![[Adobe Stock]](https://sp-ao.shortpixel.ai/client/to_webp,q_glossy,ret_img,w_750,h_375/https://www.psypost.org/wp-content/uploads/2025/12/artificial-intelligence-AI-chatbot-750x375.jpg)
A new theoretical review published in Trends in Cognitive Sciences explores whether relying on artificial intelligence compromises human intelligence. The paper suggests that while outsourcing mental tasks to algorithms can erode specific learned skills, our foundational cognitive abilities are likely more resilient. The impact of these tools largely depends on how people choose to interact with them.
Humans regularly use external tools to reduce the mental effort required to complete tasks. This behavior is known as cognitive offloading. Common examples include using a calculator for arithmetic or relying on a satellite navigation system for directions.
The recent explosion of generative artificial intelligence has brought the concept of cognitive offloading back into the spotlight. Generative artificial intelligence refers to computer programs that can instantly create text, solve problems, or analyze data based on massive amounts of existing information. Because these platforms can synthesize complex ideas in seconds, many worry that human intellect will suffer.
Trent N. Cash, Megan O. Kelly, Brooke N. Macnamara, and Evan F. Risko authored the new review to untangle the nuanced ways these programs affect the human mind. They separate human cognition into two distinct categories to better understand the risks. The first category includes specific skills, which are learned behaviors improved through practice, like flying an airplane or solving algebra. The second category involves basic cognitive abilities, which are foundational mental capacities, like working memory and attention.
Losing Skills Through Lack of Practice
When people offload their thinking to an automated tool, they skip the practice required to acquire and maintain specific skills. Cash and his colleagues argue that bypassing this necessary mental friction almost certainly compromises skill development. Without regular practice, acquired skills tend to decay because the brain loses the defense against forgetting. For example, physicians might lose their diagnostic sharpness if they overly rely on an automated imaging system without reviewing the underlying data.
This concept was recently demonstrated in a high school mathematics experiment in Turkey. Students were divided into groups and asked to complete practice problems using either their own knowledge, an unrestricted chatbot, or a highly guarded chatbot designed to act as a tutor. Students who used the unrestricted artificial intelligence scored high on the practice round but performed worse on a subsequent independent exam compared to peers who used no technology at all. The unrestricted software provided direct answers, which allowed students to bypass the productive struggle needed to understand the material.
“The effects are non-trivial and move in opposite directions depending on what you measure,” Alp Sungu, a co-author of the math study, previously told PsyPost. “With AI access, students scored 48% higher on practice problems, but once that access was removed, the same students scored 17% lower on exams than those who never had AI at all.” He added that using these systems merely as answer machines hurts actual learning.
Shallower Learning and the Illusion of Competence
Similarly, college students who used a chatbot to research a presentation remembered significantly less information about the topic forty-five days later than students who used traditional study methods. Traditional studying forces the brain to retrieve and connect information, a process psychologists call desirable difficulties. This mental exertion builds stronger memory pathways and secures long-term retention. Relying on an automated summary creates an illusion of competence, where students feel they know the material better than they actually do.
“There is an abysmal difference between delivering a piece of work and understanding the process of its creation,” researcher Andrรฉ Barcaui explained to PsyPost. He noted that without the mental friction of reading and writing, people lose the ability to articulate complex ideas and question information.
The ease of getting instant summaries also tends to result in shallower knowledge acquisition. In a series of experiments involving thousands of participants, people who learned about a topic using chatbot syntheses developed a weaker understanding of the subject than those who used a standard internet search. The researchers designed simulated environments where the core facts provided to both groups were completely identical. Despite receiving the exact same information, participants who read the automated summaries exerted less effort and reported feeling a lower sense of ownership over the knowledge.
Bypassing this self-guided exploration prevents users from developing deep, original knowledge structures. When participants in these experiments were asked to write advice based on what they had learned, the chatbot users wrote shorter, less factual, and less original responses. Recipients of this advice consistently rated the text generated by the chatbot users as less helpful and less trustworthy.
Critical Thinking and Cognitive Surrender
This passive consumption of information provides evidence of a broader decline in critical thinking. A recent survey combined with in-depth interviews found that individuals who heavily rely on algorithmic tools perform worse on critical thinking assessments. The effect is particularly pronounced among younger users, who often accept computer-generated recommendations without questioning their accuracy. On the other hand, individuals with higher education levels tended to maintain their analytical skills by cross-checking information across multiple sources.
“The findings reveal a strong negative correlation between frequent AI tool usage and critical thinking abilities, mediated by cognitive offloading,” researcher Michael Gerlich previously told PsyPost. He noted that this pattern suggests reliance on automated tools reduces opportunities for deep, reflective thinking.
Psychologists refer to this uncritical acceptance of algorithmic output as cognitive surrender. Rather than using the software as an assistant, users entirely relinquish mental control and adopt the machine’s judgment as their own. To explain this, scientists have proposed a Tri-System Theory of Cognition, which adds an external, artificial reasoning system to the human brain’s natural instinct and deliberate logic networks.
In laboratory puzzles, participants who had access to a chatbot frequently submitted incorrect answers simply because the software confidently presented flawed advice. Even when researchers offered financial bonuses for correct answers, a large portion of participants continued to accept the faulty algorithmic output. This indicates a high level of misplaced trust in technology.
