Colombia's Teachers and AI: What the Survey Actually Measures
Colombia's lower secondary teachers reported AI use and infrastructure barriers in TALIS 2024. What that evidence can tell a school planning its next step.
TL;DR
- Colombia's October teacher-cooperation announcement brings AI training into focus. The available survey describes lower secondary teachers in 2024, before this visit and the new law.
- Reported use, a teacher's perception of a barrier and a measured classroom result answer different questions. Keep them separate when planning support.
- A school can check its equipment and practise verifying teaching material with staff, without exposing children's information or treating a short exercise as certification.
A teacher can know how to use an AI tool and still have trouble getting a lesson online. Another can have a reliable connection but need help checking the answer. Colombia's recent education news gives school leaders a reason to examine both problems, rather than assume one training course will solve them.
What the October visit covered
The Ministry of National Education's 3 October 2026 release reports a four-day visit by a South Korean delegation to Bogotá, Medellín, Envigado and Ubaté. School visits and workshops addressed the pedagogical and ethical use of generative AI and Vibe Coding.
The ministry reports that the wider cooperation programme has trained 300 teachers from 25 departments over almost twenty years, reaching more than 12,000 students. Those cumulative figures are not the attendance or measured impact of the October workshops. The ministry also proposed expanding technology classrooms, student mobility and shared teaching resources.
A workshop announcement can describe what was offered. To judge what changed afterwards, a school would need different evidence: whether staff could use the material, whether it suited their subjects, how much review time it required, and what students learned. None of those outcomes follows from the number of people who attended.
Read the survey with its population attached
The OECD TALIS 2024 Colombia note concerns lower secondary teachers and principals, with responses collected in 2024. It relies on self-reports. Among teachers, 53% reported AI use at work in the preceding year, compared with an OECD average of 36%.
In the same OECD note, among non-users, 76% reported insufficient school infrastructure, versus 37% across the OECD; 64% reported lacking knowledge and skills, versus 75%. Separately, 43% of teachers reported high professional-learning need in AI skills. These groups and questions should not be combined into a single readiness score.
The figures describe reported experience, not a national equipment audit or evidence of improved student attainment. They also predate the 2026 law and October visit, so they cannot measure either one's effect. The AI-skills question does not establish exactly which kind of training each respondent wanted.
For school leaders elsewhere in Latin America, the findings offer questions to ask locally. Ask staff where a planned lesson stalls: device access, network reliability, approved software, subject knowledge, preparation time or checking an answer. Keep more than one answer possible. Treat a complaint as a reason to investigate, rather than proof that a particular purchase will fix the problem.
The World Bank’s April 2024 account of Amazonas supplies a different kind of context. Its survey interviewed 504 households and 348 businesses; more than half of the households lacked an online connection and 18% lacked coverage. Those findings concern Amazonas, not all Colombian households or schools. They do not replace a check at the place where a lesson will run.
Shadow Literacy in a staff-room discussion
Adrian Dunkley's Shadow Literacy framework describes the judgement needed to recognise plausible but wrong AI output. This article applies that idea to reviewing teaching material. It does not attribute a particular school exercise or view of Colombian policy to him.
For a history lesson, checking might mean tracing a date to an approved reference and asking whether the source supports the interpretation. For a mathematics worksheet, it means working through the solution, including units and assumptions. A fluent explanation can still contain a mistake that is difficult for a learner to identify.
The reviewer should also ask what the lesson is meant to teach. A correct answer that skips the reasoning may be unsuitable for a class practising that reasoning. Review the task, the proposed explanation and the learning objective together. Model speed alone is a poor measure of the teacher's total preparation work.
What the new law changes, and what a school still needs to ask
Ley 2626 de 2026 updates technology and computing education. Article 2 applies it to public educational institutions, while allowing private institutions to adopt the proposed model. Article 4 sets a six-month-from-promulgation period for a teacher-training programme. Articles 7 and 14 also contain six-month provisions.
MinTIC's legal compilation identifies publication in Diario Oficial 53.602 on 25 August 2026. Article 5 calls for progressive infrastructure and connectivity provision, prioritising underserved areas within existing fiscal resources. A statutory requirement is not evidence that every school already has equipment or a course place. School leaders should check current implementation guidance and their territorial timetable with the relevant education secretariat; this article is not a determination of legal compliance.
For a school in Ubaté, useful questions include which programme its teachers can access, who funds participation, whether the materials work with its current devices, and where to report a recurring connection problem. Record the answer and the office responsible. If guidance has not arrived, avoid filling the gap with a vendor's promise presented as an official requirement.
