USC Engineers Rework Courses After Faculty Could Not Tell AI Proofs From Student Work

A faculty meeting that changed the plan

The shift began with Stephen Tu, a former researcher at Google DeepMind who teaches the graduate machine learning course EE 660. He had been testing how well AI models could solve course assignments, some of which required advanced, multipage mathematical proofs. At an ECE faculty meeting, he showed colleagues two solutions side by side and asked: "Can you tell which of these proofs was done by AI, and which by a student?" According to the department, no one could.

Richard Leahy, chair of ECE, called it "a wake-up call for our entire faculty." After months of brown-bag lunches and hallway discussions, the faculty concluded that the department needed to rethink not only how students demonstrate learning, but also what and how it teaches.

The shared goal, as described in the article, is to help students develop critical thinking skills while learning to use AI tools effectively. Michelle Povinelli, vice chair of ECE, said teaching should be grounded in "rigorous critical thinking, while empowering students to leverage new tools." Leahy said industry partners, including alumni, an advisory board and research partners, report that AI literacy is becoming a standard part of the engineering workflow.

AI in assignments, and new ways to assess

Faculty are now building AI code generation and prompt engineering directly into assignments. They stress that engineers remain responsible for critical judgment. Povinelli said deciding what is worth designing, how to go about it and how to assess the impact are "fundamentally human tasks." Faculty also agree that students need a solid grasp of core concepts so they can ask the right questions, spot mistakes and validate results.

In EE 250, Distributed Systems and the Internet of Things, Professor Bhaskar Krishnamachari teaches students to use prompt engineering with AI models to generate functional Python applications. In his own lab, he used AI to port course software when moving from Raspberry Pi to ESP32 microcontrollers.

Because standard coursework can be easily completed with AI, many faculty are putting more weight on in-person lab work, live data collection, physical measurements and interactive problem solving, where instructors can observe and coach students directly. Assessment is also moving toward in-person demonstrations of skill:

• In the freshman course EE109, Introduction to Embedded Systems, Professor Mark Redekopp has students complete short oral quizzes after each lab, plus hand-coding exercises that require verbal explanations.
• Krishnamachari has students record short videos on class concepts, which mirrors real-world interview formats. AI-driven analysis of the submissions helps him find collective knowledge gaps and adjust later lectures.

Redekopp, associate chair for undergraduate programs, also convened a strategy workshop for undergraduate core faculty on adapting teaching and assessment. He notes that oral assessments double as career preparation, since they build practice in articulating technical reasoning.

A redesigned first-year course

Professors Rehan Kapadia and Yasser Khan led a major redesign of EE 105, Introduction to Electrical Engineering, one of USC's flagship first-year ECE courses. The new version introduces students to the field through hands-on, project-based learning, much of it centered on AI. A custom knowledge graph guides students through course concepts and helps them build the foundation for practical experimentation.

Kapadia said AI tools have made his PhD students far more productive and allowed his group to take on research questions that were previously impossible. He added that undergraduates can now take on more complex projects with cutting-edge hardware than they could five or 10 years ago. He directs USC's John O'Brien Nanofabrication Lab and the MOSIS 2.0 Service for rapid chip design, fabrication and testing. Khan contributes expertise in additive manufacturing and skin-like wearable and implantable devices.

The department also uses AI tools, from 3D visualization to video generation, to deliver course content in richer formats. Tu, whose experiment started the discussion, now holds dedicated class discussions on the latest AI breakthroughs, from formal mathematical proofs to neural networks in autonomous drone flight.

What it means for students

According to the article, the department's approach is to embed AI in the learning process rather than treat it as a shortcut used outside the classroom. For students, that means more graded work done in person or explained aloud, alongside explicit training in prompting and checking AI output. Povinelli said students have used AI tools to build initial prototypes in days that might previously have taken a whole semester. The department says it wants graduates ready for workplaces where AI literacy is expected.