This is the second dispatch in our series of Grand Design Challenges for 2025. You can read Part 1 here. Keep checking back for the final three challenges.
Artificial Intelligence User Interfaces
We have anthropomorphized AI. It is ChatGTP, Siri, and Alexa, and soon, a proliferation of agents decorated with personalities, possibly named Smith. There is an innate desire (sometimes crossing over into a fetish) to make AI a conscious personality. Anthropomorphic AI limits the interaction to a conversational model, which feels lazy, heedless, or, at worst, an underdeveloped first attempt at which we raise our hands and say, “Ta-da! It's done.”
Conversation is not the apex of communication and should not be the stopping point but a fragment of a whole multi-model interface. Conversational interfaces, while natural and intuitive, are also inefficient, ambiguous, and simplistic. The predominant chat interface minimizes the design to a text input component – a square box. A harshly barren UI replaces visual design with personality design, making UX an enormous problem in terms of systems design and character world-building.
A conversational UI over-emphasizes direct and explicit input (spoken or written word) while ignoring other critical components of a conversational interaction. A human-centric conversational interaction model is challenging to design and implement because a complete human conversation interaction must include facial expressions and body mannerisms that are not communicated by the human or the AI – in most interactions, the AI has no ability to sense physical human expressions, nor can it create its own “human” expressions.
Instead of forcing AI to mimic humans, let it retain its alien nature. Anthropomorphic AI fools us into thinking of the AI as an entity separate from the computing system on which it runs, which can limit the critical characteristics and advantages of both being human and being a computer. Keeping AI compumorphic does not mean computers have to retain “classific” interfaces; rather, the interfaces can be something entirely new.
Focal Points:
- What are the AI interaction models (unlikely to be just one) and design patterns for effectively applying AI?
- When is natural language ideal or appropriate for human-computer interactions?
- When is a natural language interface (NLI) the sole interface or part of a multi-model one? Further, when is an NLI best at being the primary or a secondary interface?
- When should AI be nothing more than a contributor to the interface and not the primary interface – a UX in which AI participates is indistinguishable from one solely generated by human, hand-coded software?
