Designing Human-Centric AI Experiences: Applied UX Design for Artificial Intelligence (Design Thinking)

Table of Contents
About the Author
About the Technical Reviewer
Acknowledgments
Preface
Part 1: Intelligence
	Chapter 1: On Intelligence
		Many Meanings of AI
			Thinking Humanly
			Thinking Rationally
			Acting Humanly
			Acting Rationally
			Substrate Independence
		Foundations of Artificial Intelligence
			Philosophy
			Mathematics
			Economics
			Neuroscience
			Psychology
			Computer Engineering
			Control Theory and Cybernetics
			Linguistics
			Business
			Why Is AI a Separate Field?
		Superintelligence and Artificial General Intelligence
		Narrow AI
			Rules vs. Examples
				Rules-Based
				Examples-Based
			A Fundamental Difference in Building Products
			Intelligent Everything
		User Experience for AI
			Beneficial AI
		Summary
	Chapter 2: Intelligent Agents
		Rational Agent
		Agents and Environments
			Agent
			Environment
				Simple Environments
				Complex Environments
			Sensors
			Actuators
			Goals
		Input-Output
			Learning Input-Output Mappings
				Machine Learning (ML)
					Supervised Learning
					Unsupervised Learning
					Reinforcement Learning
					Deep Learning (DL)
				Feedback Loops
					Rewards
					The Risk of Rewards
			The Probabilistic Nature of AI
		Summary
Part 2: Decisions
	Chapter 3: Incorporating Artificial Intelligence
		A Cognitive Division of Labor
			What Machines Do Better
			What Humans Do Better
		Human + Machine
			Supermind
			Collective Intelligence
				Addition
				Improvement
				Connection
			Artificial Collective Intelligence
				Improving Machines
				Automating Redundant Tasks
				Improving Machine-Machine Collaboration
				Improving Human-Machine Collaboration
			Missing Middle
				Cobots
				Roles for AI
					Tools
					Assistants
					Peers
					Managers
		Finding AI Opportunities
			Jobs
			Tasks
			Breaking Down Jobs into Tasks
				Example: Personal Running Trainer
			Mapping User Journeys
				Experience Mapping
					Characteristics
				Journey Mapping
					Characteristics
				User Story Mapping
					Characteristics
				Service Blueprints
					Characteristics
			Problem-First Approach
				When Does It Not Make Sense to Use AI
					Maintaining Predictability
					Minimizing Costly Errors
					Complete Transparency
					Optimizing for High Speed
					Optimizing for Low Costs
					Static or Limited Information
					Data Being Sparse
					Social Intelligence
					People Not Wanting AI
				When Does It Make Sense to Use AI
					Personalization
					Recommendation
					Recognition
					Categorization and Classification
					Prediction
					Ranking
					Detecting Anomalies
					Natural Language Understanding
					Generating New Data
				Identifying Tasks Suitable for AI
			Considerations for AI Tasks
				Type of Action
					Augmentation
						When to Augment
						Measuring Successful Augmentation
					Automation
						When to Automate
						Measuring Successful Automation
						Human in the Loop
					Example: Training for a Marathon
				Type of Environment
					Full or Partial Observability
					Continuous or Discrete Actions
					Number of Agents
					Predictable or Unpredictable Environments
					Dynamic or Static Environments
					Time Horizon
				Data
					Availability
					Access
						Access from Within
						External Access
					Compounding Improvements
				Cost
				Time and Effort
				Quality Improvements and Gains
				Societal Norms
		Big Red Button
			Levels of Autonomy
		Rethinking Processes
			Netflix
			Mercedes
		Summary
Part 3: Design
	Chapter 4: Building Trust
		Trust in AI
			Components of User Trust
				Competence
				Reliability
				Predictability
				Benevolence
			Trust Calibration
			How to Build Trust?
				Explainability
				Control
		Explainability
			Who Needs an Explanation?
