book

Designing Autonomous AI

by Kence Anderson

June 2022

Intermediate to advanced

245 pages

6h 55m

English

O'Reilly Media, Inc.

Read now

Unlock full access

What Is Autonomous AI?Who Should Read This Book?Process ExpertsData Scientists and Software EngineersInnovation LeadersTeachersProblem SolversWhat Can You Expect to Learn from This Book?Conventions Used in This BookO’Reilly Online LearningHow to Contact UsAcknowledgments
The Changing World Requires Adapting SkillsProblems Need Solutions, Not AIWhat Can AI Do for Me in Real Life?AI Decision-Making Is Becoming More AutonomousBeware of Data Science ColonialismThe Changing Workforce Demands Transferred SkillsExpertise Is Hard to AcquireExpertise Is Hard to MaintainExpertise Is Simple to Teach, but Requires PracticePressing Problems Demand Completely New SkillsAI Is a Tool; Use It for Good
Math, Menus, and Manuals: How Machines Make Automated DecisionsControl Theory Uses Math to Calculate DecisionsOptimization Algorithms Use Menus of Options to Evaluate DecisionsExpert Systems Recall Stored Expertise
Augmenting Human IntelligenceHow Humans Make Decisions and Acquire SkillsHumans Act on What They PerceiveHumans Build Complex Correlations into Their Intuition with PracticeHumans Abstract to Strategy for Complex TasksThere’s a New Kind of AI in TownThe Superpowers of Autonomous AIAutonomous AI Makes More Human-Like DecisionsAutonomous AI Perceives, Then ActsThe Difference Between Perception and Action in AIAutonomous AI Learns and Adapts When Things ChangeAutonomous AI Can Spot PatternsAutonomous AI Infers from ExperienceAutonomous AI Improvises and StrategizesAutonomous AI Can Plan for the Long-Term FutureAutonomous AI Brings Together the Best of All Decision-Making TechnologiesWhen Should You Use Autonomous AI?Autonomous AI Is like a Brilliant, Curious Toddler That Needs to Be Taught
Learning Multiple Skills Simultaneously Is Hard for Humans and AITeaching Skills and Strategies ExplicitlyTeaching Allows Us to Trust AIThe Mindset of a Machine TeacherTeacher More Than ProgrammerLearner More Than ExpertWhat Is a Brain Design?How Decision-Making WorksAcquiring Skill Is like Learning to Navigate by ExploringA Brain Design Is a Mental Map That Guides Exploration with Landmarks
Case Study: Learning to Walk Is Hard to Evolve, Easier to TeachSo, Why Do We Walk?Strategy Versus EvolutionTeaching Walking as Three SkillsConcepts Capture KnowledgeSkills Are Specialized ConceptsBrains Are Built from SkillsBuilding SkillsExpert Rules Inflate into SkillsPerceptive Concepts Discern or RecognizeDirective Concepts Decide and ActSelective Concepts Supervise and AssignBrains Are Organized by Functions and StrategiesSequences or Parallel Execution for Functional SkillsHierarchies for StrategiesVisual Language of Brain Design
Understanding the ProcessMeet with ExpertsAsk the Right QuestionsCase Study: Let’s Design a Smart Thermostat

Determining Which Actions the Brain Will TakePerception Is Required, but It’s Not All We NeedSequential DecisionsTriggering the Action in Your AI BrainSetting the Decision FrequencyHandling Delayed Consequences for Brain ActionsActions for Smart Thermostat
There’s Always a Trade-offThroughput Versus EfficiencySupervisors Have Different Goals Than Crews DoDon’t Prioritize Goals; Balance Them InsteadWatch Out for Expert Rules Disguised as GoalsIdeal Versus AvailableSetting GoalsStep 1: Identify ScenariosStep 2: Match Goals to ScenariosStep 3: Teach Strategies for Each ScenarioGoal ObjectivesMaximizeMinimizeReach, like the Finish Line for a RaceDrive, like the Temperature for a ThermostatAvoid, like Dangerous ConditionsStandardize, like the Heat in an OvenSmooth, like a LineExpanding Task Algebra to Include Goal ObjectivesSetting Goals for a Smart Thermostat
Teaching Focuses and Guides Practice (Exploration)Skills Can Evolve and TransformSkills Adapt to the ScenarioLevels of Teaching SophisticationThe Introductory Teacher Conveys the Facts and GoalsThe Coach Sequences Skills to PracticeThe Mentor Teaches StrategyThe Maestro Democratizes New ParadigmsHow Maestros Democratize TechnologyLevels of Autonomous AI ArchitectureMachine Learning Adds PerceptionMonolithic Brains Are Advanced BeginnersConcept Networks Are Competent LearnersMassive Concept Networks Are Proficient LearnersPursuing Expert Skill Acquisition in Autonomous AIBrains That Come with Hardwired SkillsBrains That Define Skills as They LearnBrains That Assemble ThemselvesBrains with Skills That CoordinateSteps to Architect an AI BrainStep 1: Identify the Skills That You Want to TeachStep 2: Orchestrate How the Skills Work TogetherStep 3: Select Which Technology Should Perform Each SkillPitfalls to Avoid When Teaching SkillsPitfall 1: Confusing the solution for the problemPitfall 2: Losing the forest for the treesExample of Teaching Skills to an AI Brain: Rubber FactoryBrain Design for Our Smart Thermostat
Sensors: The Five Senses for Your AI BrainVariablesProxy VariablesTrendsSimulators: A Gym for Your Autonomous AI to Practice InSimulating Reality Using Physics and ChemistrySimulating Reality Using Statistics and EventsSimulating Reality Using Machine LearningSimulating Reality Using Expert RulesSensor Variables for Smart Thermostat
Designers and Builders Working Together in Harmony (Mostly)The Autonomous AI Design Fallacy Designs but Won’t IterateThe Autonomous AI Implementation Fallacy Skips Design AltogetherSpecification for Documenting AI Brain DesignsPlatform for Machine TeachingPlatform for Wiring Multiple Skills Together as ModulesWhat Difference Will You Make with Machine Teaching?

Content preview from Designing Autonomous AI

Foreword

It gives me great pleasure to enthusiastically recommend Designing Autonomous AI, written by a leading real-world practitioner in emergent autonomous AI: Kence Anderson.

As the Corporate Vice President currently responsible for new product incubations at Microsoft, and as someone who has been closely involved with creation of new product categories like VPNs, real-time communications and now autonomous systems at Microsoft, I have learned that creating new categories is not a straightforward path.

Established product categories and their underlying technologies and business models are well honed for the real world with well understood problems and solutions. These are things we use in our daily lives.

Fundamental research is important to new category creation but lives in the ether of science, generally abstracted from real world considerations. Researchers create novel methods and demonstrate their promise through results in carefully crafted experiments—the amazing state of the art in AI is a recent example of this.

Importantly, not every method created by researchers is sufficient or workable in the real world. New categories can be created when one or more of these new methods are developed further, grounded in and applied to the real-world environment.

New category creation is a chaotic, primordial process entailing a loop of exploration, learning, and adaptation. This process is very high dimensional and domain dependent; there is no single formula here. However, ...