Abstract
When a Tesla promotional video was allegedly staged in 2016 to demonstrate capabilities the vehicle did not actually possess, it raised a deeper question about how the way we represent emerging technologies shapes how people understand, trust, and ultimately use them. This is the central question of my SEAS portfolio. The technical project, a Meeting Notes Summarizer powered by natural language processing, addresses how AI tools can be built to support rather than undermine human understanding. The STS research project examines how corporate promotional materials for autonomous vehicles construct sociotechnical imaginaries that make the transfer of control from human to machine seem natural, desirable, and inevitable, while quietly redistributing responsibility in ways that remain politically unresolved. Both projects engage the same underlying problem from different angles, one through design and one through cultural analysis, and together they reveal why STS frameworks are essential to engineering practice. Engineers who optimize only for technical performance, without attending to how their systems are culturally received and represented, risk building tools that work but are not trusted, or are trusted for the wrong reasons.
The technical portion of my thesis produced a working prototype of a Meeting Notes Summarizer, an NLP tool that uses transformer-based models, specifically BART and T5, to automatically generate structured, modularized summaries from raw meeting transcripts. Unlike existing tools such as Otter.ai or Cluely, which offer basic transcription and surface-level summaries, my system segments transcripts into coherent discussion modules, highlights action items alongside associated speakers, and embeds explainability features that allow users to trace each summary line back to its source text. This "trace-back" highlighting, combined with editable summary outputs, is the system's distinguishing design choice: it keeps the human in the decision-making role rather than presenting AI output as authoritative and final. The system is evaluated using ROUGE and BERTScore metrics alongside a user evaluation measuring whether participants recall meeting details significantly faster using the summarizer versus raw transcripts. The broader significance of this project is its demonstration that even a technically functional AI tool must be designed to earn trust through transparency and user control, not merely through accuracy.
In my STS research, I examined how companies including Tesla, Mercedes-Benz, Ford, and Cadillac frame partially automated driving systems in promotional materials. Applying multimodal discourse analysis, sociotechnical imaginaries theory, and Winner's concept of technological politics, I analyzed how textual and visual cues work together to construct a specific kind of user that is passive, relaxed, and fully trusting of the automated system. The analysis found a consistent pattern of what Dixon calls "autonowashing," where the aesthetic language of full autonomy is applied to SAE Level 2 systems that still require active driver supervision. Promotional materials emphasize effortlessness and comfort while safety disclaimers are pushed into the background, creating what Elish describes as a "moral crumple zone," where the human driver retains legal responsibility even as control appears to shift to the machine. The significance of this finding is that advertising actively shapes public expectations about safety, accountability, and risk before most people have any direct experience with these systems.
Considering both projects together through an STS lens reveals how technical reliability is not the same as trustworthiness, and the gap between them is cultural. My meeting summarizer may generate accurate outputs, but users who cannot trace or modify those outputs will distrust the system regardless of its performance, because they cannot verify where agency lies. The same logic operates in AV advertising but in reverse: promotional materials create the appearance of trustworthiness and control precisely by hiding the complexity of where agency actually resides. Winner's argument that artifacts have politics applies to both contexts. The summarizer's design is a small attempt to build a tool that is honest about what it is doing and who remains in charge, which works against the idea of using design and narrative to obscure those questions entirely. Engineers who understand both dimensions, who can build systems and critically read the cultural narratives surrounding them, are better positioned to make responsible choices about what they build and how it gets communicated to the world.
