Justine Cassell, Associate Dean for Technology, Strategy & Impact and Director Emerita, Human Computer Interaction Institute, Carnegie Mellon University
It is said that artificial intelligence (AI) is a way of modeling human behavior on the basis of data collected about human behavior. But traditionally AI is task-oriented, so this approach makes it awkward to interact with. This is because the AI’s behavior is only as good as the types of data it is fed, so as with people who have had unusual upbringings, there may be behavioral blind spots or inconsistencies.
Justine Cassell has decided that the data she is choosing to model for her AI, SARA (Socially Aware Robotic Assistant), is the data that comes from studying how we form relationships with others. For example:
- Verbal cues such as tone of voice. “I’m so pleased to meet you.” suggests sarcasm, while “I’m so pleased to meet you!” suggests excitement.
- Visual cues such as frowns and smiles, nods, and blinks
- Vocal behavior such as word choice
After collecting thousands of hours of data, Cassell developed a personal assistant whose goal is to constantly seek to improve its relationship with its user in real time. You can see a specific example in the video below.