When AI Can Do Everything, What Is Left to Learn?

Why It Matters

Employers increasingly expect graduates to supervise AI‑generated work, so curricula must teach problem framing and critical interpretation rather than just artifact creation. This shift ensures students graduate with skills that remain valuable in an AI‑augmented workplace.

Key Takeaways

• AI can produce artifacts, but not verify their correctness.
• New curricula emphasize problem framing before artifact creation.
• Students must interpret and critique AI-generated models, queries, dashboards.
• Artifact reasoning combines defining questions and critical interpretation of outputs.
• Boston University redesign shows increased engagement and deeper learning outcomes.

Pulse Analysis

The rise of generative AI has forced a fundamental reassessment of how higher‑education programs measure competence. Traditional assignments—writing essays, coding programs, building data models—once served as reliable proxies for underlying cognitive skills because producing the artifact required mastering the process. Today, AI can generate polished outputs with minimal human effort, decoupling artifact creation from true understanding. Educators therefore need to pivot from assessing the product to evaluating the reasoning that guides AI, ensuring students develop the mental models necessary to verify, adjust, and trust machine‑generated results.

At Boston University’s Questrom School of Business, Professor Dellarocas applied this insight to a core database course. Instead of asking students to simply draw an ER diagram, the class first defines what business reality the model must capture—deciding which entities, relationships, and attributes matter. Similar reframing occurs for SQL queries and dashboard design: students articulate the exact business question, clarify definitions like "revenue" or "last month," and then critique the AI‑produced solution. This structure turns AI into a collaborative tool rather than a shortcut, and classroom discussions have become more animated as students grapple with assumptions and edge cases.

The implications extend beyond data science. Writing programs, designing graphics, and even composing music now require a second‑order skill set: the ability to frame problems, interpret outputs, and identify hidden biases. Institutions that cling to artifact‑only assessments risk graduating graduates ill‑prepared for workplaces where AI handles routine production. By embedding artifact reasoning into learning outcomes—through problem‑definition exercises, critical debriefs, and iterative evaluation—schools can ensure students retain the deep expertise needed to guide intelligent systems responsibly, preserving both academic rigor and future employability.