AI For Global Education

Directions Website

AI For Global Education We provide free resources, training, and advocacy to ensure AI enhances learning

AI for Global Education is a non-profit organisation dedicated to helping educators, students, and policymakers integrate AI into global education ethically and effectively.

20/10/2026

Engineering lecturers in a new Frontiers in Education study describe homework grades near 100% alongside widely varying exam results.

Saravanan, Meissner and Glenn draw on focus groups with 16 engineering faculty at one American research university in November 2025. Tools such as ChatGPT can write working code and solve multi-step problems, the work coursework is meant to build. The authors call the lecturers’ stance reluctant engagement: they accommodate AI because it feels inevitable, rather than out of conviction.

Some lecturers moved marks from homework onto exams. The authors warn that students may lose low-stakes practice, weakening the feedback loop between practice and exam performance. Faculty said they’re adapting largely without coordinated support. The study reports what faculty described and measured no student outcomes.

What we’d take from this: when homework stops showing what students can do, moving marks to exams makes sense, yet it can cost them practice. Before your department reweights assessment, ask where students will still get feedback on work that doesn’t count. With 16 lecturers at one university, treat the study as questions to test in your own programmes.

Our courses on AI in education are at learn.aiforglobaleducation.org/courses.

Source: Saravanan, Meissner and Glenn (2026), Frontiers in Education. https://doi.org/10.3389/feduc.2026.1880863

In PISA 2025, education systems where students use AI more for schoolwork tend to score lower in science, with clear exc...
19/10/2026

In PISA 2025, education systems where students use AI more for schoolwork tend to score lower in science, with clear exceptions.

The OECD chart plots the mean science score of each system against how often its 15-year-olds use AI chatbots for schoolwork. The index covers summarising a text, preliminary research, drafting written work and help with learning. The trend line slopes down, but the spread is wide. Singapore reports high use and scores near the top. Japan reports the lowest use on the chart and also scores above the OECD average.

The OECD says policy may partly explain differences in AI use among high-performing systems. Singapore has a large-scale digital learning policy that explicitly includes AI tools, while Japan and Estonia have less comprehensive ones. It also raises a possibility running the other way: lower-performing systems may have adopted AI faster to compensate.

Our view: a country-level trend line can’t tell a university what AI does to learning, and the OECD doesn’t claim it does. If you recruit from several of these systems, expect your students to bring different school experiences of AI. We’d ask about those experiences at induction and plan first-year teaching around the answers.

Our courses on AI in international education are at learn.aiforglobaleducation.org/courses.

Chart data: OECD (2026), PISA 2025 Results (Volume I), Figure I.4.17.

An AI’s first answer to a take-home exam question would most likely have passed, yet it named key arguments without expl...
15/10/2026

An AI’s first answer to a take-home exam question would most likely have passed, yet it named key arguments without explaining them.

Brunnström and Palmqvist gave a chatbot a cognitive science exam question, ‘Do machines think?’, while playing a naive student. The answer met the assignment criteria, and they describe it as “a collection of factoids dressed in fluent academic prose”. At several moments, they felt that reading the course book would have been simpler and clearer than negotiating with the AI.

They propose that take-home exams assess how students connect and evaluate ideas, and suggest recall-based knowledge suits in-class assessment without AI. They also want teachers to teach AI-interaction literacy explicitly. The evidence is one conversation run by the authors, with no student participants.

For institutions reviewing take-home assessment, a simple check follows. Could a fluent answer that lists the right arguments without explaining them pass your criteria? If so, your criteria may be rewarding coverage over connected reasoning. We’d pick one take-home task per programme and ask what a student must explain, link or evaluate to earn more than a pass.

Our courses on AI in education are at learn.aiforglobaleducation.org/courses.

Source: Brunnström and Palmqvist (2026), Assessment & Evaluation in Higher Education.

13/10/2026

In a UNESCO IESALC survey of 200 universities, mostly in Latin America, 87 per cent report using AI but only 26 per cent have a formal AI strategy.

The working paper, produced with UNU-IAS, finds that adoption has come from the ground up. Individual academics and students drive it, and internal champions sustain it, rather than national or institutional policy. The report calls that fragile, since progress depends on the right people being in the right place.

Few of these universities check what their AI is doing. Only 9 per cent report formal evaluation, and 57 per cent report no monitoring or evaluation at all. Senior leaders answered in 2025, and the report reads its self-selected sample as indicative of trends, not statistically representative.

What we’d ask of any university, in any region: if your most active AI champion left tomorrow, what would stop? A named owner, a written strategy and a few indicators you review each term could keep that work going. Without them, you can’t tell whether the tools are doing what you bought them for.

Our courses on AI in international education are at learn.aiforglobaleducation.org/courses

Source: UNESCO IESALC, AI implementation in higher education in Latin America and the Caribbean, Working Paper 16, 2026. https://doi.org/10.54674/WP.2026.103

Gartner’s IT 2030 diagram sets pre-AI IT beside an AI-first version, with changes from software and data to vendors, ope...
12/10/2026

Gartner’s IT 2030 diagram sets pre-AI IT beside an AI-first version, with changes from software and data to vendors, operations and governance.

