Test the teaching structure behind AI
Can a questioning partner help future teachers build stronger scientific arguments? An eight-week study compared 124 biology teacher candidates in two intact classes at an Uzbek university. One group used a GPT-4-based Socratic protocol that repeatedly challenged claims and requested counterevidence; the other held conventional peer discussions. Adjusted argumentation scores were 3.91 versus 3.21 on the study’s five-level rubric. The AI group also performed better on rebuttal and counterevidence tasks. The comparison supports the complete instructional package, not a conclusion that AI outperforms people. The control group lacked an equally structured challenge protocol, so additional deliberate practice could explain part of the difference.
Why it matters for leaders
Have professional-learning teams rehearse the same rebuttal routine with peers and with AI. Compare unaided written arguments using common scoring. Budget for facilitation and source checking; a subscription alone does not reproduce the instructional package examined here.
The evidence
The primary analysis adjusted for baseline argumentation. Thirty percent of submissions received blind double scoring; reported agreement was .87.
Who was studied
Third-year university teacher candidates; indirect evidence for K-12 pupils.
What the evidence cannot tell us
One institution and two classes; allocation descriptions are inconsistent. Unequal protocol monitoring, unverified control-group AI abstinence and short follow-up weaken causal interpretation.
Read the original sources
Summary and leadership analysis from the Week ending 2026-10-16 brief. The edition’s selection notes explain source access and evidence limitations.