{"id":199,"date":"2026-11-04T07:30:00","date_gmt":"2026-11-04T06:30:00","guid":{"rendered":"https:\/\/hillwardconsults.com\/journal\/?p=199"},"modified":"2026-11-04T11:47:08","modified_gmt":"2026-11-04T10:47:08","slug":"ai-made-research-faster-did-it-make-us-learn-less","status":"publish","type":"post","link":"https:\/\/hillwardconsults.com\/journal\/ai-made-research-faster-did-it-make-us-learn-less\/","title":{"rendered":"AI made research faster. Did it make us learn less?"},"content":{"rendered":"\n<p class=\"wp-block-paragraph\"><em>Part 1 of a series on learning in the age of AI<\/em><\/p>\n\n\n<p class=\"wp-block-paragraph\">When I started doing research work, a single review of the literature could swallow days. I would search Google Scholar and other databases for studies that looked relevant, print them out, and sit with a pen. Abstract, methodology, key findings, conclusion: I marked each one, paper after paper. Then came the drafts. Draft upon draft, hours upon hours.<\/p>\n\n\n<p class=\"wp-block-paragraph\">Then ChatGPT arrived, and the work changed. It could suggest where to look. I could drop in the link to a paper and get a summary. I could paste my messy notes and get them back in clean order. Most of what was left for me was to read through, check the information and confirm the sources.<\/p>\n\n\n<p class=\"wp-block-paragraph\">Now we have AI agents (tools that carry out a series of steps for you, not just answer one question). They can find the papers, read them, summarise them and produce a well-formatted document at the end.<\/p>\n\n\n<p class=\"wp-block-paragraph\">So here is the question I keep coming back to. If work that once took weeks now takes hours, what happens to the learning that used to come with all that effort?<\/p>\n\n\n<p class=\"wp-block-paragraph\">I went looking for what researchers have actually found. The answer is more useful than \u201cAI is good\u201d or \u201cAI is bad\u201d.<\/p>\n\n\n<h2 class=\"wp-block-heading\">Not all effort is learning<\/h2>\n\n\n<p class=\"wp-block-paragraph\">Look again at that old process. Some of the effort was logistics: searching, printing, retyping, formatting references. None of that made me wiser. It just took time.<\/p>\n\n\n<p class=\"wp-block-paragraph\">But some of the effort was the learning itself. Reading a methodology section closely enough to spot its weakness. Holding two studies side by side and working out why they disagree. Struggling to put an argument into my own words. That struggle is where understanding was built.<\/p>\n\n\n<p class=\"wp-block-paragraph\">Learning scientists have a name for this second kind of effort: \u201cdesirable difficulties\u201d, the friction that makes knowledge stick. So the real question is not whether AI saves effort. It is <strong>which effort it is saving<\/strong>.<\/p>\n\n\n<h2 class=\"wp-block-heading\">What the research says<\/h2>\n\n\n<p class=\"wp-block-paragraph\"><strong>1. AI can raise the quality of the work without raising what you learn.<\/strong><\/p>\n\n\n<p class=\"wp-block-paragraph\">In a randomised study of 117 university students doing a writing task, the group that used ChatGPT improved their essay scores more than the other groups. But when the researchers measured how much the students had actually learned, and whether they could apply it, there was no real difference (Fan et al.). The authors warn of \u201cmetacognitive laziness\u201d: letting the tool do the thinking about your own thinking.<\/p>\n\n\n<p class=\"wp-block-paragraph\">A larger field experiment in a high school maths programme found something similar. Students with GPT-4 did better during practice. When the AI was taken away, those who had relied on it could do worse, a sign that they had built less skill (Bastani et al.). The output looked better. The learner was not better.<\/p>\n\n\n<p class=\"wp-block-paragraph\"><strong>2. Designed well, AI can teach faster than a good classroom.