Delhi students improve their math and Hindi scores

Mindspark. [1]

Rules written by education specialists selected lessons from students’ answers and error patterns. [1]

The study does not identify a numbered software release. [1]

Educational Initiatives. [1]

The software assessed each student and selected exercises at the appropriate level. [1]

Students used individual computers with headphones for 45 minutes, then spent 45 minutes with an instructor in a small group. [1]

Students’ answers, scores, and errors determined the next exercises and explanations. [1]

The software selected exercises automatically. Instructors could assign particular topics. [1]

Each session combined practice on Mindspark with instruction in groups of about 12–15 students. Centers offered six sessions a week. [1]

Education specialists wrote the teaching rules. Trained instructors supervised students and provided group instruction. [1]

Mindspark used rules for adaptive teaching. It did not use machine learning. [1]

Editor’s note. Treating the system as symbolic AI is an editorial classification of its expert-written rules. The authors expressly state that it did not use machine learning.

The trial evaluated software and instruction together. It did not measure the software’s separate effect. [1]

Not assessed

Students offered a voucher scored higher in math and Hindi after 4.5 months. [1]

Improved learning. [1]

Observed in a randomized evaluation. [1]

Students. [1]

Public-school students in Delhi, most of whom were in grades 6–9 and behind their grade’s curriculum. [1]

The trial enrolled 619 students and offered 314 a voucher. Researchers tested 539 students at the end. [1]

Gains were similar across initial ability levels. Students who began further behind gained more relative to their previous progress. [1]

The measured learning gains covered 4.5 months. [1]

Improved learning was the intended benefit. [1]

The centers required staff and equipment. Low enrollment and the loss of subsidies led all three to close. [1]

AIBD-0212-E01 [1]

A lottery assigned students to a free voucher or a comparison group. Researchers estimated the effect of the offer. [1]

Independent paper tests assessed math and Hindi before and after the program. Scores were placed on comparable scales. [1]

There were 314 voucher recipients and 305 controls. Endline participation was 85% and 90%, respectively. [1]

Math and Hindi test scores. [1]

Students who were not offered a voucher continued their usual schooling and activities. [1]

The voucher offer raised math scores by 0.37 standard deviations and Hindi scores by 0.23. [1]

The design did not separate adaptive software from additional teaching and practice. [1]

Participants volunteered for the study. Endline attrition differed between groups, and longer-term attainment was not measured. [1]

Average attendance was 58%, or about 50 of the 86 available days. [1]

Replication data are deposited with the American Economic Association at openICPSR. [1]

Program cost and subsidized fee. Separate estimates describe possible costs at scale. [1]

Delivery cost about INR 1,000 per student per month. The subsidized fee was INR 200, which vouchers waived. [1]

Indian rupees. The authors gave an approximate equivalent of US$15 for the monthly delivery cost. [1]

Per student per month during the Delhi pilot. [1]

The INR 1,000 figure describes pilot delivery cost. Lower costs at scale were projections. [1]

Families normally paid the subsidized fee. Donors funded the remaining cost, and vouchers covered the fee for selected students. [1]

These are delivery costs and fees, not income or net savings. [1]

The delivery estimate included premises, hardware, staff, and a share of software development costs. [1]

The authors compared learning and cost with regular public schooling and with separate research on after-school tutoring. [1]

The pilot cost was reported per student. Forecasts for schools used a deployment budget that Educational Initiatives prepared in 2017. [1]

The study does not estimate financial returns or a payback period. [1]

Educational Initiatives operated the centers. Academic researchers evaluated the program with J-PAL support. [1]

Three centers served low-income neighborhoods in Delhi, India. Students were recruited near five public schools. [1]

Educational Initiatives developed the commercial product and operated the three centers. [1]

The centers needed computers, headphones, suitable rooms, and instructors. [1]

The research involved institutions in several countries. The study does not document a cross-border transfer of students’ data. [1]

Not assessed

Students need regular access to the software and instructional support. The Delhi program also depended on continued subsidy. [1]

A separate Rajasthan school trial reported gains of 0.22 standard deviations in math and 0.20 in Hindi after 18 months. [2]

The Delhi centers closed in March 2016. Later research evaluated a different school-based delivery model in Rajasthan. [1] [2]

The three Delhi centers closed in March 2016. Mindspark continued in other settings. [1]

Donors decided whether to continue subsidizing the centers. [1]

Donors required enrollment at or above 80% of capacity. [1]

Donors ended the subsidies, and all three centers closed. [1]

At least 80% enrollment at the subsidized fee. [1]

AIBD-0212-source-0001, AIBD-0212-source-0002 [1] [2]

The final Delhi study provides the results. Its official appendix explains the teaching rules. A later paper covers Rajasthan. [1] [2]

English. [1] [2]

The Delhi evaluation and its versions are one work. The Rajasthan trial used a different setting and cohort. [1] [2]

The Delhi trial and its official methods appendix provide the original evidence. The Rajasthan paper reports a separate trial. [1] [2]

The final author manuscript and official appendix were read. The Rajasthan review covered its abstract and introduction. [1] [2]

Free access to the combined program improved students’ measured learning. [1]

The sources are linked to the findings that they support. [1] [2]

A peer-reviewed randomized evaluation supports the learning gains. [1]

Not assessed

Students’ learning. [1]

Personal: 2 [1]

Students had measurable gains in math and Hindi. [1]

Observed in a randomized evaluation. [1]

Not assessed

Not assessed

Not assessed

Independently studied [1] [2]

A randomized evaluation of an after-school program. [1]

Not assessed

Education. [1]

Mindspark matched lessons to each student’s current level. [1]

The study ran from September 2015 to February 2016. The centers closed in March 2016. [1]

The program operated in three Delhi centers during the study. [1]

SHA-256: f4c15b98fbef…

Not assessed

Not assessed

Not assessed

6 revisions. Latest: 25 October 2026.