In the AI-first version, business teams and business operations stay in place. Existing assets are transformed: software becomes agent-ready, data and BI become AI-ready, and operations become semiautomated. New assets appear beside them, including AI models, AI infrastructure, new vendors and AI agents next to IT employees assisted by AI. Governance and finance become AI-driven, and risk and security are augmented.

For institutions, this is a CIO’s picture of a corporate estate. In a university, student records, learning platforms and research computing may sit with different owners and contracts. We’d start with the layer Gartner labels non-AI legacy: which of your systems will stay that way, and who decides? Your answer shapes the vendor conversations that follow.

For university digital teams, our courses on AI are at learn.aiforglobaleducation.org/courses.

Source: Gartner, IT 2030: How CIOs Should Reinvent IT for the AI Age.

AI has moved from pilot activity to operating risk.For universities, the question is no longer whether tools are being t...
09/10/2026

AI has moved from pilot activity to operating risk.

For universities, the question is no longer whether tools are being tried, but whether leaders can evidence value, ownership, assessment redesign and staff capability. Policies are not enough: governing bodies need records of what is deployed, how it is monitored and who is accountable.

Engineering lecturers in a new American study say AI detectors cannot prove misuse, and they are moving towards assessme...
08/10/2026

Engineering lecturers in a new American study say AI detectors cannot prove misuse, and they are moving towards assessment redesign.

“You can’t prove it because none of the AI checkers are accurate,” said one lecturer in the study by Saravanan, Meissner and Glenn. Faculty also described fewer visits to office hours, shallower questions, and student teams turning on each other over suspected AI use. The authors note this effect on group work has been underreported in earlier research.

They suggest different rules for different levels. Limit or structure AI in foundational courses, where students still need productive struggle, then scaffold critical AI use once students can judge its output. They also ask for assessment that makes thinking visible, such as in-class problem solving, oral exams, design documentation and tasks that critique AI output. The study rests on 16 faculty at one university in the southern United States.

For institutions, the detector finding is the easiest to act on. If your lecturers can’t prove misuse, detection may not carry an integrity policy on its own. We’d ask each programme which foundational modules need AI limits, and which later ones could teach students to check AI output.

Our courses on AI in education are at learn.aiforglobaleducation.org/courses.

Source: Saravanan, Meissner and Glenn (2026), Frontiers in Education.

06/10/2026

Fifteen-year-olds who use AI chatbots for specific schoolwork tasks score lower in science, on average, in the OECD’s PISA 2025 results.

PISA asked students how often they use chatbots to summarise a text, research a new topic or draft written work. Across OECD countries, students who don’t use AI for these tasks outperform those who do.

The OECD is careful with this finding. It says the gap may reflect who uses AI and how, and doesn’t necessarily mean AI lowers scores. Among students who use AI to help them learn, moderate users score slightly higher than rare or heavy users. Frequent users who are taught to judge what AI produces also do slightly better, and disadvantaged students are less likely to get that teaching.

What we take from this: some of the students in these data will be in your first-year classes within a few years. The pattern is a correlation, so it won’t tell you whether to restrict AI. It does point to a question you can answer now. Where do your first-year courses teach students to check what a chatbot gives them, and does every student get that teaching?

Our courses on teaching and learning with AI are at learn.aiforglobaleducation.org/courses.

Source: OECD (2026), PISA 2025 Results (Volume I).

Students are the main users of AI in the universities UNESCO IESALC surveyed, yet its new working paper finds they are t...
01/10/2026

Students are the main users of AI in the universities UNESCO IESALC surveyed, yet its new working paper finds they are the least protected.

Nearly half of those universities, 49 per cent, report more student confusion about appropriate AI use. Only 24 per cent have academic integrity guidelines on student AI use that are clear and actively communicated to students.

The gap also runs between sectors. Private non-profit universities are almost three times as likely as public ones to have a formal AI strategy, at 39 per cent against 14. Public universities serve the largest, mostly lower-income student populations, so the paper sees equity implications beyond higher education. It recommends a governance body where student representation is structural, not advisory, and AI literacy as a graduation-level competency.

For institutions in any region, start with a question for your students. Could they say what AI use is allowed on their next assignment? If they can’t, we’d fix the guidance before the strategy, and we’d write it with students at the table. The survey covers senior leaders at 200 institutions and may not reflect your own sector.

Our courses on AI in international education are at learn.aiforglobaleducation.org/courses

Source: UNESCO IESALC, AI implementation in higher education in Latin America and the Caribbean, 2026. https://doi.org/10.54674/WP.2026.103

Address

CIO AI For Global Education The Gateway 85-101 Sankey Street
Warrington
WA11SR

Website

Alerts

Be the first to know and let us send you an email when AI For Global Education posts news and promotions. Your email address will not be used for any other purpose, and you can unsubscribe at any time.

Shortcuts

Share

Category