Deep learning spots patterns but not meaning; to build trustworthy AI we must marry learning with reasoning and common sense. Marcus and Davis tour the blind spots: vision systems fooled by a sticker; chatbots that string fluent sentences with no grasp of causes and goals; models that crumble when the context shifts.

They sketch a reboot: represent knowledge explicitly, reason over it, and use learning where it shines. That means neurosymbolic architectures, curated commonsense, curriculum‑style training, and tough, diagnostic tests—Winograd‑style puzzles over leaderboard theatre—plus design norms of interpretability, auditability, and robustness.

As AI steers medicine, transport and policy, tricks won’t do. This book offers a practical path from clever demos to systems you’d trust with real decisions.