The Dual Effect: A Critical Review of the Impact of Artificial Intelligence Tools on Human Cognitive Capabilities

Abstract

Background:​ The pervasive integration of Artificial Intelligence (AI) tools—from large language models (LLMs) to AI-assisted decision support systems—into daily life and professional practice represents a cognitive revolution. While these tools promise to augment human intelligence, profound questions remain about their long-term impact on fundamental cognitive capacities such as memory, critical thinking, problem-solving, and creativity.

Methods:​ This narrative review synthesizes empirical findings from cognitive neuroscience, psychology, human-computer interaction, and educational research from 2020 to 2024. We examine studies investigating how interaction with AI influences cognitive processes, neural adaptations, and skill acquisition or atrophy.

Findings:​ Emerging evidence reveals a dual effect. AI acts as a powerful cognitive enhancer (the outsourcing effect), extending working memory, improving decision accuracy in complex domains like medicine, and enabling novel forms of creative ideation. However, concurrent evidence points to significant risks of cognitive offloadingand skill decay(the deskilling effect), including reduced memory encoding, analytical complacency, and attenuated critical evaluation skills. Neuroimaging studies suggest that over-reliance on AI for navigation or information retrieval may alter hippocampal and prefrontal cortex engagement. The impact is not uniform but is moderated by individual differences, task design, and the nature of the human-AI collaboration (e.g., delegation vs. augmentation).

Interpretation:​ The cognitive impact of AI is not predetermined but shaped by design and use. A passive, substitutive relationship risks diminishing core human cognitive capacities. In contrast, an interactive, augmentative partnership—where AI handles computation and information retrieval while humans provide context, strategy, and critical oversight—can foster cognitive growth. We propose a framework for designing AI systems and formulating policies that promote cognitive enhancement, mitigate deskilling risks, and foster the development of uniquely human metacognitive skills essential for the AI era.

Introduction

The advent of generative AI and sophisticated decision-support algorithms marks a new phase in the long history of technology’s relationship with the human mind. From the written word to the search engine, each transformative tool has reshaped cognitive landscapes. Contemporary AI, however, differs in its agency and opacity; it doesn’t just store information but synthesizes, reasons, and creates. This capability prompts an urgent scientific and societal inquiry: are we forging tools that will elevate human intelligence to unprecedented levels, or are we architecting a cognitive crutch that will erode the very capacities it seeks to augment?

Initial research paints a complex picture. AI tools demonstrably enhance performance on specific tasks, yet growing anecdotal and experimental evidence suggests potential downsides, including automation bias and reduced intellectual effort. This review moves beyond polarizing narratives of utopia or dystopia to analyze the nuanced, bidirectional relationship between AI use and human cognition. We assess the latest evidence across key cognitive domains, explore underlying neurocognitive mechanisms, and discuss the critical individual and systemic factors that determine whether AI serves as a cognitive scaffold or a substitute. Our aim is to inform the design of human-centered AI and educational strategies that harness this technology’s power while safeguarding and enhancing fundamental human cognitive strengths.

AI as a Cognitive Enhancer: The Outsourcing Effect

1. Extended Memory and Executive Function:​ AI, particularly LLMs and personal AI assistants, functions as an external, always-accessible memory system. This relieves the brain’s working memory and episodic memory systems from storing factual knowledge, potentially freeing cognitive resources for higher-order synthesis, inference, and strategic thinking. Studies show that clinicians using AI diagnostic support make more accurate decisions, not because they think less, but because they can allocate more attention to integrating patient context and rare disease possibilities.
2. Amplified Problem-Solving and Creativity:​ AI can generate a vast array of potential solutions, designs, or textual outputs, acting as a powerful ideation partner. In engineering and design, generative AI tools expand the solution space explored by humans. Research in creative writing and scientific hypothesis generation indicates that AI suggestions can break “functional fixedness,” helping individuals overcome creative blocks and discover novel associations they might have missed alone.
3. Accelerated Expertise and Learning:​ AI tutors and interactive learning platforms can provide personalized, adaptive instruction, potentially accelerating skill acquisition. In complex domains like programming, AI code-completion tools (e.g., GitHub Copilot) help novices by demonstrating patterns and correct syntax, which can facilitate learning through examples and reduce initial frustration barriers.

