"Cognitive delegation": using AI for schoolwork produces correct answers but may leave students learning less

By Diario Libre · November 10, 2026.

Javier González, director of the Summa laboratory (Research and Innovation in Education for Latin America and the Caribbean), warned at the 6th Foro Global @ Casa de Campo, in La Romana, that using AI to complete an assignment can leave students with the right answer but without having learned the content. He calls it "cognitive delegation": the student lets the tool do the work and then feels they have understood it.

According to González, those who use AI retain less information and forget it sooner. To back this up, he cited an MIT Media Lab study with 54 participants who wrote essays in three ways: without tools, with a search engine or with ChatGPT. The researchers measured brain activity using electroencephalograms (recordings of the brain's electrical activity). According to the talk, the no-tools group showed 35% more neural connectivity than the search-engine group and about 50% more than the ChatGPT group. Over four months, the ChatGPT group underperformed on neural, linguistic and behavioural measures.

That figure should be read with caution. The study itself focused on writing tasks and its authors described the findings as preliminary, according to the article. Moreover, "less neural connectivity" does not by itself equal "less learning": it is an indirect indicator, and the piece does not detail how retention was measured.

The local context gives weight to the warning. According to PISA 2025 (the OECD's international assessment of 15-year-old students), half of Dominican students of that age use AI chatbots every week to learn. 40% use them to summarise assigned texts and 36% to write assignments. Only 9% say they never or almost never use them for the school tasks examined.

González framed AI as the third technological wave in schools. The first, computers and connectivity, expanded access without automatically improving learning. The second, mobile phones and social media, affected attention. The third is AI. He also noted that in PISA 2022 only 8% of Dominican students reached the basic level in mathematics, 25% in reading and 23% in science.

The expert is not calling for the technology to be banned. He acknowledges that AI can help with school management, data analysis, adaptive assessments and teacher training. His approach is to regulate its use and prevent it from replacing the effort of learning. He also warned that if teachers delegate their role to the algorithm, their confidence in their own abilities may weaken.

**Our take.** The underlying thesis is sound even if the evidence is still fragile: learning requires an effort that cannot be outsourced, and a tool that eliminates that effort also eliminates part of the outcome. It is the same logic by which nobody gets fit watching someone else lift weights. What is worrying is the speed: with half of Dominican 15-year-olds using it every week, the experiment is already under way in classrooms, without waiting for research to validate it.

The short term may be uncomfortable: generations that do well on homework and poorly on exams, and education systems slow to adapt. But design matters more than prohibition. In general, a tutor that asks questions, corrects and forces students to reason is very different from a tool that hands over a finished essay, and AI itself can be the former. That version, a personalised tutor available to any child, does fit the horizon of abundance we advocate: knowledge and individual attention for everyone.

The useful question for schools and families is not "AI, yes or no?", but "what part of the cognitive effort do we want students to keep doing?". Assessing in class, asking students to explain their reasoning and using AI for practice rather than for submitting work are simple measures. The data that will finally settle the debate, from broader studies and not only on writing, is yet to come.