Learning and technology, read against the evidence
Each Tuesday, one claim about how people learn with technology, from a vendor, a headline or a conference stage. I find the research behind it, read the studies themselves, and say what the evidence supports, what it does not show, and how sure I am. Every source is linked, and every citation is checked by script before the piece goes up.
Twice the learning
What one Harvard physics trial shows about AI tutors, and where the claim stops
A 2024 story asked whether AI tutors might double learning. The trial behind it is careful and real. Set beside two other trials and a vendor report, it supports a narrower claim than the one that travelled.
- The claim
- AI tutoring doubles learning compared with a classroom
- Where it started
- Harvard Gazette, September 5, 2024, before the paper was published
- Evidence base
- Three randomized studies, three tutoring reviews, one re-reading of the two-sigma claim, one independent re-analysis of the Harvard data and one vendor report
- Confidence
- Moderate for the narrow finding, low for the headline
Read the review
Christopher Ross · 18 min read · 11 sources
The five questions
- 1What exactly is claimed, and by whom
- 2What the evidence actually is
- 3What it supports and what it does not
- 4How sure I am, and what would change that
- 5What a practitioner should do with it
Questions about sourcing or how a piece is checked? How these reviews are made, including how corrections work →