Governing AI A Primer (Onur Bakiner)（Z-Library）

【One-Line Pitch】 A clear-eyed primer on how societies try—and often fail—to steer artificial intelligence, arguing that technical fixes, corporate self-regulation, and even laws are not enough unless ordinary people get a real seat at the table. Best for students, policymakers, journalists, and tech professionals who want a comparative, evidence-based map of AI governance rather than hype. 【Book Arc】 - **Opening (~0%–10%)**: Frames the core problem—AI risks and harms are not purely technical but rooted in social inequality and disempowerment—and previews the three governance models (technical solutions, business self-regulation, legal regulation) that the book will assess. - **Early (~10%–30%)**: Builds the conceptual foundation: what "AI" actually means, the OECD definition, the shift from rule-based symbolic AI to data-driven machine learning, neural networks, deep learning, and transformer architectures, plus definitional controversies over hybrid AI, anthropomorphism, and AI-as-marketing-hype. - **Early–Middle (~30%–45%)**: Surveys the vocabulary of "good AI"—ethics, responsible, safe, trustworthy, transparent, explainable, just, accountable, sustainable, robust, inclusive—and shows how these terms overlap, conflict, and often lack clear definitions or named beneficiaries. - **Middle (~45%–60%)**: Examines technical solutions such as content detection, human-in-the-loop, testing, de-biasing, and AI-for-social-good projects, concluding that because the roots of AI harm are social, technical fixes alone cannot deliver transformative change. - **Late (~60%–85%)**: Turns to business self-regulation—voluntary principles, codes of conduct, internal and external ethics boards, responsible AI teams at companies like Axon, Google/DeepMind, Meta, and Microsoft—and finds mixed evidence on their effectiveness. - **Ending (~85%–100%)**: Assesses legal regulation and its limits, including regulatory capture, neglected harms to the environment, workers, and competition, and stalled efforts on lethal autonomous weapons, closing with a call for democratic participation in AI governance. 【Key Takeaways】 - **AI governance is not just a technical problem** (Opening): The book insists that where AI fails, human societies fail—racism, sexism, inequality, and eroded trust in institutions shape AI outcomes more than code does. - **Definitions matter for regulation** (Early): Disagreements over rule-based vs. data-driven vs. hybrid AI, anthropomorphism, and hype affect how lawmakers and the public understand what they are governing. - **Today's AI is powerful but limited** (Early): Machine learning is an impressive application of statistics, calculus, linear algebra, and network theory—not human-like contextual understanding—and treating it otherwise distorts policy. - **"Good AI" vocabulary is crowded and vague** (Middle): Ethics, responsible, trustworthy, transparent, and fair are used inconsistently; "trustworthy AI" often fails to specify whose trust matters, and some scholars argue AI can only be reliable, not trustworthy. - **Technical fixes have a ceiling** (Middle): Detection, human-in-the-loop, testing, and de-biasing can help, but they cannot address harms whose roots are social and structural. - **Self-regulation yields mixed results** (Late): Voluntary principles and ethics boards have proliferated since the mid-2010s, but evidence on their accountability impact is inconclusive and varies by company. - **Legal regulation is underdeveloped and vulnerable** (Ending): There are many bills but few actual AI laws; regulatory capture by industry is a real risk, and systemic harms to the environment, workers, and competition receive almost no legal attention. - **Ordinary people need a seat at the table** (Ending): The long-term health of technology-society relations depends on empowering the public to make informed governance decisions, not leaving them to corporations and governments alone. 【Reading Tips】 - **Deep-read the opening and closing chapters**: They carry the book's central argument about disempowerment, alienation, and democratic participation; the middle chapters are more survey-like and can be skimmed for concepts. - **Use the definitional chapters as reference**: The early material on AI terminology and the "good AI" vocabulary is dense but valuable—return to it when a later chapter invokes a contested term. - **Track the three governance models**: As you read, note how each chapter tests technical solutions, self-regulation, and law against real harms; this structure is the book's analytical spine. - **Read the company case studies critically**: The ethics-board examples (Axon, Google/DeepMind, Meta, Microsoft) are useful evidence, but ask who holds power and what enforcement exists. - **Keep the comparative lens active**: The book spans multiple countries, sectors, and solution concepts; pay attention to which jurisdictions and harms get attention and which are neglected. 【Coverage Limits】 This guide is based on stratified excerpts covering roughly the first half of the book in detail, with later chapters represented mainly through outline and summary material; specific case studies, country comparisons, and the full list of AI laws and bills are not covered in depth here.