Authority Without Standing

In the search results for "AI ethics," one organization ranks for more keywords than any other single entity, with estimated organic traffic considerably above the rest of the top-10. The organization is Partnership on AI. It is not a research lab, not a regulator, not a standards body in any legally binding sense, not a university, not a journalistic outlet. It is a multi-stakeholder nonprofit founded in 2016 by Amazon, Apple, DeepMind, Facebook, Google, IBM, and Microsoft, with later additions including civil society organizations and academic institutions. Its function is convening. Its output is frameworks, recommendations, and convening statements.

By the most legible empirical measure available — organic search visibility — Partnership on AI is the most authoritative voice on AI ethics that a general-audience searcher in the United States is likely to encounter. The voice precedes the labs, the regulators, the academic ethicists, and the journalists in the search-discovery layer. It is the first thing the searcher meets.

This piece is about what that authority is and is not. It is also about a longer question: what it means for an organization to have authority in a space without having standing in it.

A distinction that matters

The words authority and standing are sometimes used interchangeably. They are not the same thing.

Authority is the capacity to have your statements treated as worth attending to. It is conferred by institutions, by reputation, by visibility, by the deference of audiences. Authority is largely a matter of position — where you sit relative to the channels through which information reaches the people deciding what to think.

Standing is something else. Standing is the relationship between the speaker and the matter at hand. A speaker has standing when their relationship to the subject — through expertise, through stakes, through having been affected, through having done the work — gives their statements a particular weight that the statements would not have if produced by someone in a different relationship to the matter.

A clinician treating patients has standing on questions about clinical practice in a way that a health policy researcher does not, regardless of how authoritative the policy researcher might be in the broader discourse. The clinician has stakes the researcher does not have, has been responsible for outcomes the researcher has not been responsible for, has had to make the actual decisions the researcher writes about. The clinician's standing is not a function of where they rank in the discourse. It is a function of where they sit in the work.

The distinction matters because authority and standing can come apart. An organization can have high authority and low standing, or high standing and low authority. The combinations are not coextensive. The popular-discovery layer of search results is more sensitive to authority than to standing, because search algorithms cannot easily measure standing. They can measure citations, backlinks, time-on-page, domain rank. They cannot measure whose stakes are most engaged.

Partnership on AI's authority

Partnership on AI's authority is real and is not in question here. The organization's authority rests on several pillars.

The first pillar is its founding membership. Being founded by Amazon, Apple, DeepMind, Facebook, Google, IBM, and Microsoft confers a particular kind of authority — the authority of being convened by the entities whose decisions matter most for the subject. Whatever Partnership on AI says about AI ethics has been said with the participation of the organizations producing and deploying the relevant AI. The participation itself is the authority signal. The reader can assume that the framework being articulated has been shaped by, and is broadly acceptable to, the entities the framework is supposed to govern.

The second pillar is its broadened membership over time. Civil society organizations (ACLU, EFF, Open Society Foundations affiliates) and academic institutions joined later. The broadening is itself an authority signal: it says that the framework is not merely industry self-regulation, that the convening has included voices outside the founding companies. The breadth of the tent confers a kind of legitimacy the founding companies alone could not.

The third pillar is its sustained production of artifacts. Partnership on AI publishes recommendations, frameworks, white papers, and convening reports on a regular cadence. The artifacts exist. They are linkable. They are citable. Other organizations cite them. The citation graph compounds. Over years, the organization accumulates a substantial body of work that journalists, researchers, and policy aides reference when they need to ground a statement about AI ethics in something more solid than their own opinion.

The fourth pillar is its convening function in practice. Partnership on AI hosts workshops, runs working groups, and produces dialogues that the participants treat as substantive. The participants include people whose participation is itself an authority signal — senior researchers from the labs, senior staff from regulators, senior figures from civil society. The convenings produce relationships that travel into other settings. The relationships are part of the organization's value, even when the artifacts of any particular convening are thin.

All four pillars are real. None of them is being criticized here. They explain why Partnership on AI has the search visibility it has. They are the legitimate basis of the organization's authority.

Where the standing question comes in

The standing question is a different question. It asks: in what relation does Partnership on AI stand to the matters its frameworks address?

The matters are, in the broadest framing, the questions of how AI is built, how it is deployed, who is affected, what should be required of those building and deploying it, and what recourse should be available to those affected. These are the substantive questions of AI ethics.

Partnership on AI's relation to these matters is convening. It does not build AI; the founding labs do. It does not deploy AI; the deploying companies do, many of which are members. It is not affected by AI in the way workers, patients, students, defendants, gig workers, or surveilled populations are affected. It does not have the legal authority to require anything of its members; participation in its frameworks is voluntary. It does not have a regulatory mandate to enforce.

The convening function is real and useful. It is also a particular kind of relation, with particular limitations.

