The Ready Reference for “Rebooting AI: Building Artificial Intelligence We Can Trust”
In their important book Rebooting AI, AI experts Gary Marcus and Ernest Davis make a compelling case that artificial intelligence needs a fundamental rethinking to build more robust, ethical, and human-aligned systems we can trust.
In their insightful 2019 book Rebooting AI: Building Artificial Intelligence We Can Trust, leading AI researchers Gary Marcus and Ernest Davis compellingly argue that the field of AI needs a major rethinking and course correction to develop trustworthy systems that align with human ethics, values and goals.
They provide an incisive analysis of current limitations in dominant AI approaches and offer thoughtful ideas for building more beneficial AI.
Marcus and Davis contend that today’s mainstream AI, powered overwhelmingly by deep learning algorithms, is too narrow, opaque and unreliable for applying to many real-world tasks.
While deep learning excels at pattern recognition in domains like image classification, its models are prone to brittleness, vulnerability to adversarial attacks, baked-in biases, and a lack of transparency that makes their reasoning inscrutable.
The authors advocate developing hybrid AI systems that thoughtfully integrate the strengths of deep learning for perception with symbolic AI approaches that enable robust reasoning, causal understanding, knowledge representation and judgment.
They envision Lifelong Learning systems that build conceptual models of the world through common sense reasoning and an integration of learning approaches.
Throughout the book, Marcus and Davis provide insightful analysis on issues of ethics, bias and transparency in AI while balancing concerns about risks with optimism about the field’s potential for good.
They underscore the urgent need for extensive AI safety research and testing protocols to ensure systems behave reliably in the messy real world.
The authors explore promising ideas like inductive programming to automatically learn understandable rules and theories from examples and experience, much as humans implicitly do.
Marcus and Davis argue today’s mainstream AI still lacks many facets of human cognition that will be critical for creating systems that comport with social norms and human values — capacities like sound reasoning, theory building, understanding causality, and employing contextual knowledge about the world.
They are critical of hype about imminent human-level AI and provide a more realistic roadmap for incremental progress on the long path toward advanced general intelligence.
Rebooting AI highlights the need for AI to be made significantly more transparent, robust and aligned with human wellbeing before being deployed in high-stakes social contexts like healthcare, law enforcement, governance and more.
Marcus and Davis make a thoughtful case for building AI that augments human capabilities and enhances our collective wisdom rather than replacing human roles and agency. Their book provides an essential rethinking of how to achieve beneficial AI through a hybrid approach.
This site contains affiliate links, which means I may earn a commission if you purchase products or services via the links provided.
This post was created with the help of AI tools.
