A new study, recently published in the American Psychological Association, suggested that while using generative artificial intelligence (GenAI) tools does not directly cause cognitive decline in users, it may affect their confidence.
For the study, users self-reported using GenAI tools (like ChatGPT, Claude, Gemini) for tasks like deciding commute options, grocery and meal planning, travel experiences, planning salary increases, doing mathematical calculations, discussing their saddest memory, etc.
These tasks were designed to mimic real-world cognitive demands such as planning under uncertainty, decision-making, and analysing experiences. They were encouraged to use commercially available large language models (LLMs) as they commonly would.
Overall, the study reported that participants who relied heavily on GenAI responses reported lower confidence in their own ability, but those who engaged with the responses and altered them to include manual inputs reported higher levels of confidence.
HOW WAS THE STUDY CONDUCTED?
The study was conducted on 1,923 adults, 946 from Canada and 977 from the United States, between the ages of 25 and 57.
Data was collected between June 2024 and December 2024, and the participants were asked to self-report after completing each task based on three categories:
“I relied on AI to complete this task.”
“I would have arrived at this answer without AI.”
“I felt confident in my own reasoning.”
Responses to each prompt were recorded on a 7-point Likert scale, where 1 represented strong disagreement and 7 represented strong agreement.
KEY FINDINGS
Participants in the study reported a strong reliance on GenAI to complete the assigned tasks. Those who relied more on GenAI tools reported a lower confidence in their own reasoning. On the other hand, participants who substantially revised, contradicted, or replaced AI-generated output with their own ideas or conclusions reported higher levels of confidence (also known as override behaviour).
According to Dr Gorav Gupta, Senior Psychiatrist at Tulasi Healthcare, Gurugram, actively using AI by questioning, refining, and engaging with its outputs can support learning and reasoning, but using it passively may reduce the need for effortful thinking, which over time can weaken deeper cognitive processes.
“Cognitive offloading becomes a concern when it turns into a default habit rather than a conscious choice,” Dr Gupta said. “If someone starts relying on AI before even attempting to think through a problem, it can resemble dependency,” he added.
But in the study, the use of GenAI was not uniform across tasks. Participants relied more on AI tools in tasks involving planning under uncertainty, multistep sequencing, and interpretation of ambiguous information. In tasks that needed introspection or linking reason to personal memories or goals, they reframed AI suggestions to include more personal responses, thus depicting higher self-confidence.
“Long-term changes can appear if people consistently depend on GenAI for planning and decision-making, including favouring speed and convenience over depth, leading to more surface-level thinking,” Dr Gupta said.
On average, participants issued 4.2 prompts to the GenAI tool per task, which showed frequent communication with the tool. The mean override frequency, which signified how often the responses were manually modified, was lower, at 0.8 per task.
Seniority in occupational roles also affected how confident participants felt in their responses. Those in senior leadership or executive roles showed higher override rates compared to participants in entry-level roles. Men also reported a higher reliance on AI on average, as compared to women in the study. However, these observations were just exploratory and not meant to test the predefined hypothesis, researcher and author Sarah Baldeo added to her paper.
Some qualitative themes were also reported by participants during the course of the study. These were: the AI structured my entire approach, the ideas didn’t feel like mine, responses were fast but not deep, awareness of over-reliance, and simultaneous appreciation of efficiency and concern about cognitive engagement.
Dr Gupta said that when people report that AI-generated ideas “didn’t feel like my own”, it points to a disconnect between effort and ownership. “Without ownership, which is tied to the process of working through ideas, making mistakes, and arriving at conclusions independently, the outcome may feel less personal, which can affect satisfaction and confidence,” he said. This can also lead to feelings of impostor syndrome, where individuals question their own contribution or ability, Dr Gupta added.
The study concluded that AI itself was not associated with reduced confidence, but instead with how participants engaged with the outputs that GenAI presented. The study does not measure actual cognitive performance or long-term outcomes, but highlights a shift in perceived autonomy, which is a psychological effect that comes with using GenAI to complete everyday tasks.
“Attention span for complex or prolonged tasks could decrease with long-term GenAI reliance, and individuals may become less aware of what they truly understand versus what they can generate with assistance,” Dr Gupta said. “At the same time, those who use AI thoughtfully as a tool to enhance rather than replace thinking may develop stronger skills in judgment and decision-making,” he added.
– Ends