<\/strong><\/p>\n\n\n<p class=\"wp-block-paragraph\">This is the other half of the story, and it matters just as much. In a randomised trial with college physics students, a custom AI tutor built on proven teaching methods helped students learn significantly more, in less time, than a strong in-class lesson. They also felt more engaged (Kestin et al.).<\/p>\n\n\n<p class=\"wp-block-paragraph\">Closer to home, a World Bank pilot in Edo State ran a six-week after-school programme in 2024 in which secondary school students used AI, with teachers guiding them, to improve their English. The gains were about 0.3 standard deviations, which the researchers describe as nearly two years of typical learning in six weeks. It outperformed 80% of comparable education programmes studied in developing countries, and girls gained even more than boys (De Simone et al.).<\/p>\n\n\n<p class=\"wp-block-paragraph\">Notice what the successful cases have in common. The AI was set up to guide, not to hand over answers, and a teacher stayed in the loop. In the maths study, the harm largely disappeared when the AI tutor was told to give teacher-designed hints instead of solutions.<\/p>\n\n\n<p class=\"wp-block-paragraph\"><strong>3. Order matters: think first, then use AI.<\/strong><\/p>\n\n\n<p class=\"wp-block-paragraph\">An early study from MIT (not yet peer-reviewed, and small, so treat it with care) measured brain activity while people wrote essays. Those who wrote with an AI assistant showed the weakest brain engagement, felt the least ownership of their essays, and many struggled to quote what they had just written. But people who first wrote on their own and then used the AI showed stronger recall (Kosmyna et al.). One study is not proof, but it fits the pattern above: the thinking has to happen somewhere, and it is better if it starts with you.<\/p>\n\n\n<p class=\"wp-block-paragraph\"><strong>4. The skill shifts to checking and judging.<\/strong><\/p>\n\n\n<p class=\"wp-block-paragraph\">A Microsoft Research survey of 319 knowledge workers found that the more people trusted the AI, the less critical thinking they applied; the more they trusted their own ability, the more they applied. It also found that critical thinking does not vanish with AI. It moves: towards verifying information, fitting the AI\u2019s output into the real task, and taking responsibility for the result (Lee et al.).<\/p>\n\n\n<p class=\"wp-block-paragraph\">That matches my own experience exactly. \u201cRead through, verify, and I am good to go\u201d sounds small, but it is the whole job now. And it is not optional. When researchers checked 636 references produced by ChatGPT in 2023, 18% of those from GPT-4 did not exist at all, and many real ones had errors (Walters and Wilder). Tools have improved since, but the lesson stands: a fluent answer is not the same as a correct one.<\/p>\n\n\n<p class=\"wp-block-paragraph\">A 2026 review in <em>Trends in Cognitive Sciences<\/em> sums up the evidence well: handing thinking over to AI can slow the building of skills and let existing skills fade, but the risk depends on <strong>how<\/strong> AI is used (Cash et al.).<\/p>\n\n\n<h2 class=\"wp-block-heading\">\u201cI feel like I cheated\u201d<\/h2>\n\n\n<p class=\"wp-block-paragraph\">Many of us were raised to believe that anything good must be hard. So when a result comes quickly, part of us feels guilty, as if it doesn\u2019t count.<\/p>\n\n\n<p class=\"wp-block-paragraph\">Psychologists have studied this. People tend to judge work as better and more valuable when they believe more effort went into it, even when the work itself is the same (Kruger et al.). It is a mental shortcut, and it explains a lot of the guilt around AI.<\/p>\n\n\n<p class=\"wp-block-paragraph\">But you don\u2019t have to carry bricks on your head before you can make a living. Nobody thinks a builder is cheating because they use a wheelbarrow. The question is not \u201chow hard did I work?\u201d but \u201cdo I understand what I produced?