Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

The System Model and the User Model: Exploring AI Dashboard Design

Fernanda Viégas, Martin Wattenberg

TL;DR

This note argues that dialogue-focused AI systems should be equipped with parallel dashboards that expose their internal state to users. It introduces the world model hypothesis and defines two universal constructs—the User Model and the System Model—as key internal representations worth visualizing. The authors outline a design space for instrumentation, including what features to display, whether displays should be static or dynamic, and how to handle imperfect or evolving models, while cautions against anthropomorphic embellishments and considers adversarial risks. The paper aims to spark targeted UX research to improve trust, safety, and usability by surfacing interpretable internal state in AI systems, potentially transforming how humans interact with complex neural networks.

Abstract

This is a speculative essay on interface design and artificial intelligence. Recently there has been a surge of attention to chatbots based on large language models, including widely reported unsavory interactions. We contend that part of the problem is that text is not all you need: sophisticated AI systems should have dashboards, just like all other complicated devices. Assuming the hypothesis that AI systems based on neural networks will contain interpretable models of aspects of the world around them, we discuss what data such dashboards might display. We conjecture that, for many systems, the two most important models will be of the user and of the system itself. We call these the System Model and User Model. We argue that, for usability and safety, interfaces to dialogue-based AI systems should have a parallel display based on the state of the System Model and the User Model. Finding ways to identify, interpret, and display these two models should be a core part of interface research for AI.

The System Model and the User Model: Exploring AI Dashboard Design

TL;DR

This note argues that dialogue-focused AI systems should be equipped with parallel dashboards that expose their internal state to users. It introduces the world model hypothesis and defines two universal constructs—the User Model and the System Model—as key internal representations worth visualizing. The authors outline a design space for instrumentation, including what features to display, whether displays should be static or dynamic, and how to handle imperfect or evolving models, while cautions against anthropomorphic embellishments and considers adversarial risks. The paper aims to spark targeted UX research to improve trust, safety, and usability by surfacing interpretable internal state in AI systems, potentially transforming how humans interact with complex neural networks.

Abstract

This is a speculative essay on interface design and artificial intelligence. Recently there has been a surge of attention to chatbots based on large language models, including widely reported unsavory interactions. We contend that part of the problem is that text is not all you need: sophisticated AI systems should have dashboards, just like all other complicated devices. Assuming the hypothesis that AI systems based on neural networks will contain interpretable models of aspects of the world around them, we discuss what data such dashboards might display. We conjecture that, for many systems, the two most important models will be of the user and of the system itself. We call these the System Model and User Model. We argue that, for usability and safety, interfaces to dialogue-based AI systems should have a parallel display based on the state of the System Model and the User Model. Finding ways to identify, interpret, and display these two models should be a core part of interface research for AI.
Paper Structure (19 sections, 2 figures)

This paper contains 19 sections, 2 figures.

Table of Contents

  1. Introduction
  2. The interpretable world model hypothesis
  3. Why the world model hypothesis is plausible
  4. Weakening the world model hypothesis
  5. User and System: universally important models?
  6. The User Model
  7. Relation to Theory of Mind
  8. The System Model
  9. Domain-specific models
  10. Instrumentation for AI: the design space
  11. Choosing which features to display
  12. Static vs. dynamic choices of displays
  13. Simple interfaces for complex systems
  14. Why show imperfect models of the world?
  15. Anthropomorphic interfaces and their pitfalls
  16. ...and 4 more sections

Figures (2)

  • Figure 1: Examples of existing machine instrumentation. Left: tachometer and speedometer in car dashboard. Center: stove top indicator of heat. Right: self-driving display in a Tesla car.
  • Figure 2: Speculative mock: A "System Model/User Model" dashboard display on the left side of a dialog system. The design imagines a system where the user can choose to see various aspects of the user and system models. We have deliberately chosen some provocative dimensions to highlight potential issues. The orange highlight indicates a rapid change, in this case a switch from factual information to a more "fictional" response. (Dialog based on ChatGPT output.)