AIEd Knowledge BaseAIEd KB
On this page

Research Article

A Workshop Series for Effective Use of AI in Uncertain Times: Building a Physics Faculty Learning Community

published 2026-09-20 · added 2026-09-22 · updated 2026-09-22

Synthesis: Perl-Nussbaum and Finkelstein document one department's answer to Generative AI arriving before anyone was ready: a faculty learning community of six biweekly sessions in which physics faculty read their own local data, tested AI against real anonymized student homework in small groups, and built a shared resource repository rather than receiving a training package. The design follows their framework for institutional change in the AI era, treating AI as an arrival technology that bypasses the usual evidence base and repositioning the workshop leaders from experts to facilitators of collective inquiry. Sessions moved through course policy, classroom conversation, homework correction, AI-integrated tasks, and assessment, with students treated as partners whose reported use set the agenda.

Key Findings

  1. Nineteen faculty participated across six biweekly sessions. Meetings of 60 to 75 minutes drew about ten attendees each, in a large public R1 physics department with a history of systematic education reform.
  2. Local data opened every session. A campus survey found 80 percent of students using AI weekly and seeking instructor guidance, and a physics-specific survey issued in five courses reached roughly 350 students.
  3. Reported use reframed the problem as shared. Roughly 29 percent of surveyed students said they sometimes or often used AI to produce an entire solution, while most reported using it for supportive tasks when stuck — data that moved discussion from policing cheating to negotiating norms.
  4. Faculty testing exposed correction limits. AI solved every problem faculty tried, but its feedback on real student work was uneven, prompt-dependent, focused on surface features, and reliable only when given explicit goals and rubrics.
  5. AI-generated content invited shallow use. After working tasks that ask students to generate code, diagrams, simulations, or explanations, faculty concluded that students prompt shallowly and modify output superficially unless productive use is modeled.
  6. Oral defenses stood out as AI-resistant assessment. A junior-level Electricity and Magnetism course used short booked "mini homework defenses" in which a Learning Assistant randomly selects a completed problem and the student defends it for an immediate pass or fail; several faculty have since adopted the format.
  7. The series produced a living repository with five entries, covering syllabus policy statements, classroom discussion materials, the student survey, AI-integrated homework tasks, and assessment structures, and the department agreed to template policies across the spectrum from no use to full use.

Facilitating collective inquiry instead of delivering best practices

The structure is the argument. Rather than presenting evidence-based AI practices the facilitators cannot yet cite, each session began with local data or department-sourced materials, moved to small-group testing, and ended in collective discussion — the same sequence whether the topic was course policy, homework, or assessment. Session 1 illustrates the posture: with no campus policy in place, facilitators printed faculty-authored syllabus statements along a spectrum from no AI use to full use, and the group compared them and named shared themes. The change agent's role shifted from brokering settled practices to curating the room's own resources.

Leading with student use

The second session put student data in front of faculty before any prescription. Campus and physics-specific surveys showed students asking for disciplinary guidance and mostly using the tools when stuck rather than to complete graded work. That evidence changed both tone and practice: the department re-issued the survey in five courses, reaching roughly 350 students from middle to upper division, and used the results as a standing agenda item. Faculty rated their own engagement with students on a 0-to-3 scale from merely having a policy to soliciting student feedback, then named the support they would need to move up. Treating students as partners here was not a slogan but a data-collection decision that shaped every later session.

Assessment as the site of disruption

Faculty raised assessment concerns in almost every session, and the fifth addressed them directly, in formative and summative forms. The formative example was oral: short scheduled defenses of a randomly chosen homework problem, scored pass or fail in the moment, which faculty judged valuable, AI-resistant, and conducive to student reasoning, though labor-intensive. The summative example was the opposite: an external AI grading system for handwritten exams whose rubric-and-solution structure constrains how problems may be posed and how students may write — a trade-off the department has not settled. Faculty also reported the department's first case of a student using smart glasses during an exam, undercutting the appeal of simply reverting to supervised in-class testing.

What this means for practice

  • Faculty developers. Start a learning community with your own students' data and your own colleagues' materials; the session structure matters more than any content you could import.
  • Faculty developers. Anchor each session on a durable disciplinary practice — checking units, reflecting on a wrong solution — rather than on a tool or platform that will be obsolete before the series ends.
  • Administrators. Expect a repository, not a policy document, as the first concrete output, and fund the facilitation time that turns individual course fixes into shared departmental resources.
  • Instructors. Model productive AI use in class before assigning AI-integrated tasks, because students in this department tended to prompt shallowly and modify generated content superficially on their own.
  • Instructors. Consider a short oral defense of one randomly selected homework problem if you need an assessment that a chatbot cannot complete for the student.

Limitations

  • The report covers one physics department at a large public R1 institution, with nineteen participants across the series and about ten at any meeting.
  • The authors present no validated curriculum and make no claims about changes in student learning, framing the series as one instantiation rather than a model to replicate.
  • The end-of-series evidence is a self-report reflection survey; faculty most valued the collective discussion, while few reported concrete changes to teaching or policy.
  • Ethics of AI use and AI in the instructional laboratories were identified in the final session as areas the series had not yet addressed.

Connected Concepts

Connected Articles

Citation

Perl-Nussbaum, D., & Finkelstein, N. D. (2026). A Workshop Series for Effective Use of AI in Uncertain Times: Building a Physics Faculty Learning Community. arXiv:2609.23887.

Connected FAQs

Metadata

Study designCase Study
DisciplinePhysics Education
LevelHigher Ed
Intended audienceFaculty Developers, Administrators, Instructors
Learning and instructionPedagogical Partnerships, Self-Directed Learning
Technologies and techniquesGenerative AI
Assessment and measurementAuthentic Assessment, Formative Assessment, Summative Assessment
Institutions and policyChange Management, Educational AI Policy, AI Governance
Equity, ethics, and responsible useAI Use and Disclosure Statements, Equity
Foundations of AI in educationEducational Development, Teacher AI Competency, Academic Integrity
Page kindFramework
SourceHistoryEditChat

Embed this page

Copy the code below to embed a chromeless version of this page in a learning management system or other website. The embedded view hides the site header, navigation, and footer.