An AI Future Built on Resilience, Not Restraint

T oday’s artificial intelligence (AI) policy discussions are infused with dystopian dread of the future. A profound pessimism — and corresponding set of extreme fear-based policy proposals — is on regular display from academics, policymakers, and even some leaders in the AI industry themselves.

In recent months, the heads of some major AI labs and their employees have issued warnings in essays, speeches, and joint letters about the disruptive effects of the AI revolution we are living through. Predictions of mass job dislocations and superintelligence run amok are heard regularly, and ambiguous calls for government action follow, including the potential need for undefined global efforts to “pace” the progress of AI capabilities.

Mark Zuckerberg, CEO and founder of Meta Platforms, released a new AI-related manifesto of his own this week that offers a quite different, and refreshing, counter-perspective: a philosophy of technological change rooted in individual empowerment, decentralization, and human adaptability, rather than fear and loathing about the future. In “The Future Is for Everyone,” Zuckerberg highlights the need “to develop a philosophy for how we can best use superintelligence to ensure it improves all of our lives, work, communities, freedom and safety,” built on “the principles of individual empowerment as the source of prosperity.”

That is the right place to begin.

Too much of today’s AI debate starts from an implicit theory of fragility — of humans, institutions, and legal systems. With that as the premise, advances in AI systems are treated as hypothetical catastrophes just waiting to happen. An unsurprising governance mindset emerges from this thinking: Restrict capabilities, slow deployment, centralize oversight, and require innovators to prove perfect safety before allowing innovation. The regulatory toolkit here involves licensing developers and gatekeeping access, through comprehensive national regulatory controls married up with global “alignment” treaties. This is the precautionary principle applied to computation.

The opportunity costs of precaution are mostly invisible, but quite real. Regulators cannot count the discoveries not made, cancers not cured, businesses not started, productivity gains never realized, or defensive technologies never developed because algorithmic innovation was delayed or prohibited. So, they ignore those benefits, preferring to address more visible worries in the present.

There is a better way: Embrace innovation and work toward a resilient response to risk. Allow experimentation to proceed, address demonstrable harms as they emerge, empower individuals and institutions to adapt, and continuously improve defenses along the way. The resilient approach is rooted in risk mitigation — not elimination — using adaptive governance, multistakeholder efforts, professional norms, technical safeguards, and existing law and ex post remedies rather than sweeping, top-down computational controls.

Zuckerberg’s manifesto reflects this approach. His overarching goal is to empower individuals through decentralized access and control. “The notion that AI is so dangerous that the only safe path is an extreme concentration of power seems inherently problematic,” he argues. He is rightly concerned that today’s debate about AI “safety” leads to a world in which everyone forgets Lord Acton’s famous admonition about how absolute power corrupts absolutely. Zuckerberg extends Acton’s wisdom to the AI age, noting that “historically, hoping that an absolute power will benevolently provide for humanity if sufficiently enlightened has not led to safe or positive outcomes.”

Requiring society to solve every technical problem in advance is not only impossible but also leads to a less safe world. We become safer through a messy process of trial and error, learning, and constant correction. Humans have repeatedly demonstrated an uncanny ability to muddle through technological disruption and develop coping mechanisms as knowledge accumulates. Problems emerged; norms adjusted; markets responded; engineers redesigned systems; courts applied existing law; lawmakers filled genuine gaps. Experience is the best teacher, as they say.

This is how resilient societies prosper and grow safer along the way. AI should be governed with the same pragmatism and humility.

That does not mean we should do nothing to address short-term worries. Zuckerberg’s manifesto offers a variety of pragmatic responses to concerns about AI and jobs, cybersecurity, and communities. Some involve practical steps that companies like his can take, while also working collaboratively with others in industry, civil society, and government to develop balanced responses on the fly.

It is a healthier vision than one rooted in top-down control. Better AI safety and security will not emerge from a government-backed conclave of technocratic elites and global bureaucrats working with protected corporate cartels to engage in computational central planning.

Zuckerberg pushes for a world of broadly distributed AI capabilities, multiple competing models, open-source systems, and personal agents serving different human goals. Open and competitive development can be a source of safety by empowering more people and organizations to familiarize themselves with the toolset, improving defenses and developing countermeasures.

This is the resilient response in practice: Harden systems, expand defensive capacity and detection, punish malicious conduct, and adapt as concrete threats become evident. New vulnerabilities will emerge, as they always have. But, as Zuckerberg observes, uncertainty is an argument for more openness and innovation, not less.