It’s Like ‘X’
How Engineering Faculty Metaphors Construct (and Constrain) AI Understanding in Engineering Education
IDEEAS Lab
2026-06-23
Project Team
Dr. Andrew Katz
Principal Investigator
Mitch Gerhardt
PhD Candidate
Kylee Shiekh
PhD Student
Benjamin Chaback
PhD Student
CAREER: Minds and Machines: Exploring Engineering Faculty Member Mental Models of Generative AI and Instructional Decisions Supported by NSF Award #2339702
Agenda
- The Ada Mandate
- Our Study
- Code-Switching
- Learn it, Use it, Teach it
- Q&A
The Ada Mandate
1975: A Defense Software Crisis
Carl H. Pforzheimer Collection of Shelley and His Circle, The New York Public Library (1835). Watercolor portrait of Ada Byron.
Problem (1970s): Hundreds of programming languages across DoD = Unshareable code, ballooning costs, and challenging updates
Fix (1980–1987): Commission and mandate one rigid language (Ada) as the common language for defense software (DoDD 3405.1)
Oops: Adoption lagged, waivers expanded, and C was never approved. The mandate was rescinded 1997 by Assistant Secretary of Defense for Command, Control, Communications, and Intelligence (C3I) Emmett Paige, Jr.
The mandate flopped, but Ada found some niches
Whitaker, W. A. (1993). Ada—the project: The DoD high order language working group. The Second ACM SIGPLAN Conference on History of Programming Languages, HOPL-II, 299–331. https://doi.org/10.1145/154766.155376
Why the DoD?
What was the error? Mistaking a heterogeneous domain for a homogeneous one, then trying to solve the problem with a single solution
Today: Reaching for one definition of “AI assistance,” “AI literacy,” or “prohibiting AI” as if an instructor, a data science TA, and a graduate student mean the same thing by “AI”
If we standardize the language of AI top-down, our faculty development will fail like Ada
Our Study
Study Details
- Semi-structured interviews with engineering instructors
- Part of a larger NSF CAREER project (VT IRB #24-572)
- Explored how STEM engineering instructors think about GAI for teaching, learning, and assessment
169
Instructors
18
Universities
Semi-structured interviews
Informed by mental models (Genter & Stevens, 1983) and the Theory of Planned Behavior (Ajzen, 1991), we designed a protocol to elicit instructor reasoning around GAI in teaching, assessment, and learning contexts
Gentner, D., & Stevens, A. L. (Eds.). (1983). Mental Models. Psychology Press | Ajzen, I. (1991). The theory of planned behavior. Organizational Behavior and Human Decision Processes, Theories of Cognitive Self-Regulation | Galletta, A., & Cross, W. E. (2013). Mastering the semi-structured interview and beyond: From research design to analysis and publication
Figurative Language Subset
- Metaphors structure understanding (Lakoff & Johnson, 1980)
- How instructors describe GAI shapes how students understand it and how instructors approach it in their teaching
Question: “If you had to think of an analogy or metaphor to describe GAI to someone, what would you say?”
3 rounds of coding of responses based on initial research-derived taxonomy of GAI metaphors, then iteratively refined through member-checking and language synthesis
57 Instructors
7 Universities
17 Disciplines
Spring 2025
Instructors in the same department, same semester, under the same policies constructed different models of what GAI is and does
Lakoff, G., & Johnson, M. (1980). Metaphors We Live By (W. a new Afterword, Ed.). University of Chicago Press. | Schmitt, R. (2005). Systematic Metaphor Analysis as a Method of Qualitative Research. The Qualitative Report, 10(2), 358–394. https://doi.org/10.46743/2160-3715/2005.1854 | Nicmanis, M. (2024). Reflexive Content Analysis: An Approach to Qualitative Data Analysis, Reduction, and Description. International Journal of Qualitative Methods, 23, 16094069241236603. https://doi.org/10.1177/16094069241236603
Results - Dimensions of AI Understanding
Example Quotes
Ontology: What is GAI?
Technical Object/Artifact: “It’s a program that’s going to get your input and based on pre-learned information, output the data.”
Social Being/Agent: “[GAI is] a person with deep knowledge who thinks far faster than I do.”
Environmental Force/Systemic Phenomenon: GAI a “paradigm shift” that does not “mean the field of greater science is going to completely go away.”
Power/Capability: What can GAI do?
Limited/Deficient: “They’re text predictors. It’s just a function, you give it some input, it does some calculation and gives you an output. It’s not a mind, it doesn’t think, and I have to warn my students [that] they can’t solve your logic problems. They don’t know how to reason, they just predict text.”
