A lot of software developers are struggling with how our skills will remain relevant in a world where most code is written by AI. So I was interested in this experiment by statistics professor Sharon Lohr. She gave three old assignments to Google Gemini and assessed the results. The AI excelled at well defined tasks, but lacked imagination and creativity. Lohr advocates using AI tools so students can focus on analysis, much like computers allowed statisticians to develop more complex models:
AI tools can free students in an introductory statistics course for non-majors from tedious calculations, so they can spend more time learning to be thoughtful consumers of statistics. Many introductory statistics textbooks still have pages and pages of exercises asking students to calculate five-number summaries or t confidence intervals for a small set of numbers. Would it not be better to have students learn how to use an AI system to do these routine tasks, and instead develop skills on how to interpret statistics and judge their quality?
What AI systems cannot replace (at least right now) is critical statistical thinking. Just as the use of electronic computers in the 1960s freed statisticians to use more complex models, handle larger datasets, and develop new methods for extracting information from the data, AI systems can free statisticians from routine programming and facilitate the development of new statistical tools. But would an AI system have been able to invent the jackknife and bootstrap?