STEM Faculty Perspectives on Generative AI in Higher Education

This paper presents findings from a focus group study of 29 STEM faculty at a US public university, revealing that while they recognize both pedagogical benefits and significant challenges regarding GenAI, its effective integration requires a comprehensive rethinking of assessment, pedagogy, and institutional governance.

The Gist

Imagine the classroom as a bustling kitchen. For decades, the chefs (teachers) have been cooking every meal from scratch, chopping vegetables, and seasoning sauces. Suddenly, a new, incredibly fast robot assistant (Generative AI) has been installed in the kitchen. It can chop, mix, and even plate a meal in seconds.

The students are already using this robot in their own home kitchens, and now they are bringing their robot-made dishes to the class. The professors at San Francisco State University (specifically in Science, Technology, Engineering, and Math) sat down to figure out: What do we do with this robot? Do we ban it, embrace it, or try to learn how to cook alongside it?

Here is a simple breakdown of their findings, using everyday analogies.

1. The Robot is Already in the Kitchen

The study found that the robot isn't just a future idea; it's already here. Most of the professors (29 of them) are already using it.

• How they use it: They aren't letting the robot take over the whole kitchen. Instead, they use it like a super-fast sous-chef.
  • They ask the robot to help write quiz questions (like asking it to chop onions quickly).
  • They use it to create practice problems or translate complex ideas into simpler language.
  • They even use it to draft emails to students, though they are careful to add their own "human touch" so the message doesn't feel like it came from a machine.

2. The "Illusion of Competence" (The Magic Trick)

This is the biggest worry for the teachers.

• The Problem: When students use the robot, they hand in their homework faster and more often. It looks like they are doing great! But the professors noticed a trick.
• The Analogy: Imagine a student asks the robot to build a Lego castle. The robot builds a beautiful castle in 10 seconds. The student hands it in. The teacher thinks, "Wow, great castle!" But if the teacher asks the student, "How does this tower stay up?" or "Can you build a different roof?", the student freezes. They don't know how the castle was built because they didn't do the building.
• The Result: The robot is masking the fact that students might not actually understand the math or the code. They are getting the answer, but they aren't learning the skill.

3. The New Job for Teachers: From "Creator" to "Editor"

The professors realized that using the robot didn't make their jobs easier; it just changed the job description.

• Before: Teachers spent hours writing every single quiz question and grading every paper from scratch.
• Now: The robot writes the draft, but the teacher has to become a strict editor. They have to read the robot's work, check if it's lying (robots sometimes "hallucinate" and make up facts), fix the errors, and make sure it's accurate.
• The Takeaway: The robot didn't eliminate the work; it just shifted the work from "making" to "verifying."

4. The "Exam Hall" vs. The "Open Book"

Because the robot is so good at writing essays and solving problems, teachers are scrambling to figure out how to test students fairly.

• The Old Way: "Don't use your phone, don't use AI, just write this essay on paper." (This is becoming hard to enforce).
• The New Strategy: Teachers are trying two things at once:
  1. The "No-Robot" Zone: They are bringing back old-school oral exams or in-class writing where students have to explain their thinking without help. It's like asking a chef to cook a dish in front of you to prove they know the recipe.
  2. The "Robot Partner" Zone: They are giving assignments where students must use the robot, but then have to critique it. For example: "Ask the AI to write a code, find three mistakes in it, and explain why they are wrong." This forces the student to be the boss, not the passenger.

5. What the Professors Need from the School

The teachers aren't asking for the robot to be taken away; they are asking for better tools to manage it.

• Training: They want workshops, not just on how to use the robot, but on how to think about it. They need to know how to ask the robot the right questions (prompt engineering) and how to spot when it's lying.
• Clear Rules: Right now, the rules are messy. In one class, using the robot is fine; in the next, it's cheating. The professors want a clear "menu" of rules so students aren't confused.
• Time: They need paid time to redesign their courses. You can't just swap a human teacher for a robot overnight; you have to rebuild the whole curriculum, and that takes time and money.

The Bottom Line

The paper concludes that Generative AI is like a powerful new engine in a car. You can't just ignore it, and you can't just let it drive the car by itself.

The professors are realizing that to get the best out of this technology, schools need to stop asking, "How do we stop students from cheating?" and start asking, "How do we teach students to drive this car safely, so they don't crash when they get a real job?"

It's not about banning the robot; it's about teaching the human how to be the captain of the ship, with the robot as a very helpful, but sometimes unreliable, crew member.

Technical Summary

Here is a detailed technical summary of the paper "STEM Faculty Perspectives on Generative AI in Higher Education" by Akila de Silva et al.

1. Problem Statement

The rapid proliferation of Generative Artificial Intelligence (GenAI) in higher education has created a reactive environment where adoption is largely student-driven, forcing instructors to adapt to technologies already embedded in classroom practices. While some faculty embrace GenAI for pedagogical innovation (e.g., content generation, assessment support), others express significant skepticism regarding academic integrity, the erosion of critical thinking skills, and the reliability of AI outputs (hallucinations).

