"A mistake we keep repeating is arguing about what machines can do instead of what values they encode." A member-only essay on code.likeagirl.io uses that line to reframe current AI debates away from raw capability and toward the social choices built into deployed systems. The author frames the argument with a personal memory of arriving in New York a few weeks before 9/11 and with the TV series Person of Interest, using both to show how automated systems sort people into risk-based groups. Practical takeaway: shift from whether models can do tasks to which tasks they should do, and under what constraints and accountability.

"A mistake we keep repeating is arguing about what machines can do instead of what values they encode," the member-only essay on code.likeagirl.io begins, and that single line sets the article's claim as plainly as any statistic. The writer anchors the argument in a personal recollection: arriving in New York a few weeks before 9/11, then watching how security and policy changed in the months that followed. What mattered, the essay says, was not only the immediate inconvenience of new checks and restrictions. It was the lasting sense of being treated as a category rather than as an individual.

From personal memory to institutional habit

The essay uses those memories to show a pattern institutions repeat when faced with uncertainty. Faced with ambiguity, systems expand classifications, increase oversight, and search for patterns to manage perceived risk. The writer argues institutions prefer procedures that privilege pattern-finding and broad classification over individualised judgment. That logic, the piece warns, standardises responses, flattens human complexity, and hardens social categories into administrative fact.

To make the point vivid, the author borrows a popular narrative. Person of Interest, the television series, isn't offered as a policy blueprint but as an illustrative narrative. In the show a machine scans vast amounts of data to flag persons of interest. That fictional device, the essay says, mirrors real-world tendencies in surveillance, classification, and preventive intervention: automated systems draw lines around people, sort them into risk-based groupings, and invite institutional responses that treat those groupings as if they were fixed realities.

Design choices matter more than benchmark scores

The central analytical claim is blunt: future outcomes will be shaped less by raw model performance and more by the values encoded in system design and deployment. When organisations entrust decisions to large-scale systems, those systems inherit both explicit rules and implicit assumptions about which traits matter. The essay presses that point against the common tech debate that foregrounds model size, accuracy, or benchmark performance. Technical benchmarks matter, the author concedes, but they're not the decisive variable for social effect.

Instead the piece calls for attention to the normative trade-offs embedded at the design stage. Who's classified as risky? Which proxies are acceptable when direct measures are unavailable? What oversight procedures exist when the system errs? These are the questions that decide whether a deployed system amplifies respect for individual dignity or reduces people to data points.

Mastheads

The author warns that emergency-driven deployments of algorithmic tools often follow a familiar arc: temporary measures become routinised, producing long-term shifts in how people are monitored and regulated.

That procedural problem is the essay's through-line. Using the post-9/11 experience, the writer shows how measures introduced in crisis can harden into standard practice. Applied to algorithmic systems, the lesson is stark. If designers and policymakers don't explicitly interrogate the values a system embodies, the system's classification logic will calcify institutional behaviour and alter daily life at scale. The essay doesn't offer a single policy fix; it stresses that design choices, institutional incentives, and legal frameworks together determine outcomes.

The author links this institutional tendency to debates about AI safety and governance without reducing those debates to technical metrics. Safety and governance do include questions about model robustness and reliability, but they must also treat classification logic and the choice of operational tasks as central. The most concrete takeaway the essay gives is a reorientation of priorities: move from asking whether a model can do a task to asking which tasks it should do, under what constraints, and with what accountability mechanisms.

That shift is urgent, the essay argues, because the social consequences of embedding classification logic at scale are already visible across domains from security to social services. Where automated systems intervene, the choices made at design time ripple through procurement, legal responsibility, and everyday administrative practice. If those choices are left implicit, the system's encoded values will become the default. If they're made explicit, designers and leaders have a chance to steer a different path.

Person of Interest returns at the essay's edges as a narrative device: it's there to make abstract institutional patterns recognisable. The show crystallises how a fictional machine's sorting logic reshapes ordinary life. The writer's memory of New York a few weeks before 9/11 does the historical work: it shows how emergency measures recalibrate institutions and individual experience long after the immediate threat fades.

For readers in tech and policy, the piece offers a clear disciplinary nudge. Spend less time arguing about capability in isolation.

Spend more time naming the tasks we want machines to take on, clarifying acceptable proxies, and building accountable oversight into deployments. Those aren't purely technical problems; they're social and political choices that design teams and regulators must make together.

Related Articles

The essay's final instruction is concrete: stop asking whether a model can do a task and start asking which tasks it should do, under what constraints, and with what accountability. Originally reported by code.likeagirl.io.

This article was created with AI assistance.