronger than they expected.
But the study doesn’t claim that AI thinks like humans or feels anything. The researchers are very clear on that point. AI might show its “work” through long chains of written steps, but that doesn’t mean it thinks in language. Deep inside, the calculations are happening in a completely different way.
Still, this overlap tells us something important. It suggests that certain problems simply require more reasoning effort, no matter whether you’re a human or a machine. And as newer AI systems become better at reasoning, understanding how they tackle challenges might even help us understand how our own brains work.
This research comes at a time when people are still debating whether AI can truly reason or if it is mostly pattern-matching. The MIT findings don’t settle that debate, but they do show that when the difficulty increases, both humans and AI feel the strain in their own ways.
