Stanford AI Professor Returns to Written Exams After Students Demand Human Assessment

Stanford Professor Returns to Paper Exams Amid AI Disruptions

Jure Leskovec, a Stanford computer science professor and machine-learning expert, faced an unexpected challenge when artificial intelligence started reshaping education. After nearly 20 years of teaching and deep involvement in AI research, he witnessed a significant shift sparked by the release of GPT-3 and the rise of large language models.

Stanford's computer science program often provides a glimpse into emerging technological trends, but the sudden AI breakthrough triggered an existential moment among students. They questioned their future roles in a world where AI might soon handle research and problem-solving. This uncertainty led to extensive discussions at the PhD level about purpose and adaptation.

Students Push for Traditional Testing

Surprisingly, the call for change came from the students themselves, particularly teaching assistants who had recently been undergraduates. Their proposal was straightforward: return to paper exams. This move was a direct response to the challenges AI posed to conventional take-home, open-book assessments.

Leskovec’s prior testing approach allowed students to use textbooks and online resources but prohibited sharing code and solutions. With AI tools like GPT-3 and GPT-4 becoming widely accessible, the integrity and effectiveness of these exams were questioned. The paper exam, while more labor-intensive to grade, was seen as the best way to accurately assess individual understanding.

AI as Both a Tool and a Challenge

Leskovec, known for his work with graph data and AI in biology, admits this shift increased the workload for him and his teaching staff. Grading hundreds of paper exams is time-consuming, but it ensures students demonstrate genuine knowledge rather than AI-assisted answers. Despite the irony of using more paper in a digital age, the move underscores the current limitations of AI as a reliable standalone evaluator.

He firmly rejects using AI to help grade, emphasizing the importance of human judgment in evaluating student work. This approach contrasts with some institutions that have banned AI outright or experimented with oral exams to counter cheating concerns.

Balancing AI and Human Skills in Education and Work

Leskovec draws a parallel between AI and calculators, suggesting both are powerful tools that change how we test skills. Just as math exams differ depending on calculator policies, AI requires educators to rethink assessment methods. The goal is to test both the ability to use AI tools and independent critical thinking.

This dilemma reflects broader questions about workforce skills. What counts as a human skill versus an AI skill? Experts from MIT and Google describe AI tools as straddling automation and collaboration — they automate tasks but also need human engagement to be effective.

The Workforce’s Mixed Signal on AI

While AI adoption faces hurdles — with studies showing many generative AI projects fail and entry-level hiring slowing — demand for AI-related freelance skills is surging. Platforms like Upwork report a 40% increase in AI and machine learning job postings, especially for contract work in creative fields, writing, and fact-checking AI outputs.

Fact-checking has become crucial as AI-generated content can be inaccurate. According to Upwork’s research, human oversight remains essential to validate AI outputs, highlighting the value of domain expertise.

Kelly Monahan from Upwork notes that human skills are gaining premium status because AI "hallucinates" too often to fully replace human involvement. Legal and specialized fields see rising demand for experts to verify AI-generated work and prevent errors.

Reskilling: The Path Forward

Leskovec agrees that reskilling is vital. As AI transforms industries, human expertise becomes more important, yet younger workers struggle to gain the necessary domain knowledge. This gap points to a need for revamped education and corporate training programs focused on effective AI collaboration.

He stresses the importance of investing in young talent and structured training. Without it, organizations risk losing the pipeline of skilled workers who can work alongside AI.

Early Days in Adapting to AI

Reflecting on the current state of AI integration in education and work, Leskovec concludes that we are still in the "coming-up-with-solutions" phase. The return to hand-graded paper exams is one such solution, emphasizing the human role in knowledge verification amidst AI’s rise.

For professionals and educators looking to build skills in AI and adapt to this evolving landscape, exploring updated courses can be beneficial. Resources like Complete AI Training’s latest courses offer practical paths to understanding AI tools and their applications.