				Decision-Makers
				Affected Users
				Regulators
				Internal Stakeholders
			Guidelines for Designing AI Explanations
				Make Clear What the System Can Do
				Make Clear How Well the System Does Its Job
				Set Expectations for Adaptation
				Plan for Calibrating Trust
				Be Transparent
				Build Cause-and-Effect Relationships
				Optimize for Understanding
			Types of Explanations
				Data Use Explanations
					Guidelines for Designing Data Use Explanations
					Types of Data Use Explanations
						Scope of Data Use
						Reach of Data Use
						Examples-Based Explanations
							Generic Explanations
							Specific Explanations
				Descriptions
					Guidelines for Designing Better Descriptions
					Types of Descriptions
						Partial Explanations
						Full Explanations
				Confidence-Based Explanations
					Guidelines for Designing Confidence-Based Explanations
					Types of Confidence-Based Explanations
						Categorical
						N-Best Results
						Numeric
						Data Visualizations
				Explaining Through Experimentation
					Guidelines to Design Better Experimentation Experiences
				No Explanation
			Evaluating Explanations
				Internal Assessment
				User Validation
					Qualitative Methods
					Quantitative Methods
		Control
			Guidelines for Providing User Control
				Balance Control and Automation
				Hand Off Gracefully
			Types of Control Mechanisms
				Data Control
					Global Controls
					Editability
					Removal and Reset
					Opting Out
				Control over AI Output
					Provide a Choice of Results
					Allow Users to Correct Mistakes
					Support Efficient Dismissal
					Make It Easy to Ignore
		Borrowing Trust
		Opportunities for Building Trust
			Onboarding
				Set the Right Expectations
				Introduce Features Only When Needed
				Clarify Data Use
				Allow Users to Control Preferences
				Design for Experimentation
				Reboarding
			User Interactions
				Set the Right Expectations
				Clarify Data Use
				Build Cause-and-Effect Relationships
				Allow Users to Choose, Dismiss, and Ignore AI Results
			Loading States and Updates
			Settings and Preferences
				Provide Global Data Controls
				Clarify Data Use
				Allow Editing Preferences
				Allow Users to Remove or Reset Data
				Allow Opting Out
			Errors
				Adjust User Expectations
				Hand Off Gracefully
				Allow Users to Correct AI Mistakes
				Allow Users to Choose, Dismiss, and Ignore AI Results
		Personality and Emotion
			Guidelines for Designing an AI Personality
				Don’t Pretend to Be Human
				Clearly Communicate Boundaries
				Consider Your User
				Consider Cultural Norms
				Designing Responses
					Grammatical Person
					Tone of Voice
					Strive for Inclusivity
					Don’t Leave the User Hanging
				Risks of Personification
		Summary
	Chapter 5: Designing Feedback
		Feedback Loops in Artificial Intelligence
		Types of Feedback
			Explicit Feedback
				Using Explicit Feedback
				Guidelines for Incorporating Explicit Feedback
			Implicit Feedback
				Using Implicit Feedback
				Guidelines for Incorporating Implicit Feedback
			Dual Feedback
		Align Feedback to Improve the AI
			Reward Function
			Collaborate with Your Team
		Collecting Feedback
			Consider the Stakes of the Situation
			Make It Easy to Provide Feedback
				Encourage Feedback During Regular Interactions
				Allow Correction When the AI Makes Mistakes
			Explain How Feedback Will Be Used
				Guidelines for Explaining Feedback Use
			Consider User Motivations
				Reward
					Symbolic Rewards
					Material Rewards
					Social Rewards
				Utility
				Altruism
				Self-Expression
		Responding to Feedback
			On-the-Spot Response
				Connect Feedback to Changes in the User Experience
				Clarify Timing and Scope
				Set expectations for adaptation
				Limit Disruptive Changes
			Long-Term Response
		Control
			Editability
			Removal and Reset
			Opting Out
			Make It Easy to Ignore and Dismiss
			Transparency
		Human-AI Collaboration
		Summary
	Chapter 6: Handling Errors
		Errors Are Inevitable in AI
