Foreword xvii

Acknowledgments xxi

Glossary xxix

Abbreviations xxxi

Overview 1

  In brief 1

  “ . . . AI will be either the best, or worst thing, ever to happen to humanity.” 4

  AI as a general-purpose technology: Tuning out the noise 5

  AI in developing countries: Adopt and adapt, then advance 11

  The promise and peril of AI: Assessing its economic, social, and political impacts 14

  Shaping AI: Governments as enablers, users, and regulators 23

  The future of AI’s impacts depends on the choices made today 31

  Notes 32

  References 35

Part 1. Decoding AI: Capabilities, Concentration, and Complements 43

Chapter 1. Capabilities: The Power of AI to Alleviate Skill Shortages 45

  Main messages 45

  Introduction 45

  The technology: AI is not new, but it is expanding in scope, improving continuously, and diffusing rapidly 46

  From technology to tasks: AI is now able to perform nonroutine cognitive tasks, facilitating decision-making 50

  From tasks to alleviating scarcity: AI increases capabilities that are relatively scarce in developing countries 53

  From promise to peril: Gains are neither equal nor guaranteed 57

  Notes 60

  References 61

Chapter 2. Concentration: Opportunities and Constraints in the AI Value Chain 65

  Main messages 65

  Introduction 65

  The layers: AI’s capabilities are developed through a series of specialized steps 66

  The landscape: The leading AI companies are concentrated in just a few economies 69

  The reality: Few economies can build the most advanced AI hardware, data centers, and models 69

  The overlooked: Participating in certain segments of the AI value chain plays to the strengths of developing countries 73

  The prospect: Adopting and adapting AI costs far less than building it from scratch 75

  The future: The favorable conditions today may not last 77

  Notes 79

  References 80

Chapter 3. Complements: Customizing AI to Local Realities and Needs 83

  Main messages 83

  Introduction 83

  The technology that learns: The risk of contextual mismatches is amplified because AI’s capabilities are derived from the data used to train it 84

  The technology that fits: Tailoring AI to the local context requires the right complementary factors 87

  The long journey: Putting complements in place is a slow and continuous process 92

  The roadblock: Access to complements may be limited 92

  The road not taken: Even where complements are in place, AI applications may still not be adopted 93

  The pathways: Adopt, adapt, and then advance 94

  Notes 95

  References 96

Spotlight 1. Small AI, Big Impact 101

Part 2. Assessing the Promise and Peril of AI: Its Economic, Social, and Political Impacts 107

Chapter 4. AI’s Economic Impact: Transforming Jobs, Productivity, and Growth 109

  Main messages 109

  Introduction 110

  AI as an input: AI holds significant promise to raise productivity, but uneven patterns of adoption are concentrating both disruptions and gains 112

  AI as an output: The AI value chain is creating growth opportunities in developing countries, but only a few firms are capturing value 130

  Cascading impacts: AI is also having indirect ripple effects across sectors and countries 133

  Future impacts: The most important impacts of AI may be unpredictable 134

  Notes 139

  References 141

Spotlight 2. The Power of AI to Enhance Agricultural Advisory and Extension Services 147

Chapter 5. AI’s Social Impact: Improving Public Service Delivery 155

  Main messages 155

  Introduction 155

  In the front end of government, evidence-backed AI applications can increase the quality of public services and expand access 157

  In the back end of government, AI can solve problems at scale to make public administration more effective, efficient, and precise 173

  AI can be successfully integrated into the public sector by prioritizing accountability, oversight, and procurement that fits government capacities 182

  Notes 183

  References 184

  Spotlight 3. Generative AI and the Erosion of Knowledge, Learning, and Human Skills 193

Chapter 6. AI’s Political Impact: Reshaping Power Within and Across Countries 201

  Main messages 201

  Introduction 201

  Dynamics between countries: A technology with geopolitical stakes 202

  Dynamics between governments and corporations: Entanglements could impede development 207

  Dynamics between corporations and individuals: Market structure determines who gains from AI 211

  Dynamics between citizens and the state: AI strengthens the state’s hand, for better or worse 215

  Policy choices will shape power imbalances among governments, corporations, and individuals 220

  Notes 220

  References 222

Spotlight 4. AI and the Environment 227

Part 3. Shaping AI: Governments as Enablers, Users, and Regulators 233

Chapter 7. Governments as Enablers: Establishing the Analog and Digital Complements of AI 235

  Main messages 235

  Introduction 236

  Adopting AI: The analog foundations must be in place 237

  From adopting AI to adapting AI and advancing AI: The analog and digital foundations must be in place 244

  Sequencing AI policy for firms and workers: Not everything can happen at once 257

  Notes 260

  References 262

Spotlight 5. Upgrading Social Safety Nets as AI Transforms the Nature of Work 267

Chapter 8. Governments as Users: Enhancing Capabilities to Deploy AI 273

  Main messages 273

  Introduction 274

  Getting AI right starts with matching it to the problem and the agency’s capacity to sustain it 276

  Investing in staff capacity is essential to make AI work and last 278

  Building AI-ready data requires targeted investments in digitization and digital public infrastructure 282

  Evaluating and monitoring performance ensures that AI improves services 290

  Procuring AI effectively involves keeping control and oversight 293

  Treating AI as an ongoing operational function of government requires different ways of thinking and funding 296

  Notes 297

  References 298

  Spotlight 6. Evaluating and Monitoring the Performance of AI Solutions 301

Chapter 9. Governments as Regulators: Ensuring Trustworthy AI to Manage the Concentration of Power 307

  Main messages 307

  Introduction 308

Trust is needed to ensure AI lives up to its promise 308

  Voluntary standards are a powerful starting point to build trustworthy AI, but market forces alone may not be enough 310

  There is no single best-practice approach to AI governance 311

  AI governance begins with an assessment of existing laws, rules, and regulations 316

  Closing governance gaps involves addressing harms of AI 318

  Governments need strong enforcement and compliance capacity to avoid having rules that govern on paper without providing safeguards in practice 322

  International cooperation gives developing countries greater leverage when acting alone is not enough 324

  Notes 330

  References 332

Spotlight 7. Overcoming Government Fragmentation in AI Policy 337