The PAIR project is an initiative of Google to research about AI where human input plays an early role in the learning process of an AI system. According to its website, PAIR is “interested in the full spectrum of human interaction with machine intelligence, from supporting engineers to understanding everyday experiences with AI.”
PAIR provides a guidebook to help build human-centered AI products, which should focus on five aspects:
- User needs + defining success: identify user needs, find AI opportunities, and design your reward function
- Mental models: introduce users to the AI system and set expectations for system-change over time
- Feedback + control: design feedback and control mechanisms to improve your AI and the user experience
- Data collection + evaluation: decide what data are required to meet your user needs, source data, and tune your AI
- Explainability + Trust: explain the AI system and decide if/when/how to show model confidence
- Errors + graceful failure: identify and diagnose AI and context errors and communicate the way forward.
This people-centric approach is fundamentally different from the conventional data-centric approach of machine learning. “Advancing the state of the art in machine learning means thinking beyond narrowly-defined objective functions. It requires weaving user needs and societal values into the design and evaluation of these systems”, said Fernanda Viegas, a Senior Staff Research Scientist on the PAIR team.
Jess Hozbrook, a PAIR lead and one of the creators of the People + AI Guidebook, compares today’s AI with the early days of designing products for personal computing, web, and mobile. “You experiment, your product looks really cool, and then you stumble by building an experience that doesn’t address a real human need or aspiration. We’re reminding people of what we know works, which is to put people first and work from there… If you start with people then any exploration, product design, or research you do will have a fruitful path.”
Putting people first in AI development will change the way we develop AI products. AI designers can harness and humanize AI’s vast potential.
AI World Society (AIWS) is a 7-layer model, in which Ethics is in Layer 2, and create AIWS Ethics Index.