Teaching Practical Skills in the Age of AI

By Erez Buchweitz, Statistics

Teaching Effectiveness Award Essay, 2026  

When I began my role as a GSI for my department’s statistical machine learning class, I examined the labs’ curriculum and noticed a disconnect with our field’s rapidly changing reality. The existing labs were built around step-by-step coding walkthroughs, focusing in large part on technical skills that AI tools are increasingly able to handle for us. As we prepared for the semester, it became clear to the instructor and me that the current syllabus was not optimally designed to bring added value to the students. Drawing on my previous work experience as an industry data scientist before I started my PhD at UC Berkeley, I was motivated to do the best service I could in preparing them for their future careers, be it in academia or industry. From a teacher’s perspective, I believe AI is both a blessing and a curse. It’s becoming unnecessary, and increasingly impossible, to teach students things that an AI can do for them, and the temptation to use AI tools to avoid making an effort is great. On the other hand, AI can be used to ramp up the learning curve and remove obstacles, thus enabling the class to focus on the important things. Recognizing both the upsides and downsides of this new reality, the instructor and I agreed that the labs needed redesigning from scratch, and I took the task upon myself.

The new lab design focused on multi-week mini-projects that required the students to research and develop a solution to a problem from end to end. The idea was that instead of testing them on small subparts of the problem, one at a time and without context, I wanted to test them in the same scenario in which they would be tested in practice. Instead of me lecturing, they spent the majority of their time directly practicing applying their skills, experiencing the hardships and challenges in a safe and managed environment where a supervisor could guide them on how to overcome them. To complement the difficulty of being thrown into the deep end, the mini-projects were structured so that the level of difficulty rose from one to the next, I let the students work in groups, and the groups received a lot of hands-on guidance from me during labs. The new lab design benefited from both the up and down sides of AI – I let students use AI for their project work, which helped them bridge technical gaps faster and aligned them with industry practice. On the other hand, the mini-projects were designed to be at a level where AI could not execute the project from start to finish. Consequently, the soft skills of a data scientist, which AI tools do not currently possess, became the focus of the labs. Statistical thinking and how to apply it to real problems became much more emphasized.

Since the lab syllabus was new, the instructor and I felt a need to closely monitor its implementation. For this, we initiated a mid-semester course evaluation administered through Gradescope, which included the usual overall questions about lab effectiveness alongside specific questions about the new lab structure. The reviews were overwhelmingly positive, with some students noting that “this class actually teaches us skills that we need.” The reviews provided valuable insight and allowed us to adjust the projects and guidance styles for the second half of the semester.