Alondra Nelson

In the [YOU] on AI Field Guide

The cycle that began with [YOU] on AI asks what it would mean to see the machine clearly, without hype or paralysis. Nelson answers a prior question the cycle cannot avoid: even if we see it clearly, who decides what to do about it? Her lens is not pessimism about the technology but precision about power. Where most AI commentary asks what these systems can do, she asks who was in the room when the foundational choices were made—about training data, objective functions, safety definitions—and whether the people most affected by those choices had any standing to contest them.

Her concept of thick alignment reframes the cycle’s central stakes. Algorithmic governance now touches every institution that shapes daily life, and Nelson’s argument is that “alignment” is too thin a word if it means only matching a model to its makers’ intentions. A thickly aligned system is one whose builders’ wants have been subjected to scrutiny by the broader public whose lives are at stake. Thin alignment can be achieved inside a company. Thick alignment cannot be, by definition, because it requires the participation of people who do not work there and were never consulted.

She stands in the cycle’s gallery as the thinker who insists that the question of governance is not downstream from the technical work but coextensive with it. Where sociotechnical imaginaries shape which futures get built, Nelson argues that those imaginaries are produced by a narrow slice of humanity and that expanding who authors them is not a soft inclusion gesture but a structural requirement for building systems that serve rather than surveil. Her Afrofuturist inheritance adds a generative dimension: the future is not delivered to us from a technological elsewhere; it is made, and those who claim the right to imagine it are the ones who will build it.

The cycle asks whether these tools can belong to all of us. Nelson’s answer is: only if all of us insist on a hand in shaping them. The deepest question AI poses is not what it can do to us but what we are willing to become in the way we choose to build and govern it—and that question is finally not a technical one but a democratic one.

Origin

Nelson built her reputation through empirical work that had nothing to do with computers. Studying the Black Panther Party’s free health clinics, she watched a marginalized community treat medicine itself as a political terrain—not a neutral science to be accessed but a system of power to be contested. Studying genetic ancestry testing, she watched DNA, that supposedly objective molecule, become the carrier of meanings about identity, belonging, and historical injustice that no sequencing machine could detect. Each case taught the same lesson from a different angle: the technical artifact and its social meaning are not two things. They are one thing seen from two sides.

When she moved into AI policy, she was not inventing a new vocabulary. She was carrying a hard-won one into a new domain. The conviction that science is never neutral, that its benefits and harms fall unevenly along the fault lines of race and class and gender, that the people most affected by a system deserve a voice in its design—these were the load-bearing beams of her scholarship long before the phrase “large language model” entered the language. She brought them to the White House, where the team she led produced the Blueprint for an AI Bill of Rights in October 2022: five principles articulating what the public should be entitled to expect as automated systems are woven into the institutions that govern daily life.

Key Ideas

Thick alignment. Nelson’s most pointed intervention into the AI discourse is the distinction between thin and thick alignment. Thin alignment asks whether a system does what its builders intend—a technical question answerable inside a company. Thick alignment asks whether what the builders intend is itself aligned with the values, contexts, and lives of the people the system will touch. On that question, the affected public is also an expert, and a field that lets engineers answer it by default has mistaken a part for the whole. The distinction reframes every debate about sociotechnical safety: the hard problem is not how to make the model reliably pursue the specified objective but who specifies the objective and by what legitimate process.

The social life of technology. A model is trained, evaluated, and shipped with a particular set of intended uses. But the moment it enters the world it begins to acquire a social life its designers neither anticipated nor controlled. A hiring algorithm becomes a gatekeeper that reshapes who gets to work. A facial recognition system becomes an instrument of surveillance that lands differently on different communities. Nelson’s method, developed through her study of genetic ancestry and medicine, is to follow the technology into the world and report honestly on what it becomes there. The social life is not noise to be filtered out so the real signal can be studied. It is the signal, and it is largely invisible to the metrics by which the field measures itself.

Data as a record of power. The field treats data as raw material—neutral, abundant, to be refined into intelligence. Nelson sees something different. A dataset is a fossil of the society that produced it, preserving the contours of that world including its injustices. When a decision is made by an algorithm trained on such data, its bias acquires a veneer of mathematical impartiality that makes it harder to see and harder to contest. The system did not discriminate, the argument goes; it merely found patterns in the data. But the patterns are the residue of past discrimination, and a system that faithfully reproduces them is laundering historical injustice through the authority of computation.

Democracy is not an obstacle. The most persistent adversary in Nelson’s thinking is a story about technology—that progress is inevitable, that it moves on rails, and that the public’s role is simply to adapt. She has named this fallacy repeatedly and countered it with a democratic claim: the outcomes many people hope for from AI are not inherent features of the technology. They are choices a society must fight to make. A determined future has no need of citizens. A contingent one cannot do without them. Technologies which lack public legitimacy are fragile; systems imposed without consent generate the kind of resistance and backlash that ultimately undermine the very deployment they were meant to accelerate.

The Afrofuturist inheritance. Before she worked on AI policy, Nelson was one of the scholars who helped establish Afrofuturism as a serious field of inquiry—the practice of insisting that people of African descent are not merely the subjects of technological history but its authors. The framework she pushed against held that Black communities stood on the wrong side of a digital divide, perpetually behind. Nelson argued this framing did a subtle violence, constructing Blackness as inherently oppositional to technological progress. Afrofuturism was a refusal of that story. It treats the future as material to be worked—as something that can be imagined otherwise and therefore built otherwise. This is the disposition Nelson brings to AI: neither the booster’s certainty that the technology will save us nor the doomer’s certainty that it will destroy us, but the maker’s conviction that the outcome is open and that imagination is a form of power over it.