Keeping yourself in the loop when working with agentic AI

0:00 Pick a problem of your own to work along with 1:02 What your Obsidian setup should look like 2:02 The eight steps, from importing data to the exported report 3:04 The two documents: the report and the appendix 4:04 Margaret Mitchell, model cards and data documentation 5:04 The paper: AI agents push humans out of the loop 6:06 Why staying in the loop gets harder with each generation 7:07 From a single score to more text than you can read 8:08 Agentic AI takes steps it never asks you about 9:08 A non-exhaustive list of what you have to check 10:09 Holding the plan and the details at the same time 11:09 Approval fatigue, and the two things built in against it 12:09 Strategic friction: hypothesis first, chart before test 13:10 The disclaimer before export, and what we cannot make you do 14:11 Exploratory analysis versus confirmatory analysis 15:11 The risks of exploring: data dredging and p-hacking 16:11 HARKing, and the garden of forking paths 17:12 Stopping rules, switched outcomes, and getting going Agentic AI for Statistics and Data Science, from the AI research group at Uppsala University. https://wordalisation.com/