BRENNEN

Industries / Government, education and community / Education / EdTech

Making course material

Writing, recording, licensing, translating and adapting the lessons, videos and textbooks that courses are built from, and making them usable by every student.

6

Milestones achieved, from 1967 to 2026.

1

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1967

    A language for lesson authors

    Paul Tenczar created TUTOR, at first briefly named TEACHER, a programming language for writing PLATO lessons, and it was open to any interested University of Illinois student, and beyond, after 1967. Tenczar wrote in 1974 that TUTOR was "designed to allow teachers lacking computer experience to author complex materials without the aid of a programmer". Some companies were still using educational programs written in TUTOR in the early 2000s.

    distributedmuseum.illinois.edu

    Credit

    Paul Tenczar and the PLATO group, University of Illinois

  2. 2002

    MIT puts courses online

    On 4 April 2001 MIT announced OpenCourseWare, a plan to publish the materials of nearly all its courses on the web, free to anyone. The pilot site opened on 30 September 2002 with 32 courses, among them Gilbert Strang's linear algebra course with more than 30 video lectures.

    news.mit.edunews.mit.edu

    Credit

    MIT (President Charles M. Vest; OpenCourseWare executive director Anne H. Margulies)

  3. 2002

    Licenses for sharing

    Creative Commons, a nonprofit founded in 2001 by Lawrence Lessig, Hal Abelson and Eric Eldred, published its first licenses in 2002. They are free, public licenses that let a creator keep copyright while allowing others to copy and adapt the work on stated terms, in place of the default "all rights reserved". Creative Commons says its licenses are now used on tens of billions of works.

    creativecommons.org

    Credit

    Lawrence Lessig, Hal Abelson and Eric Eldred, Creative Commons

  4. 2012

    Free textbooks, peer reviewed

    OpenStax at Rice University grew out of Connexions, an online library of open teaching materials that Richard Baraniuk started in 1999, and began publishing its own free, peer-reviewed college textbooks in 2012. By OpenStax's own figures, it had more than 80 textbooks by 2023, 71% of US colleges and universities use its books each year, and students have saved $3.8 billion in all; its impact page does not explain how these figures are counted.

    openstax.orgopenstax.org

    Credit

    Rice University (Richard G. Baraniuk)

  5. 2017

    Berkeley hides 20,000 lectures

    After the US Justice Department found that UC Berkeley's free public lecture videos and podcasts had to meet higher accessibility standards to stay public, Berkeley announced on 1 March 2017 that it would move its more than 20,000 legacy recordings on YouTube and iTunes U behind a campus login instead.

    LessonMaking years of old recordings accessible cost more than the university chose to spend, so the public lost them; Berkeley said it would make new public material accessible from the start.

    news.berkeley.edu

    Credit

    UC Berkeley (Cathy Koshland, vice chancellor for undergraduate education)

  6. 2026

    Accessibility deadline pushed back

    A 2024 US Justice Department rule under the Americans with Disabilities Act requires public schools and colleges to make their websites, apps and digital textbooks meet WCAG 2.1 level AA, a widely used web accessibility standard. In April 2026 the department moved the first deadline, for places with 50,000 or more residents, from 24 April 2026 to 26 April 2027, citing education groups' concern that compliance "requires significant resources and staff time".

    LessonThe requirement arrived before most institutions were ready; in a 2025 survey by the National School Public Relations Association, 14% of respondents said their districts had completed or nearly completed the work.

    k12dive.com

    Credit

    US Department of Justice

Watch for

  1. Lectures dubbed by AI

    Coursera says it used generative AI to translate and dub the video of more than 100 popular courses into Spanish, French, German and Brazilian Portuguese in a reproduction of the original instructor's voice, and that learners finish translated courses at higher rates and "nearly 25% faster" than courses offered only in the original language. Its post gives no method or sample size and does not say whether people review the dubbed audio. Duke University repeated the completion claim when it put two of its Coursera specializations into the program.

