Is AI eroding our thinking?

I recently read that millennials (the students I taught!) might be the last generation to be able to think and write without using AI. So I gave chatGPT the following prompt:

generate a literature review with links to sources and an extensive executive summary of peer reviewed research regarding the possibility that using large language models diminishes the human ability to think well

The first paragraph of the executive summary:

The best peer-reviewed evidence does not support a simple claim that large language models (LLMs) inevitably make people “think worse.” What it does support is a more nuanced conclusion: LLMs can reduce cognitive effort, encourage cognitive offloading, and impair retention or independent reasoning under some conditions; but when used with strong scaffolding, they can also improve performance, engagement, and some forms of higher-order thinking.

The summary offers 4 supporting citations for this paragraph, starting with ChatGPT as a cognitive crutch: Evidence from a randomized controlled trial on knowledge retention. Part of the abstract from that citatio :

...This suggests that unrestricted ChatGPT use impaired long-term retention, likely by reducing the cognitive effort that supports durable memory. The findings align with cognitive offloading theory and the ‘desirable difficulties’ principle: while AI assistance may ease initial learning, it appears to undermine the effortful processes needed for robust learning. These results have important implications for how generative AI tools should be integrated into higher education.

The rest of the chatGPT executive summary:

The strongest evidence for possible harm comes from three converging lines. First, a 2025 randomized controlled trial found that students who used ChatGPT as a study aid had significantly worse long-term retention on a surprise test 45 days later than students who used non-AI study methods. Second, cross-sectional and survey-based studies report associations between heavier AI dependence and weaker critical-thinking scores, with cognitive offloading as a plausible mediator. Third, several reviews and meta-analyses report that ChatGPT tends to reduce mental effort, which is not automatically bad, but raises a real concern when learning depends on effortful retrieval, elaboration, and self-explanation. 

The strongest evidence against a blanket “diminishes thinking” thesis is that multiple peer-reviewed reviews and experiments show gains in learning performance, engagement, argumentation, and higher-order-thinking measures when LLM use is structured as tutoring, debate support, or guided collaboration rather than passive answer generation. In other words, the evidence increasingly points to a use-pattern effect rather than a uniform technology effect: LLMs appear most risky when they replace problem formulation, retrieval, synthesis, or judgment, and most beneficial when they support those activities without substituting for them. 

A critical distinction in the literature is between performance and competence. LLMs often improve the quality or speed of an immediate output, yet the question the user is asking is about whether they reduce the human ability to think well. The current evidence suggests that output quality can improve even while the user contributes less effort, remembers less later, or practices less independent reasoning. This is the core concern in the literature: better short-term products may coexist with weaker long-term internalization. 

The literature is still immature. Most studies are in higher education, many use short interventions, many rely on self-report or course-based outcomes, and there are very few longitudinal studies that track durable changes in reasoning ability over months or years. So the most defensible conclusion is: the possibility is real and already supported in specific tasks and contexts, but there is not yet decisive evidence that LLM use broadly or permanently diminishes human thinking ability across the population.

The entire literature review is long and full of links to source citations. The link to the entire chatGPT lit review is at the top of this post.

All of which begs the question: How do educators and parents help children and students use this powerful new cultural technology without diminishing their capacity to learn, think, and write?

It looks like Chinese education researchers are digging deeply into this question: From MOOC to MAIC: Reshaping Online Teaching and Learning through LLM-driven Agents:

Since the first instances of online education, where courses were uploaded to accessible and shared online platforms, this form of scaling the dissemination of human knowledge to reach a broader audience has sparked extensive discussion and widespread adoption. Recognizing that personalized learning still holds significant potential for improvement, new AI technologies have been continuously integrated into this learning format, resulting in a variety of educational AI applications such as educational recommendation and intelligent tutoring. The emergence o ntelligence in large language models (LLMs) has allowed for these educationa enhancements to be built upon a unified foundational model, enabling deeper integration. In this context, we propose MAIC (Massive Al-empowered Course), a new form of online education that leverages LLM-driven multi-agent systems to construct an Al-augmented classroom, balancing scalability with adaptivity.

Beyond exploring the conceptual framework and technical innovations, we conduct preliminary experiments at Tsinghua University, one of China's leading universities.

Drawing from over 100, 000 learning records of more than 500 students, we obtain a series of valuable observations and initial analyses. This project will continue to evolve, ultimately aiming to establish a comprehensive open platform that supports and unifies research, technology, and applications in exploring the possibilities of online education in the era of large model AL. We envision this platform as a collaborative hub, bringing together educators, researchers, and innovators to collectively explore the future of Al-driven online