			Humble Machines
		Guidelines for Handling AI Errors
			Define “Errors” and “Failures”
			Use Feedback to Find New Errors
			Consider the Type of Error
				System Errors
				User Errors
				User-Perceived Errors
			Understand the Stakes of the Situation
			Indicate That an Error Occurred
			Don’t Blame the User
			Optimize for Understanding
			Graceful Failure and Handoff
				Provide Appropriate Responses
				Use Errors as Opportunities for Explanation
				Use Errors as Opportunities for Feedback
				Disambiguate when Uncertain
				Return Control to the User
				Assume Intentional Abuse
		Strategies for Handling Different Types of Errors
			System Errors
				Data Errors
					Mislabeled or Misclassified Data
						Error Resolution
					Incomplete Data
						Error Resolution
					Missing Data
						Error Resolution
				Relevance Errors
					Low-Confidence Results
						Error Resolution
					Irrelevance
						Error Resolution
				Model Errors
					Incorrect Model
						Error Resolution
					Miscalibrated Input
						Error Resolution
					Security Flaws
						Error Resolution
				Invisible Errors
					Background Errors
						Error Resolution
					Happy Accidents
						Error Resolution
			User Errors
				Unexpected or Incorrect Input
					Error Resolution
				Breaking User Habits
					Error Resolution
			User-Perceived Errors
				Context Errors
					Error Resolution
				Failstates
					Error Resolution
		Recalibrating Trust
		Summary
	Chapter 7: AI Ethics
		Ethics-Based Design
		Trustworthy AI
			Explainable AI
				Black Box Models
				Transparency
			Bias
				Facial Recognition
				Causes of Bias
					Bias in Training Data
					Lack of Team Representation
				Reducing Bias
			Privacy and Data Collection
				Protect Personally Identifiable Information
				Protect User Data
				Ask Permissions
				Explain Data Use
				Allow Opting Out
				Consider Regulations
				Go Beyond “Terms and Conditions”
			Manipulation
				Behavior Control
				Personality
					Risks of Personification
		Safe AI
			Security
			Accountability and Regulation
				Accountability
				Law
					Liability
					Independent Review Committees
		Beneficial AI
			Control Problem
			Beneficial Machines
				Principles of Beneficial Machines
			Human in the Loop
		Summary
	Chapter 8: Prototyping AI Products
		Prototyping AI Experiences
			Desirability
			Usability
				Types of Usability Prototypes
					Using Personal Examples
					Wizard of Oz Studies
					Minimum Viable Product
			Explainability
				Internal Assessment
				User Validation
			Relevance
		Hardware Prototypes
		Summary
Part 4: Teamwork
	Chapter 9: Understanding AI Terminology
		Key Approaches for Building AI
		AI Techniques
			Supervised Learning
			Unsupervised Learning
			Reinforcement Learning
			Deep Learning and Neural Networks
			Backpropagation
			Transfer Learning
			Generative Adversarial Networks (GANs)
			Knowledge Graphs
		AI Metrics
			Accuracy
			Precision
			Recall
			Precision vs. Recall Tradeoff
		AI Capabilities
			Computer Vision (CV)
			Natural Language Processing (NLP)
			Speech and Audio Processing
			Perception, Motion Planning, and Control
			Prediction
			Ranking
			Classification and Categorization
			Knowledge Representation
			Recommendation
			Pattern Recognition
		Summary
	Chapter 10: Working Effectively with AI Tech Teams
		Common Roles in an AI Product Team
			Machine Learning Engineer
			Machine Learning Researcher
			Applied ML Scientist
			Software Engineer
			Data Engineer
			Data Scientist
			Product Manager
			Product Designer
		Effective Collaboration
			Easy Things Are Hard
			Collaborate; Don’t Dictate
				Share Problems, Not Solutions
			Motivation
				Build User Empathy
				Transparency About Product Metrics and User Feedback
				Storytelling
			Encourage Experimentation
				Hypothesis Validation
			Gathering Better Functional Requirements
				Data Requirements
				Feedback Mechanisms
				Understand Limitations
			Highlight Ethical Implications
		Summary
Epilogue
Contact Author
Index