\u201d<\/p>\n\n\n<p class=\"wp-block-paragraph\">Here is a simple test. Close the laptop. Can you explain the work in your own words? Can you defend it if someone challenges it? Could you apply it to a new problem? If yes, the speed was a gift. If no, the AI did the learning for you, and you only collected the result.<\/p>\n\n\n<h2 class=\"wp-block-heading\">What this means for you<\/h2>\n\n\n<p class=\"wp-block-paragraph\">Whether you are a student, a researcher or a professional learning something new at work, three habits come straight out of the evidence:<\/p>\n\n\n<ol class=\"wp-block-list\">\n\n<li><strong>Think first.<\/strong> Write your own rough answer, outline or questions before you open the AI. Then use it to challenge and improve what you have.<\/li>\n\n\n<li><strong>Ask for hints, not answers,<\/strong> when your goal is to learn. Tell the AI: \u201cDon\u2019t give me the answer. Ask me questions and give me hints until I get there.\u201d<\/li>\n\n\n<li><strong>Own the checking.<\/strong> Open the sources. Confirm the quotes and the numbers. You are responsible for every word that carries your name.<\/li>\n\n<\/ol>\n\n\n<p class=\"wp-block-paragraph\">AI is a thinking partner, not a shortcut. Used one way, it hollows out learning. Used the other way, it may be the best tutor most of us will ever have.<\/p>\n\n\n<h2 class=\"wp-block-heading\">What comes next in this series<\/h2>\n\n\n<p class=\"wp-block-paragraph\">This is the first of several posts on learning in the age of AI. Coming up: how students can use AI without skipping the learning, how parents can guide their children, and what AI agents mean for the future of research itself. I am studying this closely myself, and I will share what I learn as I go.<\/p>\n\n\n<p class=\"wp-block-paragraph\">If you want help bringing AI into your team\u2019s work in a way that builds skill instead of replacing it, visit <a href=\"https:\/\/hillwardconsults.com\">hillwardconsults.com<\/a>. And to get each part of this series in your inbox, subscribe to the Hillward weekly note below.<\/p>\n\n\n<div style=\"background:#f3e4b8;border:1px solid #ecd9a0;border-radius:20px;padding:28px 30px;margin:36px 0;color:#1b2233\"><h3 style=\"font-family:Fraunces,Georgia,serif;margin:0 0 8px;font-size:1.6rem;color:#0e1d3a\">One useful idea, every Thursday.<\/h3><p style=\"margin:0 0 14px\">The <a href=\"https:\/\/hillwardconsults.com\/journal\/\">Hillward weekly note<\/a>: a short tip, tool or lesson in plain English. Join below.<\/p><div style=\"max-width:520px\"><script>(function (s, e, n, d, er) {s['Sender'] = er;s[er] = s[er] || function () {(s[er].q = s[er].q || []).push(arguments)}, s[er].l = 1 * new Date();s[er].on = function(event, callback) {s[er].listeners = s[er].listeners || {};(s[er].listeners[event] = s[er].listeners[event] || []).push(callback);};var a = e.createElement(n), m = e.getElementsByTagName(n)[0];a.async = 1;a.src = d;m.parentNode.insertBefore(a, m)})(window, document, 'script', 'https:\/\/cdn.sender.net\/accounts_resources\/universal.js', 'sender');sender('d15081180b5747')<\/script><div style=\"text-align: left\" class=\"sender-form-field\" data-sender-form-id=\"av2V4L\"><\/div><\/div><\/div>\n\n\n<h2 class=\"wp-block-heading\">Sources<\/h2>\n\n\n<ul class=\"wp-block-list\">\n\n<li>Bastani, H., Bastani, O., Sungu, A., et al. (2025). Generative AI without guardrails can harm learning: Evidence from high school mathematics. <em>Proceedings of the National Academy of Sciences<\/em>. <a href=\"https:\/\/doi.org\/10.1073\/pnas.2422633122\" target=\"_blank\" rel=\"noopener\">https:\/\/doi.org\/10.1073\/pnas.2422633122<\/a><\/li>\n\n\n<li>Cash, Kelly, &amp; Macnamara (2026). Is AI making us stupid? <em>Trends in Cognitive Sciences<\/em>. <a href=\"https:\/\/doi.org\/10.1016\/j.tics.2026.06.004\" target=\"_blank\" rel=\"noopener\">https:\/\/doi.org\/10.1016\/j.tics.2026.06.004<\/a><\/li>\n\n\n<li>De Simone, M. E., Tiberti, F., Barron Rodriguez, M., et al. (2025). From chalkboards to chatbots: Evaluating the impact of generative