I feel this is important, because the vast majority of attention to date has been on the harmful effects of how students use AI themselves, to game the system or even outright cheat.

From the paper:

Teachers are rapidly adopting generative AI, yet there is little causal evidence on whether AI-assisted teaching improves or harms student outcomes. We conducted a randomized field experiment across a chain of middle and high schools in Turkey, assigning teachers to receive a generative AI teaching support tool or business-as-usual control. The majority of teachers used AI for lecture material, syllabus, and exercise creation.

AI can be an effective support tool for a teacher. Old way = 30 students, with teacher at the front explaining/showing/etc. Occasional questions/responses. Some are shy, some not following, many baffled, most silent. New way, 30 students each in their own alcove with dual 20 inch monitors and head phones with the same recorded explanation with AI all on their own time track so the AI can monitor responses and assess progress. AI has the smarts to query points to assess progressm and to present an alternate approach one or more times if needed, and escalate to a supervisory teacher/assistant as needed. Each student can progress at their pace, fast/slow. Poor students can be spotted and get added attention and faster ones can progress more quickly. The big problem today is lack of teacher one on one time, so many just languish, Some form study groups, others are auto-didacts and the book will do and too many drop out. The teaching community has a chance to embrace well implemented AI in the classroom. I hope they do better this time than they did with computers when they first entered classrooms in the late 80's. I recall in the early 80's when departments at U of T were not allowed to buy IBM and Apple PC's to perform modelling etc because the University had a large IBM system and we had to prepare programs, test them and them present them for the TSO system. (Time Shared Option) and the staff would run and return your batch/results the next AM. You would then assess data and fix it and run again(as there were frequent batch fails). Most of the grad students and post docs had access to IBM/Apple PC's and passed programs around, the early internet and sneakernet was in full flower. Eventually the UofT gave in to realty and soon every desk had a PC. Laptops -?? Not yet. We had luggables, Kaypro's and then Dells 25 to 40 poundsm CRT based 5.25 floppy, rare 5 Meg hard drives. High Schools had near zero computers, they slowly came in over 20 years. We had all manner of non standard ones, Commodore did bestm but the Gov of Ontario killed them by adopting their ICON system, https://en.wikipedia.org/wiki/ICON_(microcomputer) This was a form of graft for Canadian computer companies who would build and sell these and = $$$. Of course, it launched with fanfare, and fell from orbit ablaze all the way.