“To what extent should our lives be directed and controlled by powerful digital systems—and on what terms?” That question, posed by Jamie Susskind in Future Politics, sits at the center of modern life as artificial intelligence moves from novelty to infrastructure. It is an apt provocation: a single prompt can now produce a personalized message, a targeted political pitch, or an automated decision that affects benefits, employment, or public safety. The choices we make about how, where, and by whom AI is used will determine whether those capabilities amplify opportunity or entrench harm.
AI is not merely faster software. It is a multiplier of agency: for managers, marketers, campaigners, and everyday users, it can instantly draft, tailor, and distribute content across channels to thousands or millions. That power is double-edged. On one hand, properly designed systems can speed service delivery, reduce fraud, and free human workers for higher-order tasks. On the other, poorly governed deployments can reproduce bias, harden opaque decisions, and enable adversaries to weaponize personalization at scale. The trade-offs are technical, institutional, and ethical—and they are already playing out across government and industry.
Technologists argue the answer is engineering: better data pipelines, explainability features, continuous monitoring, and adversarial testing. NIST’s AI Risk Management Framework is often cited as a pragmatic guide for building governance, measurement, and communication into procurement and deployment practices, helping practitioners create audit trails and testability rather than black boxes. These are concrete mitigations for predictable technical failures and biased outputs .
Policymakers face a different calculus. They must balance innovation and public interest: speeding up procurement to modernize public services while protecting civil liberties and national security. Oversight bodies such as the Government Accountability Office have repeatedly warned about the risks of procuring off-the-shelf AI without sufficient governance; Congress and agency leaders are wrestling with acquisition rules that were not designed for rapidly evolving algorithmic systems. The unresolved tension is clear: move too fast and risk systemic harm; move too slowly and risk stagnation of vital public services .
Design and deployment choices matter for users. In ideal implementations, AI augments human judgment—triaging medical cases to improve throughput, streamlining benefits processing, or routing emergency response more efficiently. In poor implementations, automated systems can bottleneck appeals, misclassify eligibility, or make decisions that are difficult to audit or contest. The public’s trust in government services will hinge on transparency, redress mechanisms, and the visible alignment of AI outcomes with human values .
Adversaries complicate the equation. Criminals and hostile states view AI both as a tool to be used and a system to be attacked, making resilience and security non-negotiable. Integrating AI into defense, intelligence, and critical infrastructure introduces new vectors for disruption. Security demands—from hardened edge sites to secure telemetry and redundancy—must accompany any large-scale adoption to prevent malicious exploitation and cascading failures .
The practical choices organizations face fall into several categories:
- Data and governance: Prioritize data modernization alongside pilots. Models are only as reliable as the data they ingest; rigorous audits and remediation mechanisms are essential to prevent amplification of existing disparities .
- Human-centered workflows: Invest in workforce retraining and hybrid systems that keep humans in the loop for high-stakes decisions. That preserves accountability and leverages human judgment where nuance is required .
- Standards and interoperability: Adopt and operationalize standards—such as those developed by NIST—to enable predictable sharing of context without exposing sensitive information, and to make systems testable and certifiable over time .
- Security and resilience: Fund adversarial testing and security reviews from project inception, and design architectures that support graceful degradation and rapid recovery in the face of attack or failure .
- Iterative governance: Favor adaptive, iterative regulatory approaches—regular audits, certifications that evolve with technology, and public reporting—over rigid rules that can rapidly become obsolete .
Real-world pilots reveal both promise and peril. City traffic systems using AI-enabled sensors show measurable gains in flow and incident response, yet raise privacy concerns; hospital triage tools can improve throughput but provoke liability and acceptance issues among clinicians. These examples underline a central lesson: success depends not on raw algorithmic power alone but on coherent alignment across technical, infrastructural, and human layers .
Different stakeholders see different risks. Technologists emphasize robustness and explainability. Policymakers worry about civil rights, procurement speed, and national security. Users demand clarity and the ability to contest decisions that affect their lives. Adversaries see vectors to corrupt or weaponize the same mechanisms intended to help. Bridging these perspectives requires institutional redesign as much as technical fixes—procurement pathways that are agile but accountable, oversight mechanisms that can keep pace, and a public conversation about acceptable trade-offs.
There are no silver bullets, only choices. The sensible path is iterative: start with small, well-governed deployments that generate empirical evidence about harms and benefits, scale what proves safe and useful, and keep robust redress and audit mechanisms in place. Speed without oversight risks long-term liabilities; excessive caution risks leaving services inefficient and people underserved. The right balance will vary by mission—predictive maintenance in logistics is a different risk profile than automated eligibility decisions for social benefits .
If the question for the last century was how much of our collective life should be determined by state versus market, the question now is whether we will let powerful digital systems decide for us, and under whose authority. The stakes are not abstract: they touch elections, livelihoods, civil liberties, and national security. The policy, technical, and civic choices we make will determine whether AI becomes an instrument of public flourishing or a force that deepens fragility and mistrust.
So what should we demand? Transparent procurement and auditability, iterative governance that evolves with technology, human-centered design that preserves contestability, and security investments that anticipate adversaries. Those are not easy choices, and they will not satisfy every stakeholder. But they are necessary if we are to keep control over the systems increasingly shaping our lives.
In the end, the question is as simple—and as profound—as Susskind’s: who decides, and on what terms? If we leave those decisions to opaque systems or to the fastest private actors alone, we cede a form of collective agency that is hard to reclaim. That is the risk worth confronting now, before the architectures of control become the architectures of default.
Source: https://www.schneier.com/blog/archives/2025/12/like-social-media-ai-requires-difficult-choices.html