When you convene, your continued ability to convene depends on your participants continuing to participate. If you take positions that the most powerful participants find objectionable, those participants may leave or, more subtly, may reduce their engagement. The convening function thus exerts a structural pull toward positions that are acceptable to the most powerful participants. This is not a moral failing of any individual convener; it is a property of the convening role.

The pull does not mean convening organizations cannot take strong positions. Some do. It means that the strong positions tend to be the ones the powerful participants either agree with or are willing to tolerate. The strong positions that the powerful participants would find intolerable are not produced. The convener's authority is, in part, the price of not producing those positions.

This is what we mean by saying Partnership on AI has authority but limited standing on the substantive AI ethics questions. The organization speaks with the voice of the convened. The voice has weight. The weight is purchased by the constraint of speaking in ways the convened can continue to participate in. The constraint is structural, not personal.

What this means for the discourse

Several things follow.

The first is that Partnership on AI's frameworks should be read as artifacts of a particular kind of conversation — a conversation among entities with substantial overlapping interests and substantial differing interests, mediated by an organization whose continued existence depends on the conversation continuing. The frameworks are real outputs of real conversations. They are not the same as the position someone would take who was free of the conversational constraint.

The second is that the absence of certain positions from the popular-discovery layer of AI ethics search results is not random. The positions absent are the ones the convening function structurally suppresses. The labs' more aggressive critics, the workers whose jobs are restructured by AI deployment, the affected populations whose stakes are most acute, the academics whose analyses do not lead to acceptable-to-all-parties frameworks — these voices have lower authority in the popular-discovery layer because they are not what the high-ranking organizations produce. The voices exist; they rank elsewhere or do not rank at all. The popular-discovery layer's framing reflects the convening function's constraints.

The third is that thinking clearly about AI ethics, as a general-audience reader, requires distinguishing between the authority you encounter first and the standing on the substantive questions. The first thing you encounter is not necessarily the speaker whose stakes are most engaged in the matter. It is, often, the speaker whose institutional position has the most search-visibility infrastructure.

This is not a unique property of AI ethics. The dynamic is general. Search-discovery layers across many topics privilege the convening, the standardizing, and the framework-producing organizations over the directly-affected and the substantively-disagreeing. The pattern repeats in environmental policy, in labor policy, in health policy, in technology policy generally. The convening function produces visible artifacts; the affected populations often do not.

What it would mean to read the discourse with the distinction in mind

For a reader of AI ethics commentary, the practical move is to ask, of any given source: what is this source's standing on the question it is addressing, distinct from its authority?

A research lab has standing on questions about model behavior and limitations in ways that a convening organization does not. The lab built the model, ran the evaluations, knows the failure modes. The convening organization has authority that draws on having included the lab in its conversations, but does not have the lab's direct knowledge.

A worker affected by AI deployment has standing on questions about deployment effects in ways that a corporate AI ethics officer does not. The worker is the population the deployment is affecting; the ethics officer's job is to articulate frameworks the deploying organization can adopt. The officer has authority; the worker has standing.

A regulator has standing on questions about what can be enforced in ways that an advocacy organization does not. The regulator can require things and impose consequences. The advocacy organization can recommend. Both may have authority. Only one has direct standing on the enforcement question.

The distinction does not collapse all questions into ones about standing. Authority matters. Synthesis, framing, and convening are real intellectual labor and produce real value. The point is that the value they produce is a different value than the value produced by direct engagement with the matter at hand. Reading the discourse well requires noticing which kind of value is being offered, and not mistaking one for the other.

What this publication's claim is

The Human and I has neither the authority of Partnership on AI nor the standing of a worker directly affected by AI deployment. It has something narrower: a particular vantage on AI work as it is actually done in a small operation, plus the willingness to take positions that the convening function would not produce.

The vantage is partial. It does not substitute for the lab's standing on model behavior, or the worker's standing on deployment effects, or the regulator's standing on enforcement. What the vantage offers is a position from which the substantive AI ethics questions can be engaged without the constraint of acceptable-to-all-participants framing. The constraint is what produces the popular-discovery layer's particular shape. Operating outside the constraint produces something different.

Whether what we produce has value depends on whether the something different is useful. We do not assume it is. We do think the distinction between authority and standing is one of the most important conceptual tools for reading AI ethics commentary honestly, and that the discourse is healthier when readers have the tool than when they do not.

The popular-discovery layer will continue to look the way it looks. The labs will continue to be absent, the convening organizations will continue to be present, the affected populations will continue to rank elsewhere or not at all. Within that landscape, this publication's bet is that some readers are looking for the substantive engagement the layer's structure forecloses, and that the engagement is worth producing whether or not it ever ranks well.

Authority is one kind of warrant. Standing is another. The discourse benefits when both are visible.

The Foragentis team publishes The Human and I from outside the AI labs, outside the major convening organizations, and outside the consulting layer that serves the deploying customer. The positioning has costs and has the freedom that comes with the costs. ForIntel, our intelligence research catalog, applies the same analytical posture to market and policy intelligence on AI deployment.