Guidebook materials

Exercises for cross-functional team members to help guide key discussions, expedite design iteration, plan for user research, and set effective AI development timelines, along with guidebook chapters in easily printable pdfs

User Needs + Defining Success: chapter | worksheet

Data Collection + Evaluation: chapter | worksheet

Mental Models: chapter | worksheet

Explainability + Trust: chapter | worksheet

Feedback + Control: chapter | worksheet

Errors + Graceful Failure: chapter | worksheet

Need it all, offline? Download everything | All chapters | All worksheets

About PAIR

People + AI Research (PAIR)
All about the work, publications, and people from the PAIR team.

PAIR channel on the Google Design blog
Articles from the PAIR team about how to take a multidisciplinary and human-centered approach to designing with AI.

AI resources

Google’s AI Principles
Google’s recommended objectives for AI applications, a list of AI applications Google won’t pursue, and Google’s long-term vision for AI.
Referenced in - User Needs + Defining Success, Data Collection + Evaluation

Google’s Responsible AI Practices
Google’s general recommended best practices for AI.
Referenced in - User Needs + Defining Success, Data Collection + Evaluation

Google’s AI tools
An aggregation of all of Google’s open-source AI tools including Tensorflow, ML Kit, and more.

Google AI Education
Courses, guides, and videos to help you learn the technical skills to build AI.

Google Material Design’s Machine Learning Patterns for ML
Design guidelines and patterns for machine learning-powered features, created in partnership with ML Kit.

Google’s perspectives on issues in AI governance
Proposal for self-regulation guidelines related to AI development.

Google’s Machine Learning Fairness Overview and Crash Course
Learn how Google aims to develop the benefits of machine learning for everyone.
Referenced in - User Needs + Defining Success, Data Collection + Evaluation

AI for Everyone
Overview of AI from Andrew Ng covering what AI is, the basic terminology, the value it can add to a business, and other high-level considerations about how to use AI responsibly.

Existing datasets for training machine learning models

Use these to develop your models so you can test working prototypes and refine your product.
Referenced in - Data Collection + Evaluation

Cloud AutoML
Pre-trained ML models for use in unsupervised learning.

Google Dataset Search
Open-source public data on a range of domains.

A searchable compendium of labeled datasets with tags.

Resources for building datasets to train machine learning models

Learn how to collect enough of the right data to teach a model and develop better products.
Referenced in - Data Collection + Evaluation

Data Split Example
Split your data into training sets, validation sets, and testing sets.

Imbalanced Data
Identify and rectify data imbalances.

The Size and Quality of a Data Set
Understand dataset size and quality affect your model.

Resources for examining datasets for bias

Use these interactive websites to slice your data by demographic or category, or see what happens if you add or remove data.
Referenced in - User Needs + Defining Success, Data Collection + Evaluation

Explore two robust visualizations to help you understand and analyze your machine learning datasets.

What If tool
Analyze a machine learning model without writing any further code with an interactive visual interface for exploring model results.

Other resources for human-centered design

Design ideation methods from IDEO
Repeatable approaches to human-centered design.
Referenced in - User Needs + Defining Success

Error Messaging Guidelines from the Nielsen Norman Group
General best practices for communicating errors and helping user move forward.
Referenced in - Errors + Graceful Failure

Ethical OS
A toolkit for identifying and discussing ethical questions and considering long-term social impact of in-development technology.

A method for evaluating early design concepts that is particularly helpful for AI products.