education performance gap
Standard deviations · business problem solving task
Loading chartTask-performance difference between education groups
AI in practice
What changed, in which study?
Education gaps with AI assistance
1,174 adults aged 25–45 · Randomized online experiment · Working paper
Loading chartEducation-group performance gaps on one business problem-solving task, not national wages or long-run productivity. Follow-up gains are not assigned numerical values from the abstract.
Writing tasks
−40%task time
+18%rated quality
Customer support
+15%issues resolved / hour
Experienced developers
+19%task time
Different tasks and historical tools; these effects are not pooled or estimates of whole-job productivity.
Diagnostic reasoning
Median rubric score · %Loading chart50 physicians · 2023 cases · not statistically significant (adjusted +2 pp, 95% CI −4 to +8). Vignette scores, not patient outcomes.
Learning with an AI tutor
Median post-test score · pointsLoading chart194 eligible students · two physics lessons · 2023. Median immediate test scores; repeated observations, not long-term retention.
Data-centre electricity demand
TWh per year · all data centresLoading chart2025: estimate · 2030: projection. AI facilities are a subset; this is not AI-only electricity use.
