XXIX Salon: Principles of Language for Autonomous Systems
Salon: Principles of Designing for Autonomous Systems: Communication 11/11/19
1. Premise: language is crucial in newness
- How the system communicates to users
- Branding of AI
- Communicating responsibility and legibility of the system — just because they’re hidden / hard to pinpoint doesn’t mean we can’t do interpretive work / rectify
2. Everyone wants to know the future: hunting for predictions
Subliminal desire for futurism
Next / new psychology
- The culture of hope
- Other terms for prediction:
- Segmentation — individual vs. collective
- What language can we use to identify that we are part of a segment?
- Personas and archetypes — is this the bias that led here in the first place?
- Human factors that led to cultural segmentation
- Articulating the unknowns in your prediction
- “people like you like”
3. Ethical Issues with False Definitions
- “My system is going to predict x, and x affects people” — problematic
- Human filter, bias
4. Avoid the hype (machines can’t be smart)
- Given: machines can’t be smart
- Through language, is there a way to help people be thoughtful in creating systems?
5. Gradients of machine autonomy: the use of language in setting expectations
- Machine work --------------- Human work
- What’s explicit, what’s insinuated?
- Agency: single-turn, multi-turn… how many turns before one can lose human agency?
- How do you tell somebody that their car is going to drive itself? How do you communicate
- A task is a decision point that progresses the system’s goal
- Communication → perception = decision (Signal → receiving the signal = result)
- High stakes ↔ low stakes / human ↔ machine
- Not a point, a vector
A Model for Types and Levels of Human Interaction with Automation Raja Parasuraman, Thomas B. Sheridan, Fellow, IEEE, and Christopher D. Wickens (Sam Kellogg) (Claire Mitchell)
6. Language in User Interfaces
- The ambiguity of language when using an autonomous system
- Translation, literal and nuanced of user intent
- Accounting for non-binary terms in a binary system
- Sentiment (volume…) : untranslatable affects