Human Interaction with Conversational AI

---

Conversational AI platforms give rise to new forms of human interaction with computers. Whereas command-based interfaces were told what to do by typing command lines (in DOS or Linux) or by clicking visual elements (in Macintosh or Windows), Large Language Models (LLMs) rely on intention-based interfaces which respond to users’ prompts about what they want. Far from instantaneous, human and computer take turns, engaging in a back-and-forth dialogue over time. They become partners in conversation.

Anthropology and AI is my next book. It's an anthology published by Routledge and edited by Angela K. VandenBroek, Matt Artz, and Lora Koycheva, in which we demonstrate how anthropologists have consistently participated in shaping, critiquing, and reimagining AI systems over the past 70 years since cybernetics research. Anthropological perspectives remain essential for contextualizing AI as a complex sociotechnical phenomenon shaped by cultural assumptions.

My research shows how the study of natural conversation can guide the design of intention-based interfaces in order to make them more engaging partners in conversation. As engineering advancements bring Chat-GPT, Claude, Perplexity, and other generative dialogue platforms to emulate human-like speech, they succumb to the challenges that we already encounter in everyday conversation. When humans speak with one another, we can achieve mutual understanding, deeper bonds, and empathetic connection; but always at the risk of incomprehension, failure, and rejection – risks at any moment of a to-and-fro exchange unfurling through time. Far from perfect, conversational AI is the fragile creation of its fallible creators. In Anthropology and AI, I draw on case studies from customer service agents, dating apps, task automation Application Programming Interfaces (APIs), writing aids, legal assistants, flight training platforms, and healthcare chatbots to show how, in each case, the user and the platform are partners in dialogue. Intention-based interfaces involve turn-taking exchanges over time. And these exchanges partake the difficulties that humans face when expressing intentions, following the circuitous pathways of an open-ended dialogue whose outcome neither partner can secure in advance. 

My first foray into the anthropology of AI appeared as a 2023 article for Psyche, "Chatbots remind us that natural conversation is artificial too." I make the case that conversational AI magnifies the dynamics of natural conversation. Efforts to correct hallucinations, to authenticate authorship, and to verify citations reflect normal anxieties that we experience when doubting the meaning behind other humans' words. The techniques that we already use to navigate those uncertainties in daily conversation offer product developers a guide to relieve the doubts posed afresh by synthetic text.

I developed design guidelines for enhancing intention-based interfaces in my article for Ethnographic Praxis in Industry Conference, "Quit Playing Language Games with My Heart; or, Conversational AI and Knowledge Sharing." I study the language learning apps, Duolingo and Babbel, which transformed their language games in 2023 from rule-based to generative dialogue exercises built with transformer models and Natural Language Processing. Dynamic, back-and-forth dialogues came to appear alongside rigid multiple-choice and drag-and-drop vocabulary games. The apps combine elements of both command-based interfaces and intention-based interfaces.

Conversational AI platforms stand to benefit from a similar approach to interface design. Two elements stand out. First, Facets appear beneath the dialogue frame and provide filtered navigation when a user is unsure how to proceed. Second, translation overlays enable users to hover over words and reveal their translation, which proves useful when multiple meanings might be at play. To ensure that future generative dialogue systems remain useful and open to diverse forms of conversation, I suggest that it's imperative for UX teams to integrate intention-based interfaces with command-based features.

We are just beginning to witness conversational AI platforms’ transformation of human interaction with computers. Progress depends on learning from the possibilities and pitfalls of natural interactions among humans.