Impromptu: Amplifying Our Humanity Through AI would the two Wes Moores’ lives have changed if they had swapped schools? If we want our schools to lift up the children of low-income parents, we know that technology alone is not enough. More than a few ed tech companies have commissioned randomized, controlled trials only to learn that their products did not make any difference. We also know that money alone won’t do the trick. America is one of several nations that have substantially increased public school spending over decades without seeing gains for the chil- dren of low-income parents. According to the National Center for Education Statistics, total expenditures for public elemen- tary and secondary schools in the United States were $800 billion in the 2018–19 school year: per pupil, about a third more than the rich-world average, double the real dollars we spent in 1980, and four times the real dollars per pupil in 1950. So, can anything really improve schools at scale, and if so, what would it look like? I asked GPT-4, but by the time I provided enough context for GPT-4 to properly engage, the mathemat- ical engines got overwhelmed and GPT-4’s answers became drenched in hallucination and incoherence (a phenomenon we see in early integration of GPT AI with edge-case search engine users). But human experts (such as Andreas Schleicher at the Organ- isation for Economic Co-operation and Development and Sir Michael Barber at the University of Exeter), who have worked with literally dozens of national school systems to determine what improves results at scale for poor children, say the fast- est-improving school systems use technology, along with all 36