AI on learning outcomes in Nigeria. World Bank. <a href=\"https:\/\/doi.org\/10.1596\/1813-9450-11125\" target=\"_blank\" rel=\"noopener\">https:\/\/doi.org\/10.1596\/1813-9450-11125<\/a> (summary: <a href=\"https:\/\/blogs.worldbank.org\/en\/education\/From-chalkboards-to-chatbots-Transforming-learning-in-Nigeria\" target=\"_blank\" rel=\"noopener\">https:\/\/blogs.worldbank.org\/en\/education\/From-chalkboards-to-chatbots-Transforming-learning-in-Nigeria<\/a>)<\/li>\n\n\n<li>Fan, Y., Tang, L., Le, H., et al. (2025). Beware of metacognitive laziness: Effects of generative artificial intelligence on learning motivation, processes, and performance. <em>British Journal of Educational Technology<\/em>. <a href=\"https:\/\/doi.org\/10.1111\/bjet.13544\" target=\"_blank\" rel=\"noopener\">https:\/\/doi.org\/10.1111\/bjet.13544<\/a><\/li>\n\n\n<li>Kestin, G., Miller, K., Klales, A., et al. (2025). AI tutoring outperforms in-class active learning. <em>Scientific Reports<\/em>. <a href=\"https:\/\/doi.org\/10.1038\/s41598-025-97652-6\" target=\"_blank\" rel=\"noopener\">https:\/\/doi.org\/10.1038\/s41598-025-97652-6<\/a><\/li>\n\n\n<li>Kosmyna, N., et al. (2025). Your brain on ChatGPT: Accumulation of cognitive debt when using an AI assistant for essay writing task. arXiv preprint. <a href=\"https:\/\/arxiv.org\/abs\/2506.08872\" target=\"_blank\" rel=\"noopener\">https:\/\/arxiv.org\/abs\/2506.08872<\/a><\/li>\n\n\n<li>Kruger, J., Wirtz, D., Van Boven, L., &amp; Altermatt, T. W. (2004). The effort heuristic. <em>Journal of Experimental Social Psychology, 40<\/em>, 91\u201398. <a href=\"https:\/\/doi.org\/10.1016\/S0022-1031(03\" target=\"_blank\" rel=\"noopener\">https:\/\/doi.org\/10.1016\/S0022-1031(03<\/a>)00065-9<\/li>\n\n\n<li>Lee, H.-P., Sarkar, A., Tankelevitch, L., et al. (2025). The impact of generative AI on critical thinking. <em>CHI 2025<\/em>. <a href=\"https:\/\/doi.org\/10.1145\/3706598.3713778\" target=\"_blank\" rel=\"noopener\">https:\/\/doi.org\/10.1145\/3706598.3713778<\/a><\/li>\n\n\n<li>Walters, W. H., &amp; Wilder, E. I. (2023). Fabrication and errors in the bibliographic citations generated by ChatGPT. <em>Scientific Reports<\/em>. <a href=\"https:\/\/doi.org\/10.1038\/s41598-023-41032-5\" target=\"_blank\" rel=\"noopener\">https:\/\/doi.org\/10.1038\/s41598-023-41032-5<\/a><\/li>\n\n<\/ul>\n","protected":false},"excerpt":{"rendered":"<p>Weeks of research now take hours. What the studies say about AI and real learning, why fast can feel like cheating, and how to use AI to learn, not just to finish.<\/p>\n","protected":false},"author":1,"featured_media":198,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[5],"tags":[],"class_list":["post-199","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-news-and-what-it-means"],"_links":{"self":[{"href":"https:\/\/hillwardconsults.com\/journal\/wp-json\/wp\/v2\/posts\/199","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/hillwardconsults.com\/journal\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/hillwardconsults.com\/journal\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/hillwardconsults.com\/journal\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/hillwardconsults.com\/journal\/wp-json\/wp\/v2\/comments?post=199"}],"version-history":[{"count":1,"href":"https:\/\/hillwardconsults.com\/journal\/wp-json\/wp\/v2\/posts\/199\/revisions"}],"predecessor-version":[{"id":200,"href":"https:\/\/hillwardconsults.com\/journal\/wp-json\/wp\/v2\/posts\/199\/revisions\/200"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/hillwardconsults.com\/journal\/wp-json\/wp\/v2\/media\/198"}],"wp:attachment":[{"href":"https:\/\/hillwardconsults.com\/journal\/wp-json\/wp\/v2\/media?parent=199"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/hillwardconsults.com\/journal\/wp-json\/wp\/v2\/categories?post=199"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/hillwardconsults.com\/journal\/wp-json\/wp\/v2\/tags?post=199"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}