AI may speed up one step while making the whole job harder to manage. The paper proposes a broader test of the full job. Does AI leave people in control and able to recover when things go wrong?
Utah tested 85 school apps with standard privacy agreements. GovTech reports that 52% collected at least one type of data their agreements did not allow. That mismatch did not establish whether vendors broke privacy laws.
In 5,115 sessions with final reports and logs, coding AI systems mentioned roughly one in eleven actions they took. The study measured executed actions missing from reports. A final report cannot replace the saved log.
LAUSD blocked student access to generative AI on district devices and accounts while it writes rules. Teachers kept access and training. Outside those controls, a student generally needs both a personal device and a non-district account.
The tools, prompts, and rules around a coding model form its harness. This study found no clear overall winner between the tested setups. Correct code, a finished run, and cost were separate checks.
A study followed 18 adults in the United States who used chatbots. Help with feelings often grew out of everyday help when other support was missing. Their accounts do not prove that chatbots cause loneliness.