Human-Level/Comparable: “One time I used it as a therapist and it’s actually really, really good. I can be like, ‘No, that’s terrible, I don’t want to hear that,’ [but] you would never say that to a therapist. It was very strangely affirming.”
Superhuman/Superior: “If you have to create a literature review, there’s only X number of papers that you can read over the next three days. And you [will] only have a 3% or 10% of that access of information in that time, where [GAI] might be able to cover 80% more because they have more access to a lot more data so they can provide you with a summary with a lot more information and content.”
Code-Switching
Code-Switching as Interpretive Work
Downey & Lucena (2004): Engineers routinely code-switch by calibrating vocabulary, framing, and emphasis to build legitimacy simultaneously in professional, popular, and global spheres
Adapted for faculty development: Multiple meaning-making systems (e.g., student, faculty, and administrator), with their own grammar for “AI”
Code-switching is how engineers (and faculty developers) build credibility across communities without abandoning any of them
Downey, G. L., & Lucena, J. C. (2004). Knowledge and Professional Identity in Engineering. History and Technology, 20(4), 393–420. https://doi.org/10.1080/0734151042000304358 | Bucciarelli, L. L. (2003). Engineering philosophy. DUP Satellite.
Returning to the Ada Mandate
Ada’s Lesson: Imposing one solution on a heterogeneous domain compromises the solution
Our Study: Instructors’ meaning-systems are not homogeneous: they hold all stances even within the same department
The Pitch: Faculty developers must move fluently among existing codes, meeting colleagues where they operate, while building similar capacities in others
The Taxonomy as Phrasebook
Not: A dictionary to enforce one correct meaning of “AI” for all faculty
Not: A rubric to grade faculty understanding about “AI” and sort them into “good” and “bad” buckets
Is: A diagnostic for locating where colleagues stand, so you can meet them where they are and build from there
Hear “it just predicts text”
→ Locate: Mechanical/Pattern-Matching + Technical Object
Hear “I use it like a personal assistant”
→ Locate: Collaborative/Partnership
Hear “It is breaking our students and higher education”
→ Locate: Adversarial/Threat + Environmental Force/Systemic
Learn it, Use it, Teach it
Learn it, Use it, Teach it
Learn it: Use the five dimensions to surface your own default codes and recognize others’
The goal is not to “correct” your colleagues’ mental models, but to understand them and build from there
Use it: Design programming that targets or meets the Social-Being optimist and the Technical-Object skeptic where they stand
Social-Being frame: “Your AI collaborator has a communication style and biases, so here are ways to work with it.”
Technical-Object frame: “Let’s look at the model specs your tool uses because that determines what it can do and how it behaves.”
Teach it: Equip faculty to code-switch with their students and across departments
The goal is fluency across perspectives
So, how do you describe GAI?
Questions?
Appendix: Initial Taxonomy (V1)
Essence / Nature of AI — What is GAI?
Black box · Brain/neural structure · Pattern-matching machine · Statistical aggregator · Crowd/zeitgeist · Simulator · God/supernatural
AI Operation — How does GAI work?
Stochastic parrot · Next-word predictor · Learning from examples · Magician/transformer · Chewing/digesting information
Human-AI Relationship — How do we relate to GAI?
Calculator/spell-checker · Copilot/e-bike · Companion/peer · Bridge/diplomat · Drug/stimulant · Mirror
Societal Function — What does GAI do in society?
Librarian/curator · Drop-in remote worker · Cultural bridge · Illuminating paths · Generally enabling technology
AI Impact — How does GAI affect us?
Wave/tsunami · Nuclear explosion · Digital plastic · Double-edged sword · Extended mind · Race
V1 at a Glance
- Derived deductively from prior literature on AI metaphors
- 69 distinct metaphors organized into 5 descriptive categories, guided Maas (2023)
- Research question: What metaphors are already cataloged?
- Captures surface-level language, not underlying beliefs
- Overwhelmingly non-educational or student-focused
From Metaphors to Dimensions
- 5 descriptive categories
- 69 metaphors from prior literature
- Deductive coding only
- What metaphors appear in the data?
& team discussion →
- 5 categories retained
- ~35 entries remaining
- Similar metaphors merged
- Deductive + Inductive
in research question →
- 5 analytical dimensions
- 3 positions each
- Diagnostic, not descriptive
- Fully Inductive Structure
V1 → V2: Metaphors describing the same underlying belief were merged
e.g., “stochastic parrot” + “next-word predictor” + “probabilistic text generator” → Mimicry/Simulates Understanding
V2 → V3: Shifted the research question
Changed from “what metaphors do people use?” to “what do the metaphors reveal about the speaker’s mental model?”
Result: V3 is a diagnostic tool, with each dimension capturing a plane of variation across instructor mental models
It’s Like “X”