The psychological experiments on cognitive surrender also revealed that certain personality traits offer a degree of protection. Participants with high fluid intelligence, which is the ability to solve unfamiliar problems, showed more resistance to blindly accepting algorithmic output. Additionally, individuals who naturally enjoy engaging in deep, effortful thinking were better at recognizing and rejecting incorrect answers. However, time constraints and the engaging, conversational nature of modern software still pushed many users toward uncritical reliance.
“People are not just asking AI for information; they are often letting it structure their thoughts, explanations, and decisions,” Steven Shaw, who studies human reasoning, explained to PsyPost. He suggested that people slip into cognitive surrender without realizing it, which shifts their intellectual agency over to the machine.
The Risk of Cognitive Debt
Other experts warn that treating technology as a cognitive prosthesis could stunt higher-order executive functions over time. Executive functions are the complex mental processes that enable planning, problem-solving, and decision-making. By allowing a computer to generate complete plans from start to finish, users miss out on the mental exercises necessary to develop these advanced capabilities.
“Just as one cannot become skilled at basketball without actually playing the game, the development of complex intellectual abilities requires active participation and cannot solely rely on technological assistance,” Umberto Leรณn Domรญnguez previously told PsyPost. He stressed that cognitive effort remains an absolute requirement for success in modern life.
This lack of active participation creates what psychiatrist Sรธren Dinesen รstergaard calls a cognitive debt. He argues that outsourcing scientific reasoning to machines threatens the fundamental skills required for academic discovery. Recent brain imaging studies provide evidence for this concern, showing that people utilizing algorithmic assistance display significantly lower brain activation in networks usually engaged during mental tasks.
รstergaard highlights the developers of AlphaFold, a protein-structure prediction program that won a Nobel Prize, as a prime example of rigorous human reasoning. He questions whether those scientists would have achieved such breakthroughs if automated systems had done their thinking for them during their formative education. Scientific reasoning is not an innate talent, but rather a skill sharpened through the tedious practice of reading, thinking, and revising.
Basic Abilities Resilient Despite Risks
Despite these alarming trends, Cash and his co-authors point out that basic cognitive abilities appear stubbornly resistant to change. Foundational capacities, like working memory, generally do not shrink or expand drastically based on task-specific training. While people might lose their proficiency in long division or spelling, the underlying mental hardware supporting those tasks will likely remain intact. A dystopian future where humanity loses its fundamental capacity to think seems unlikely.
Cash and his colleagues note that many questions remain unanswered about long-term interactions with these tools. Scientists still need to uncover how prolonged use affects metacognitive skills, which involve monitoring and understanding one’s own thought processes. There is a risk that users might experience source monitoring errors, misattributing computer-generated ideas as arising from their own cognition. The researchers also question whether exposure to automated assistance during critical developmental phases, such as early childhood, poses unique hazards.
The researchers emphasize that the specific design and application of these technologies will dictate their cognitive impact. When software is programmed to act as a collaborative tutor, providing hints rather than direct answers, skill acquisition is preserved. In the high school math study, students who used the restricted software performed just as well on the final exam as those who used textbooks.
If people remain actively engaged in the cognitive loop, they can mitigate the risks of offloading while leveraging the benefits of automated assistance. Users can structure their interactions with these tools to boost learning, such as asking for feedback on an original idea instead of demanding a fully formed solution. Engaging in a thoughtful partnership with technology provides a path forward that preserves human intellect.
As Cash and his co-authors concluded: “In sum, there is clearly a risk that AI can make us โstupidโ by compromising our skills (and knowledge) if we completely offload them to AI. However, AI may be less likely to diminish the foundational cognitive capacities that underpin our ability to be smart, rather than โstupidโ, in the first place.”
The study, “Is AI making us stupid?,” was authored by Trent N. Cash, Megan O. Kelly, Brooke N. Macnamara, and Evan F. Risko.
The study, “Experimental evidence of the effects of large language models versus web search on depth of learning,” was authored by Shiri Melumad and Jin Ho Yun.
The study, “Generative AI without guardrails can harm learning: Evidence from high school mathematics,” was authored by Hamsa Bastani, Osbert Bastani, Alp Sungu, Haosen Ge, รzge Kabakcฤฑ, and Rei Mariman.
The study, “AI Tools in Society: Impacts on Cognitive Offloading and the Future of Critical Thinking,” was authored by Michael Gerlich.
The study, “Thinking – Fast, Slow, and Artificial: How AI is Reshaping Human Reasoning and the Rise of Cognitive Surrender,” was authored by Steven D Shaw and Gideon Nave.
The study, “ChatGPT as a cognitive crutch: Evidence from a randomized controlled trial on knowledge retention,” was authored by Andrรฉ Barcaui.
The study, “Potential cognitive risks of generative transformer-based AI chatbots on higher order executive functions,” was authored by Umberto Leรณn Domรญnguez.
The study, “Generative Artificial Intelligence (AI) and the Outsourcing of Scientific Reasoning: Perils of the Rising Cognitive Debt in Academia and Beyond,” was authored by Sรธren Dinesen รstergaard.