    Achieved whenAn outside study compares completion and test results between dubbed and original versions of the same courses, and counts translation errors in the dubbed audio.

    blog.coursera.orgonline.duke.edu

    Claimed by

    Coursera, April 2025; restated by Duke, October 2025

Open

  1. 01

    Errors in AI-built language courses

    An outside count of the errors in language courses built with generative AI. Take the first two units of three of the courses Duolingo launched in April 2025, choosing language pairs whose native speakers are easy to find, such as German taught to Japanese speakers. Two native speakers per pair mark every sentence and translation by error type and severity, using MQM (Multidimensional Quality Metrics, a public scheme for classifying translation errors), and mark the same units of an older Duolingo course in the same language for comparison. Work only from what a test account shows.

    The brief
    Credit · unclaimed
  2. 02

    Check licenses inside open textbooks

    A checker that lists every image and figure in an open textbook, traces each to its original, and flags any whose license conflicts with the book's license, or whose credit line is missing or incomplete, for a librarian to confirm.

    The brief
    Credit · unclaimed

Open 01 · the brief

Errors in AI-built language courses

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

An outside count of the errors in language courses built with generative AI. Take the first two units of three of the courses Duolingo launched in April 2025, choosing language pairs whose native speakers are easy to find, such as German taught to Japanese speakers. Two native speakers per pair mark every sentence and translation by error type and severity, using MQM (Multidimensional Quality Metrics, a public scheme for classifying translation errors), and mark the same units of an older Duolingo course in the same language for comparison. Work only from what a test account shows.

Why it is still open

Duolingo launched 148 new courses in April 2025, mainly for beginners at levels A1 and A2 of the Common European Framework of Reference, and its chief executive, Luis von Ahn, said generative AI let it build almost 150 courses in one year, against about twelve years for its first 100. He said the company's staff would focus on monitoring and improving quality. This research found users' complaints about the new courses but no outside count of their errors, and no count for Coursera's AI-dubbed courses either.

heise.de

Built with what exists

A free Duolingo account and screen recordings; the MQM error categories; an AI reviewer to propose candidate errors for the native speakers to confirm (the Note "Seven more flagged" covers a Claude-based reviewer of long documents); a shared spreadsheet and Cohen's kappa, a standard measure of how far two reviewers agree.

  1. First step

    Record the first unit of one new course with a test account and have one native speaker mark 100 sentences to settle the categories.

  2. Done looks like

    1,000 sentences from three new courses and 1,000 from older courses, each marked by two native speakers; errors per 100 sentences by severity for each group, with the reviewers' agreement reported and the full list of errors sent to Duolingo.

Open 02 · the brief

Check licenses inside open textbooks

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A checker that lists every image and figure in an open textbook, traces each to its original, and flags any whose license conflicts with the book's license, or whose credit line is missing or incomplete, for a librarian to confirm.

Why it is still open

Tel Amiel and Tiago Chagas Soares audited 50 repositories of open educational resources in 2016 and found a "high level of incongruity" between the licenses shown on resources, their catalog pages and the sites' policies, but they checked whole resources, not the images inside them. A course led by the Commonwealth of Learning lists incomplete credit lines and incompatible licenses in remixed work among the common mistakes. This research found no tool that checks the media inside an open book.

irrodl.orgpacificopencourses.col.org

Built with what exists

The books' openly licensed source files, for example from Pressbooks or OpenStax; the Openverse and Wikimedia Commons programming interfaces to find licensed originals; TinEye's paid reverse image search interface; a vision-language model to read credit lines in captions; Creative Commons' published chart of which licenses can be combined.

  1. First step

    Pull every image from one open textbook on Pressbooks, check 50 by hand against Openverse and Wikimedia Commons, and count how many credit lines are complete.

  2. Done looks like

    Ten open textbooks with about 2,000 images checked; a librarian reviews a random 300 flags and 300 passes; the checker catches at least 90% of the real problems with no more than one false flag in five; confirmed problems reported to each publisher.