2025-09-21 –, Space 2
This talk traces how the uncanny—that productive tension between familiar and strange—shapes effective prompt engineering. I'll explore how Weizenbaum's, Mori’s and Turkle's pioneering work reveals psychological dynamics we can strategically leverage, and how understanding this lineage leads to more powerful human-AI interactions.
Long before ChatGPT, Joseph Weizenbaum's ELIZA program (1966) revealed something profound: users formed inexplicable emotional attachments to simple text interfaces, attributing understanding where none existed. Weizenbaum grew concerned about these effects, becoming one of AI's earliest critics. Similarly, roboticist Masahiro Mori (1970) identified the 'uncanny valley'—that distinctive cognitive response when something appears almost, but not quite, human.
While these pioneers saw the uncanny as a problem to be avoided, Sigmund Freud's earlier analysis in 'The Uncanny' (1919) had explored this psychological phenomenon more neutrally as a complex interplay between the familiar and strange. In the 1980s, Sherry Turkle described user-computer interactions as ‘uncanny’, noting how people formed relationships with machines that were artificial yet emotionally significant.
This talk reveals an overlooked connection: what many considered a problem to avoid is behind some of our most effective prompt engineering techniques. I'll demonstrate how these techniques leverage uncanny psychological dynamics, highlighting the theoretical principles that help explain why these approaches work so well in practice.
I'll begin by examining the psychological foundations of human-computer interaction through ELIZA's architecture, briefly illustrating how this simple program's pattern-matching mechanisms triggered profound psychological effects. By analysing key elements of ELIZA's code structure, I'll show how the same technical techniques that alarmed Weizenbaum can be viewed through Turkle's theoretical lens of the ‘second self’ as creating productive engagement rather than just deception.
Building on these contrasting perspectives, I'll demonstrate how established prompt engineering techniques leverage these uncanny dynamics, while their theoretical foundations remain largely unexplored. Using Python examples with the OpenAI API, I'll show how methods like chain-of-thought prompting (Wei et al., 2022) and role prompting position AI in the productive uncanny space between computational tool and simulated agent.
Attendees will leave with an understanding of how current best practices in prompt engineering strategically harness the uncanny tension in human-AI interaction, along with a deeper theoretical framework for understanding why these techniques work. By recognising both the risks and the productive potential of the uncanny, developers can make more informed design choices that leverage psychological principles that have influenced human-computer interaction for over fifty years.
Basic
I'm an AI Training & Adoption Strategist and academic researcher with software engineering experience. My work bridges the creative and technical, drawing on two decades in academia to bring rigour, curiosity and communication skills to software development and AI. I’m currently building a real-time platform for collecting feedback during live workshops and events, and I speak and write regularly about the overlaps between performance and code.