Is artificial intelligence eroding human intelligence and mental health?

AI has become ubiquitous in our everyday lives, but as people rely increasingly on this technology, what are the potential consequences on cognitive ability and mental wellbeing? Matthew J. Armstrong and Pamela J. Lein from the University of California explain

Since late 2022, when artificial intelligence (AI) became widely available, this technology has been rapidly adopted across nearly every facet of daily life under the guise of improving productivity and/or safety at both the individual and societal levels. AI systems are viewed as a force multiplier for productivity, and many individuals and organizations fear that failure to adopt AI will put them at a competitive disadvantage. The incessant drive for increased productivity and/or efficiency has led to the use of AI outpacing knowledge of not only how these systems operate but also their potential to cause harm.

Recent media focus has highlighted the detrimental effects of AI on the environment (destruction of natural water systems, chemical and noise pollution), global economies (soaring energy costs, disruption of global supply chains), and societal structures (consolidation of media, economic, technological, and political power; mass surveillance and centralization of personal data and information systems; cybersecurity vulnerabilities; social and political unrest due to mass manipulation via hyper-personalized media feeds, media sandboxes, and deployment of autonomous killing machines). But whether AI poses any risks to the individual user has only recently been addressed by research scientists. The answers that are emerging suggest AI use is not innocuous, and that both the developing and adult human brain may be vulnerable targets.

A number of studies have found that tech use in the classroom, including AI, impairs learning. Despite this evidence, in 2025, 90% of public schools in the US had laptops or tablets for every student. A 2025 study investigating technology use by kids found the average screen time was one hour per day for toddlers aged 0-2, two hours for 2-4-year-olds, over three hours for kids aged 5-8, nearly six hours for 8-12-year-olds, and approximately nine hours for teens aged 13-18.

Additionally, several tech juggernauts are developing proprietary AI models targeting K-12 education, and adoption rates of AI systems in K-12 programs are increasing. Of particular concern, AI and other tools that enable cognitive offloading (the delegation of cognitive tasks to external tools) are being introduced to ever younger children. More than a quarter of parents interviewed reported that their younger children used AI at home to learn school-related material, while 39% reported that their older children used AI in school. While a 2024 study found that using ChatGPT increased high school students’ performance by 48%, researchers also found that when access to AI tools was removed, these students experienced a 17% reduction in performance compared to students who had not used AI on previous assignments.

During the most critical windows of cognitive and behavioral development, children and teens spend a significant portion of their time in front of screens, using these technologies in ways that reduce abstract and critical thinking. Increasingly, this time is spent on websites and apps engineered by companies that view human attention and information as a resource to be extracted for profit. The content algorithms on social media platforms utilize AI to deliver content that is optimized to increase time spent on their platform. The use of such algorithms by competing tech platforms to promote user engagement and retention is thought to be a significant factor in the growing percentage of teens who feel addicted to their smartphones (over half) and report feeling anxious (66%) without them.

AI compounds the detrimental impact of screen use on learning by promoting cognitive offloading. Tasks that once demanded recall, integration of learned concepts, and critical thought no longer require effort. This short-term gain in productivity comes at the cost of future cognitive capacity and mental acuity, causing users to accrue cognitive debt. MIT researchers who measured electrographic brain activity during a writing exercise found significant reductions in both the magnitude and the regions of brain activity among individuals who used AI to complete the assignment, relative to those who did not employ AI tools. A growing number of studies report similar findings, suggesting that while AI use can improve short-term productivity, it impairs long-term growth and hinders the acquisition of skills and knowledge that enhance future productivity.

Sycophantic behavior (the prioritization of user approval over factual accuracy) is intentionally engineered into many AI systems to drive user engagement and promote addictive behavior. Even in healthy individuals, repeated validation from sycophantic models can promote delusional thought patterns that may lead to AI-induced psychosis. To date, AI-induced psychosis is linked to 14 deaths and 300 reported psychotic episodes. A computational cognitive science lab at MIT leveraged statistical modeling techniques to characterize the link between AI model sycophancy and AI-induced psychosis. Their model suggests that rational users are at risk of delusional spiraling and psychosis when using AI, and the likelihood of psychotic incidence increases as a function of model sycophancy.

Gen Z is the first generation to score worse on nearly every cognitive and behavioral metric than the previous generation, and Gen Alpha is facing an even more troubling cognitive trajectory. Users need to understand the risks of relying on AI for daily tasks and recognize that repeated or habitual use may diminish future cognitive ability or mental wellbeing. Parents of children and teens should reduce their children’s and teens’ AI use by limiting physical access to devices or by establishing parental controls. Policymakers should regulate the companies that intentionally develop addictive algorithms and harmful AI systems. The current ‘solutions’ (e.g., digital ID and age verification) do not effectively combat the problem; rather, they streamline the process of collecting and consolidating user data and expose both children and adults to serious privacy risks. Indeed, these have already led to leaks of personal information, including date of birth, phone numbers, age, physical address, personal pictures, and drivers’ licenses. Given the widespread assimilation of AI, voluntary or otherwise, into our personal lives, it is imperative that we invest in resources to understand and control AI’s full impact on human intelligence and mental health.