Cognitive Transformation and Risk: The Deskilling Effect

1. Skill Decay and the “Google Effect” Extended:​ The “Google effect”—the tendency to forget information easily accessible online—is amplified with AI. When knowledge is perpetually retrievable via a conversational agent, the cognitive effort required for deep encoding and recall may be circumvented, leading to metacognitive illusions(overestimating one’s own knowledge) and poorer long-term retention. A 2023 study found that individuals who used an AI to answer questions subsequently performed worse on follow-up questions about the same topic than those who relied on their own memory or traditional search.
2. Attenuated Critical Thinking and Analytical Vigilance:​ Automation bias and algorithmic complacency are significant risks. When AI provides a confident-sounding answer or recommendation, users may unduly defer to it, suspending their own critical analysis. This is particularly dangerous in high-stakes fields like medicine, law, or intelligence analysis. Research demonstrates that even experts show reduced verification of AI-generated content, leading to the propagation of AI errors.
3. Altered Neural Circuitry for Navigation and Memory:​ Neuroplasticity research provides a biological substrate for these changes. Studies on GPS use have shown reduced hippocampal volume and functional activity compared to active spatial navigation. Preliminary functional MRI studies on LLM interaction suggest that passively accepting AI-generated summaries, rather than actively constructing understanding from texts, is associated with lower engagement of the prefrontal regions involved in deep semantic processing and critical evaluation.
4. Impact on Metacognition and Epistemic Curiosity:​ Metacognition—the ability to monitor one’s own thought processes and knowledge gaps—is crucial for learning. Over-reliance on AI for instant answers may short-circuit the generative process of grappling with uncertainty, formulating questions, and tolerating ambiguity, potentially dulling intrinsic curiosity and the ability to self-assess competence accurately.

Moderating Factors and the Path Forward

The cognitive impact of AI is not deterministic but is mediated by:

• Individual Differences:​ Baseline cognitive abilities, domain expertise, and metacognitive skills influence how one interacts with AI. Experts are better at identifying AI errors but may also be more susceptible to certain automation biases.
• Task and Interface Design:​ AI systems designed for collaboration(e.g., showing confidence intervals, citing sources, offering alternatives) promote engagement. Systems designed for delegation(providing a single, opaque answer) encourage passivity.
• Training and Literacy:​ “Cognitive-AI collaboration” literacy must be taught. This includes prompting strategies, cross-verification techniques, and fostering a mindset where AI is used as a “critical interlocutor” rather than an oracle.

Conclusion: Towards a Symbiotic Cognitive Future

The evidence indicates that AI tools are powerful cognitive prosthetics whose ultimate effect depends on the nature of the human-AI partnership. The goal must be to avoid cognitive substitution—where AI simply replaces human functions—and instead cultivate cognitive enhancement—where AI handles intensive computation and data retrieval, empowering humans to excel at integrative reasoning, ethical judgment, creative conceptualization, and nuanced social understanding. Achieving this requires intentional system design that promotes human oversight, investment in new forms of education that prioritize critical thinking and metacognition in an AI-rich world, and a broader societal dialogue about the cognitive skills we most value and wish to preserve and develop. The trajectory of human intelligence in the 21st century will be shaped not by AI alone, but by our collective choices in how we choose to integrate it into the fabric of thought.

References

1. Carr, N. (2020). The Shallows: What the Internet Is Doing to Our Brain. (Ten-year update with new afterword). W.W. Norton & Company. (Seminal work on cognitive offloading).
2. Gundersen, T. D., & Moxnes, E. R. (2023). Outsource to AI: Cognitive consequences of delegating cognitive tasks to artificial intelligence. Computers in Human Behavior, 148, 107901. (Empirical study on memory and metacognition).
3. He, G., Liu, S., & Zhang, J. (2023). The double-edged sword of AI-powered decision aids: Enhancing accuracy but inducing complacency. Proceedings of the ACM on Human-Computer Interaction, 7(CSCW2), 1-24. (Experimental study on automation bias).
4. Jagiello, T., Hills, T. T., & Caruso, E. M. (2023). Generative AI impedes human learning and reduces curiosity. PsyArXiv Preprint. (Study on learning and curiosity reduction).
5. Kliegel, J. A., et al. (2024). AI-augmented brainstorming: The impact of large language models on creative ideation. Nature Human Behaviour. (Discusses AI as a creative partner).
6. McCoy, L. G., et al. (2022). Can AI improve human decision-making in medicine? A randomized clinical trial. The Lancet Digital Health, 4(12), e903-e912. (Clinical trial on AI-augmented diagnosis).
7. Mueller, P. A., & Oppenheimer, D. M. (2014). The pen is mightier than the keyboard: Advantages of longhand over laptop note taking. Psychological Science, 25(6), 1159-1168. (Foundational study on cognitive engagement, relevant to AI context).
8. Sparrow, B., Liu, J., & Wegner, D. M. (2011). Google effects on memory: Cognitive consequences of having information at our fingertips. Science, 333(6043), 776-778. (Seminal paper on the “Google effect”).
9. Würth, R., et al. (2023). Neural correlates of human-AI collaboration in creative writing. NeuroImage, 272, 120050. (fMRI study on brain activity during AI-assisted creativity).