Rebooting AI

Rebooting AI critiques modern artificial intelligence's overreliance on narrow machine-learning systems and advocates for building AI with robust commonsense reasoning. The book argues current approaches (like deep learning) fail in open-ended environments, proposing hybrid models that integrate cognitive science to create trustworthy, human-aligned AI.

This book is essential for AI researchers, tech policymakers, and readers interested in AI’s societal impact. It offers technical insights for professionals and accessible critiques for general audiences concerned about AI’s limitations in healthcare, autonomous vehicles, and decision-making systems.

Yes—it’s a vital counterpoint to AI hype, emphasizing systemic flaws in current models. Marcus and Davis provide actionable solutions for developing AI that adapts to real-world complexity, making it valuable for anyone seeking a balanced understanding of AI’s capabilities and risks.

The authors highlight AI’s fragility in unstructured environments, overreliance on big data, and lack of causal reasoning. They argue systems like deep learning excel in controlled tasks (e.g., board games) but fail at contextual understanding, leading to errors in healthcare diagnostics or self-driving cars.

The book advocates hybrid architectures combining neural networks with symbolic reasoning and cognitive models. Solutions include prioritizing causal inference, lifelong learning, and transparency to reduce errors in critical applications like medical diagnosis.

Marcus and Davis identify commonsense reasoning—interpreting context, cause-effect relationships, and tacit knowledge—as AI’s missing foundation. They argue systems without this cannot safely navigate real-world unpredictability, like domestic robots handling unfamiliar objects.

It shifts focus from sci-fi “superintelligence” risks to immediate dangers of flawed systems making catastrophic errors (e.g., misdiagnosing patients). The authors stress rigorous testing and ethical frameworks to prevent reliance on unreliable AI in high-stakes scenarios.

The “AI Chasm” refers to the gap between current narrow AI (specialized tasks) and general intelligence. Marcus and Davis attribute this to inadequate cognitive modeling, unrealistic expectations, and overconfidence in data-driven approaches.

While Stuart Russell’s Human Compatible focuses on aligning AI with human values, Rebooting AI emphasizes rebuilding AI’s cognitive foundations. Both agree on avoiding speculative risks, but Marcus/Davis prioritize fixing today’s unreliable systems over long-term existential threats.

“As long as the dominant approach is focused on narrow AI and bigger data sets, the field may be stuck playing whack-a-mole indefinitely.” This underscores their call for paradigm shifts toward hybrid, explainable AI systems.

Despite AI advancements, core challenges like hallucination in LLMs and autonomous vehicle failures persist. The book’s warnings about data-centric limitations remain pertinent, offering frameworks for addressing reliability gaps in modern generative AI.

Gary Marcus (NYU cognitive scientist, founder of Robust.AI) and Ernest Davis (NYU computer science professor) combine decades of research in human cognition and AI ethics. Their interdisciplinary expertise underpins the book’s critiques of machine learning’s shortcomings.