Taming Silicon Valley: How We Can Ensure That AI Works for Us

Gary Marcus frames the current AI race as a managerial and institutional challenge rather than a question of whether AI should exist. The technology may accelerate science, medicine, and engineering, but it can also amplify misinformation, cyber threats, political manipulation, and social instability faster than public safeguards can adapt.

The book criticizes a market race in which unstable systems may be released before their risks are understood and argues that voluntary self-regulation does not adequately protect data, employment, or trust in institutions. Key governance gaps include lobbying that delays enforceable rules, opaque models and training data, and unclear responsibility when a deployed system causes harm at scale.

Marcus calls for stronger data rights, independent oversight before and after release, transparency about limitations and failure modes, tax reforms, alignment with human rights, and civic pressure for enforceable accountability. Engineering leaders can translate that agenda into risk review, red-team exercises, data-quality controls, explicit prohibited uses, escalation paths, and named owners within the delivery lifecycle.