One Click Away: AI Tutoring with Khanmigo in a Two-Year School Experiment
Added to the Kunskapsbank on 2026-09-07 but not on the old published site.
Lower secondary (grades 7–9) Across ages AI and learning
Links to claims
Supports (1)
| Claim | Basis |
|---|---|
| K26 (candidate) | Note carries the bank kill sentence behind K26 (group K26); no change in review |
Review state: first-pass draft, pending human review.
Source note (Swedish, verbatim from the Kunskapsbank)
Syntes: Oreopoulos & Low (2026) — Khanmigo two-year RCT (AI-015)
Källa: Oreopoulos, P., & Low, N. (2026). One Click Away: AI Tutoring with Khanmigo… NBER WP 35620. https://doi.org/10.3386/w35620
Original: 01-originaldokument/oreopoulos-low-2026-khanmigo/ (ABSTRACT; NBER PDF manuell)
ID: AI-015
Tier: RCT
Closed-book follow-up: delvis (nationella math percentiles; coach-mode, ej open GenAI)
HE vs K–12: grundskola-hog (18 TN middle schools; 2 år)
Evidensstyrka: hög ekologisk validitet; effekt liten; engagement-mekanism tydlig.
Metadata
- ID:
AI-015 - Tier: RCT
- N / design: cluster-RCT (2 år; Khan Academy + Khanmigo coach-mode under remedial math)
- Population: grundskola-hog (middle school)
- Closed-book follow-up: delvis (nationella percentiler / achievement; AI coach-konfigurerad, ej open ChatGPT)
- HE vs K–12: grundskola-hög (middle school, Tennessee; 18 skolor)
- Subject: One Click Away: AI Tutoring with Khanmigo in a Two-Year School Experiment
Finding (+ comparator)
Assignment till Khan+Khanmigo (coach) under daily remedial math: +1,3 national percentile ranks/term (~0,06–0,08 SD/år); full aktivt år ≈0,14 SD. Comparator: skolans remedial session utan denna assignment. Effekten liknar Khan practice utan AI. 96 % provade AI; median använde det ~1/3 av practicedagar; endast 17 % av mistagssessioner hade substantiv dialog.
Mechanism
Bindande constraint = engagement, inte modellkapacitet. “Access ≠ use ≠ tutoring dialogue.” Pedagogisk konfiguration (coach) otillräcklig om elever klickar/svarar kort.
Limits
Cluster RCT; specifikt Khanmigo+Khan; ej Bastani-stil closed-book inom samma lektion; NBER WP.
This claim dies if …
K–12-RCT där coach-AI med hög substantiv dialog-rate (≥ median elev) fortfarande ger nolleffekt vs samma platform utan AI på standardiserad math.
Swedish implication
Stop räkna inlogg/licens som evidens för AI-tutor; start logga misstag→dialog-andel; measure percentile/NP + usage fidelity innan uppskalning.
Links
- Relaterade: AI-014, AI-003, AI-011, AI-010
- DOI: https://doi.org/10.3386/w35620
Metadata as recorded
- ID
- AI-015
- Tier
- RCT
- Closed-book follow-up
- delvis (nationella percentiler / achievement; AI coach-konfigurerad, ej open ChatGPT)
- HE vs K–12
- grundskola-hög (middle school, Tennessee; 18 skolor)
- Titel
- One Click Away: AI Tutoring with Khanmigo in a Two-Year School Experiment
- Författare
- Oreopoulos, P.; Low, N.
- År
- 2026
- Källa/DOI
- NBER Working Paper 35620. https://doi.org/10.3386/w35620
- URL
- https://www.nber.org/papers/w35620
- Typ
- cluster-RCT (2 år; Khan Academy + Khanmigo coach-mode under remedial math)
- Åldersgrupp
- grundskola-hog (middle school)
- Teman
- ai-och-larande
- Sparade fil
- ABSTRACT.md (NBER PDF kan kräva inlogg/manuell)
- Hämtad
- 2026-09-07
- Paywall
- NBER (abstrakt OA)
- Obs
- Assignment ≈+1,3 NPR/term (~0,06–0,08 SD/år); liknar Khan utan AI. Bindande constraint = engagemang (median AI-meddelanden få; 17 % av mistagssessioner).
Original files held in the project (not published): ABSTRACT.md