A recent study presented at the 2026 CHI conference in Barcelona has uncovered a vital nuance in our relationship with artificial intelligence: the impact of AI on your intellect depends largely on when you turn it on.
Researchers have found that while AI can act as a powerful productivity booster under pressure, using it too early in a problem-solving process may actually undermine your ability to think deeply and reason independently.
The Study: A Test of Logic and Speed
To understand the relationship between AI and cognition, computer scientist Mina Lee (University of Chicago) and her team conducted an experiment involving 393 participants. The study was designed to test how different patterns of AI usage affected decision-making and information retention.
Participants were tasked with a complex simulation: acting as a city council member deciding whether to approve a company’s proposal to fix water contamination. To make an informed choice, they had to analyze seven different documents and write a persuasive essay.
The researchers divided the participants into two primary scenarios:
1. Sufficient Time: 30 minutes to complete the task.
2. Insufficient Time: 10 minutes to complete the task.
Within these groups, participants were further categorized by their AI usage: early access, continuous access, late access, or no access at all.
The Results: The “Late Access” Advantage
The findings revealed a striking distinction between those who used AI as a starting point versus those who used it as a finishing tool.
1. Deep Reasoning and Argumentation
When given enough time, the most successful participants were those who waited until late in the process to consult the GPT-4o chatbot. By working through the documents manually first, they built a foundation of knowledge that allowed them to use the AI to refine their arguments rather than replace them. These participants produced essays with more valid arguments and a better grasp of diverse perspectives.
2. Information Retention
Interestingly, the group that never used AI at all performed best in terms of raw memory. They retained the most details from the provided documents, suggesting that the act of manual reading and synthesis is superior for long-term information storage compared to delegating that task to an algorithm.
3. The Speed vs. Depth Trade-off
The study also highlighted a pragmatic reality: under tight deadlines, AI is a lifesaver. In the “insufficient time” groups, those who used AI early scored the highest on their essays. However, this came at a cost.
“When you are under time pressure and use AI to boost your performance, you are basically risking taking the AI’s framing,” says Mina Lee. “That reduces the kinds of arguments you make and your engagement with the information.”
Why This Matters: Slow vs. Fast Thinking
The results mirror a classic psychological concept known as the two modes of learning:
* Slow, effortful reasoning: Building a deep, deliberate understanding of a subject.
* Fast, automatic thinking: Relying on quick judgments and existing habits.
By engaging with a problem before turning to AI, humans trigger “slow learning.” This creates a mental framework that allows the person to remain the “pilot” of the conversation. When AI is used too early, users often fall into “fast thinking,” where they passively accept the AI’s logic and structure, effectively outsourcing their critical thinking to the machine.
Conclusion
The study suggests that the goal is not to avoid AI, but to develop AI literacy. To maintain high-level cognitive skills, users must learn to recognize when a task requires deep, independent reasoning and when it is appropriate to use AI as a tool for speed.
The takeaway: Use AI to polish your thoughts, not to create them from scratch.
