By Andreas Ehstand
AI-generated text based on Andreas Ehstand’s public project materials.
A model returns an answer. An interface displays it. Between those two events, a design decision has already been made: what status will the user give this output?
That question appears in several directions of my work: concepts translated into art, the experience of AI conversations, and human participation in teams that include software agents and robots. Here are three design proposals inspired by those public project themes.
1. What does this image represent?
The AUGMANITAI art collection includes AI-generated artistic interpretations of named concepts. A concept and an image give the viewer two things to examine together.
For an interface built around that relationship, I would make three elements easy to find: the starting concept, the resulting work, and an explanation of the choices connecting them. If an image emphasises tension through distance or contrast, that choice can be described in ordinary language.
This makes a useful design question available: can the viewer tell which relationships came from the starting idea and which were introduced during its interpretation? The answer could guide captions, comparison views and the placement of explanations. These are proposals for making the process inspectable.
2. Where did the question change?
Imagine asking an AI system to help express an idea. Its response introduces a phrase. You adopt the phrase and continue. The next exchange now has a slightly different starting point.
My work on human–AI interaction includes attention to what happens during such exchanges. An interface could help by keeping the original question, the model's suggestion and the user's revision available for comparison.
The design problem extends beyond displaying a text difference. A shorter sentence may preserve the intention; a small change in wording may redirect it. A useful comparison view would let the user explain the change and return to an earlier formulation.
3. When does a suggestion become an instruction?
Consider a hypothetical task involving a person, a software agent and a robot. The agent proposes a step; the robot reports a condition; the person decides what should happen next.
SWAMANITAI explores human participation in mixed teams, including questions of attention, roles and responsibility. For developers, one practical starting point is the transition between suggestion, acknowledgement and action.
A proposed event view could distinguish observations, proposals, decisions and actions. Each entry would identify its source and show what remains pending. The interface would make it clear when an acknowledgement was received and when an action actually occurred.
The question to test is specific: can a person reconstruct who contributed what, and why the next action followed?
Across these examples, a useful interface gives people something concrete to inspect. Which transition in your own AI application is hardest for a user to explain afterwards?
Top comments (0)