Industries / Government, education and community / Education
Classroom teaching
A teacher leads a group of students through lessons in a room, from preschool to the university lecture hall.
7
Milestones achieved, from 859 to 2020.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
859
The oldest working university
Al-Qarawiyyin in Fez, Morocco, dates its founding to 859 and still teaches. Guinness World Records lists it as the oldest educational institution still operating, and lists the University of Bologna as the oldest university in Europe.
LessonFounding dates can be set after the fact; Bologna's own history calls its 1088 date "conventional", fixed for an 800th-anniversary celebration held in 1888.
Credit
al-Qarawiyyin, Fez; University of Bologna
1798
One teacher, hundreds of pupils
Andrew Bell in Madras and Joseph Lancaster at Borough Road in London ran schools in which older pupils, called monitors, taught small groups while one master oversaw the whole room. Lancaster advised that one master could teach 1,000 pupils, and Britain's first government grant to schools, £20,000 in 1833, went to the two societies that ran schools this way.
LessonMonitors were children, not trained teachers, and in the 1840s England set up a paid pupil-teacher scheme and teacher exams to improve on the system.
tes.combritishschoolsmuseum.org.ukparliament.ukopen-education-repository.ucl.ac.uk
Credit
Andrew Bell; Joseph Lancaster; the British and Foreign School Society and the National Society
1840
Froebel opens a kindergarten
On 28 June 1840 Friedrich Froebel founded the "General German Kindergarten" in Blankenburg, Thuringia, where young children learned through play with set materials he called gifts. It closed for lack of money in 1844, and Prussia banned kindergartens in 1851. The Froebel Trust calls him the inventor of the kindergarten.
Credit
Friedrich Froebel
1985
Small classes, chosen by lottery
Tennessee's Project STAR randomly assigned about 11,600 children in 79 public schools, from kindergarten to third grade, to small classes of 13 to 17 pupils or regular classes of 22 to 25, starting in 1985–86. The economist Alan Krueger found that test scores rose about 4 percentile points in a child's first small-class year, and a follow-up with Diane Whitmore found small-class pupils, especially minority students, more likely to take a college-entrance exam.
Credit
State of Tennessee; Alan Krueger and Diane Whitmore, Princeton University
2007
Whiteboards without test gains
From 2003–04 England paid for interactive whiteboards, large touch screens wired to a computer, in London secondary schools. An Institute of Education evaluation for the Department for Education and Skills found "no impact on pupil performance" in the first year that departments were fully equipped.
LessonEquipment alone did not change results; the evaluators advised that discussion of teaching come before discussion of the technology.
Credit
Gemma Moss, Carey Jewitt, Ros Levačić and colleagues, Institute of Education, University of London
2017
Teaching children at their level
The Indian organization Pratham tests children and, for part of the day, groups them by what they can actually read and calculate rather than by grade. Researchers at J-PAL, a poverty research center at MIT, tested the approach in randomized trials from 2001 and in 2017 reported two versions that raised learning inside government schools at scale; J-PAL counts more than 60 million students reached in India and Africa.
LessonSome designs failed inside regular schools before workable ones were found.
Credit
Pratham (Rukmini Banerji); Abhijit Banerjee, Esther Duflo and colleagues at J-PAL
2020
Schools close worldwide
COVID-19 school closures affected more than 1.6 billion students, by UNESCO's count. In the US, 9-year-olds' average scores on the national long-term trend test fell 5 points in reading and 7 in math between 2020 and 2022, the first drop in math that test had recorded.
LessonLong closures showed up as measured learning losses two years later.
Credit
UNESCO; US National Center for Education Statistics
Watch for
AI scores lessons on Teach
Teach is the World Bank's tool for scoring the quality of a lesson from classroom observation. World Bank researchers say the best of eight setups of Google's Gemini 2.0 and 2.5 Flash, scoring typed transcripts of 199 Peruvian primary-school lessons, met Teach's certification rule (within one point of a certified observer) in 92.9% of simulated exams. The comparison was against a single certified field observer, not the program's master coders.
Achieved whenA ministry uses the model's scores in its routine observation program, or an outside team repeats the result on new lessons against master coders.
Claimed by
Carolina Lopez, Ezequiel Molina and colleagues, World Bank, July 2026
Open
01
Read the board aloud
A camera at the back of a classroom that reads what a teacher writes on the board, including math, and sends it to a blind student's refreshable braille display or earphones within seconds, with a way to move back through earlier lines. Point the camera at the board only, so no students are filmed.
The briefCredit · unclaimed02
Teach scores from a recording
A pipeline that takes an audio recording of a lesson, made on a phone with a lapel microphone on the teacher, transcribes it in the languages used in class, and scores it on the World Bank's Teach tool. Test it in a country where lessons mix English or French with a local language. Collect recordings only through a school or ministry partner, with consent and ethics approval.
The briefCredit · unclaimed
Open 01 · the brief
Read the board aloud
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A camera at the back of a classroom that reads what a teacher writes on the board, including math, and sends it to a blind student's refreshable braille display or earphones within seconds, with a way to move back through earlier lines. Point the camera at the board only, so no students are filmed.
Why it is still open
The projects found are ideas or prototypes without measured results. An Arizona State University student team, EquiBraille, planned in December 2022 to trial board-reading cameras in entry-level math and English courses and reported no testing results.
Built with what exists
A webcam or phone on a tripod; a vision-language model that reads handwriting and math, such as Gemini or Claude; MathCAT, an open-source library that turns math into speech and braille codes such as Nemeth and UEB, and is used by the NVDA and JAWS screen readers; a refreshable braille display.
First step
Film the board during three lessons of a consenting math teacher, run the frames through a vision model and MathCAT, and count symbol errors by hand.
Done looks like
On 20 filmed lessons with handwritten math, at least 95% of written symbols reach the braille display correctly within 10 seconds of being written, checked line by line against a typed copy of the board; and three braille readers answer questions about the lesson as accurately from the system as from a sighted person reading the board aloud.
Open 02 · the brief
Teach scores from a recording
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A pipeline that takes an audio recording of a lesson, made on a phone with a lapel microphone on the teacher, transcribes it in the languages used in class, and scores it on the World Bank's Teach tool. Test it in a country where lessons mix English or French with a local language. Collect recordings only through a school or ministry partner, with consent and ethics approval.
Why it is still open
The July 2026 World Bank study scored typed transcripts from one country, in Spanish, and its raw recordings are not shared because they identify people. This research found no study that scores Teach from classroom audio, or in a classroom that mixes languages.
Built with what exists
The World Bank's Teach observer manual; speech-to-text that covers the classroom's languages, such as Meta's MMS model or ElevenLabs Scribe v3 (see the Note "Fewer word errors"); an AI model such as Gemini given the full Teach rubric, which the Peru study found was the main driver of agreement.
First step
Ask the Peru study's authors for the de-identified data they say is forthcoming, and record three lessons of one consenting teacher to measure transcription errors.
Done looks like
On 50 lessons each scored by a certified Teach observer, the pipeline meets Teach's within-one-point rule in at least 90% of simulated certification exams, close to the Peru study's 92.9%, with code and scoring prompts published.