There is a critical gap in understanding how STEM faculty specifically perceive, integrate, and negotiate the role of GenAI. Existing literature lacks comprehensive insights into the collective experiences of STEM educators at access-oriented, public institutions serving diverse and underrepresented populations. Without this understanding, institutions cannot develop effective policies, pedagogical strategies, or governance frameworks to balance innovation with educational integrity.

2. Methodology

The study employed a qualitative focus group design to investigate faculty perspectives.

• Participants: 29 faculty members from the College of Science and Engineering (CoSE) at San Francisco State University (SFSU), a large public, Hispanic-serving institution. The cohort included lecturers and tenure/tenure-track faculty from Computer Science (41.9%), Engineering, Psychology, Mathematics, Chemistry, Environmental Science, Physics, and Biology.
• Data Collection:
  • Seven Focus Group Sessions: Conducted remotely via Zoom, lasting approximately 90 minutes each.
  • Structure: Three sessions were discipline-specific (Computer Science only), and four were interdisciplinary.
  • Protocol: Semi-structured discussions guided by three Research Questions (RQs), preceded by a demographic questionnaire.
  • Analysis: Sessions were recorded and transcribed. Data was anonymized and processed using Google NotebookLM for exploratory thematic analysis. Researchers manually verified AI-generated themes against original transcripts and notes to ensure accuracy.
• Research Questions:
  1. How do faculty integrate GenAI into course design and learning activities?
  2. What benefits and challenges do instructors observe regarding student learning?
  3. What institutional resources and policies are needed to support effective GenAI adoption?

3. Key Contributions

The paper makes three primary contributions to the field of educational technology and higher education policy:

1. Contextual Insights: Provides empirical data on faculty perspectives from a large, access-oriented, Hispanic-serving public university, offering a nuanced view distinct from elite or private institutions.
2. Pedagogical Taxonomy: Identifies specific pedagogical benefits and challenges, moving beyond general speculation to concrete observations of how GenAI alters the instructional labor and student learning dynamics in STEM.
3. Policy Recommendations: Proposes faculty-informed recommendations for institutional support, emphasizing the need for a balance between university-wide guardrails and departmental autonomy.

4. Results and Findings

RQ1: Integration into Course Design and Activities

Faculty usage (93% of participants) falls into three categories:

• Course Design & Preparation: Used for generating quizzes, creating real-world scenarios, refining assignment rubrics, and improving instruction clarity. However, this shifts labor from creation to curation; faculty spend significant time verifying, refining, and formatting AI-generated content to meet course standards.
• Direct Student Support: Students use GenAI for brainstorming, code generation (e.g., translating pseudocode to Python), and debugging. Faculty observe a "dual strategy" in assignments: some require students to compare human vs. AI solutions, while others restrict AI use to in-class exams.
• Administrative Tasks: Used for summarizing email threads and drafting responses. A key concern is the potential erosion of trust if communication feels overly automated.

RQ2: Benefits and Challenges for Student Learning

• Benefits:
  • Bridging Technical Gaps: Helps students overcome coding barriers, leading to higher assignment completion rates.
  • Efficiency: Accelerates idea generation and project iteration.
  • Accessibility: Acts as a "24/7 private teaching assistant," particularly beneficial for students balancing work/family responsibilities.
• Challenges:
  • The "Illusion of Competency": High submission rates often mask a lack of conceptual understanding. Students struggle to debug AI-generated code because they did not construct the underlying logic.
  • Prompting Burden: Students spend excessive time crafting prompts rather than learning core concepts.
  • Academic Integrity: Detection tools are unreliable. Faculty are increasingly abandoning "AI-proofing" assignments in favor of traditional assessments (oral exams, pen-and-paper tests) to ensure genuine learning.

RQ3: Institutional Resources and Policies

Faculty identified three critical areas for institutional support:

• Training & Resources: Need for professional development on LLM mechanics, prompt engineering, and task-specific applications (e.g., rubric creation). A centralized repository for shared prompts and case studies is requested.
• Policy & Guidelines: Current policies are inconsistent across departments, causing student confusion. Faculty call for clear, department-level guidelines on acceptable use, citation standards, and academic honesty, rather than rigid, top-down mandates.
• Curricular Change: Proposals include a mandatory university-wide AI literacy course (covering ethics and usage) and a restructuring of major requirements to account for AI's impact on foundational skills.

5. Significance and Implications

The study concludes that the integration of GenAI in higher education is not merely a technical adoption issue but a fundamental shift in pedagogy, assessment, and governance.

• Shift in Labor: GenAI transforms the faculty role from content creator to expert curator, requiring new skills in verification and refinement.
• Assessment Crisis: Traditional assessment methods are failing to detect learning gaps created by AI reliance. There is an urgent need to redesign evaluation to prioritize critical thinking and oral defense of work.
• Governance Balance: Effective policy requires a hybrid approach: university-wide ethical guardrails to ensure consistency, paired with departmental autonomy to allow discipline-specific adaptation.
• Future Outlook: The paper warns against premature structural changes, advocating for iterative, evidence-based policy evolution as the technology and its impact on the workforce continue to unfold.

This research underscores that successful GenAI integration requires sustained investment in faculty development, a rethinking of what constitutes "learning" in an AI-saturated environment, and a collaborative approach to institutional governance.