For Caribbean readers, the transferable idea is the sequence of checks, not a Colombian percentage. Another Caribbean or Latin American school can ask about its own connection, curriculum and review practices without assuming its staff have the same needs. The local education authority's rules and the school community's experience still determine what is appropriate.
A small, supervised exercise for this month
The dates below are suggested planning dates, not official deadlines. Agree the work with the school leader and technical staff first. Keep pupils' names, grades, messages and other personal information out of the exercise.
- Check access by 21 October. With staff and approved school devices, try the same public learning page on up to ten devices at once. Record the number that load, elapsed time and any accessibility problems. Repeat at another relevant time. Do not use pupils' accounts or ask them to provide personal devices. This is a limited diagnostic check, not a certification of the network.
- Test five teaching examples by 23 October. A subject specialist prepares fictional, curriculum-relevant answers, including two deliberate errors, with a checked answer key. Invite colleagues to explain what they would verify and why. Record anonymous group counts of missed errors and time spent checking. Participation should support learning, not staff ranking.
- Discuss the checks, then try fresh material. Show the evidence for each correction. Use a separate set later, so improvement is not simply recall of the original answers. A small group can surface useful problems, but no minimum of ten participants makes the result representative or proves teaching competence.
- Ask about implementation. Send the education secretariat a focused request through the school's usual channel about training access and the local timetable. If the school chooses 30 October as a follow-up date, label it as its own planning date. An unanswered message alone does not establish a legal breach.
Compare the exercise with an ordinary preparation task, including the time needed to find sources and correct drafts. If a tool produces an answer quickly but takes longer to check, that result matters. Keep an offline version of the activity available so a connection failure does not exclude staff from the discussion.
Protect the learner while testing the tool
AI may help a teacher draft alternatives or organise material, but any time saving needs to include review. An incorrect worksheet can confuse a class, and an unsuitable explanation may hide rather than resolve a learning difficulty. Keep a qualified teacher responsible for material before it reaches pupils. The staff-room exercise cannot validate automated grading, admissions, safeguarding or other consequential decisions about children.
Privacy needs a separate check. Do not paste identifiable student work, health information, grades or private messages into an unapproved service. Removing a name may leave a pupil identifiable through circumstances or a combination of details. Use invented examples, follow school policy and obtain appropriate advice before processing real records. Consent alone does not settle every duty involving children.
Access also affects whether an activity is fair. Offer a paper or teacher-led alternative where needed, check assistive-technology requirements and avoid grading pupils on access to a paid tool. Staff should be able to raise concerns without being labelled resistant to innovation. An honest report of a limitation is evidence the school can use.
What this analysis cannot establish
The weakest step would be treating a teacher's report of infrastructure difficulty as a diagnosis of a specific school's equipment. We have not inspected that school. The proposed exercise can help it document a problem, but it cannot establish national prevalence, prove a causal effect on learning or rank schools against one another.
Publisher disclosure: The Genius Project is part of the StarApple AI network. The suggested exercises are the publisher's applications of a general framework. They are not a ministry programme, a validated assessment or a finding that a particular classroom product is safe.
Frequently Asked Questions
How should a teacher record the practice results?
Keep only the group information needed to improve the exercise, such as which example caused confusion and how it was checked. Avoid names and individual rankings. Agree who can see the notes and when they will be deleted. A training discussion should not silently become a personnel assessment.
Can this exercise run without an AI account?
Yes. A teacher can prepare printed examples and reference extracts in advance. Colleagues can practise checking them on paper. If a model is used to draft material, an approved adult reviewer should verify the examples and answer key before the session; participants do not need individual accounts.
What should a school ask a training provider?
Ask for the curriculum connection, prerequisites, accessibility arrangements, cost and expected staff time. Request examples of how sources are checked and mistakes corrected. Also ask what evidence supports claimed benefits, which learners it concerns, and whether the provider will process any personal information.
How can pupils take part safely later?
Use age-appropriate, teacher-led activities under school policy, with approved materials and no requirement to create an account or share private information. Invite pupils to explain how they checked a claim. Do not use deliberate misinformation in an assessed task without clear teaching context and correction.
Where can readers find the project’s own activities?
The Genius Project’s programme pages describe its activities and participation requirements. Read the current rules with a parent, guardian or school where appropriate, and check safeguarding and privacy arrangements before taking part. A link from this article does not make a programme an official Colombian curriculum requirement.
Related reading: The Genius Project blog, StarApple AI and the Caribbean AI Association.
Supported by StarApple AI.