It is summer 1945. The Manhattan Project has produced two functional weapons. The bomb is in the box. The question on Truman's desk is whether to use it.
The standard frame for that question is means-and-ends. The end is to compel Japanese surrender; the means is dropping the bomb. Are these means proportionate to that end? Are they justified? The frame asks the operator to evaluate the means against the end as if the two were separable items in a moral ledger.
R.G.H. Siu walks straight at the frame and breaks it. Given a nuclear bomb, an annihilated city is certain to follow; given spacecrafts in the stratosphere, spies in the skies and weapons platforms in the heavens are certain to follow; given computerized memory banks, increased invasion of privacy is certain to follow. Ends realized are nothing more than means expressed.1
Read that last sentence twice. Ends realized are nothing more than means expressed. Once you have built the bomb, the annihilated city is no longer a separate moral question. The annihilated city is what the bomb is. The decision was made when the bomb was built; the decision in 1945 was the unfolding of a decision that had already been made. The means and the end were not separable. The means was the end, in slow motion.
This is Siu's central argument in Op#74, and it is the argument that grounds the only place in The Craft of Power where Siu sets aside operational neutrality and writes a normative claim. There is no way around it. Power is a thoroughly moral phenomenon. It is effective, defective, or deceptive morality — as the case may be.2
The standard amoralist position on power is compact and tempting. Power is just the means; only the ends carry moral weight; therefore the operator's job is to acquire and wield power efficiently and worry about ethics at the level of the goals he chooses. Siu names this position to demolish it.
With some justification, one might be inclined to brush aside the question of the morality of power on the premise that power is but the means to an end and therefore basically amoral.3 The amoralist position has a kind of internal logic. It would hold if means and ends were genuinely separable.
Siu's first move is to point out that the determination as to the kinds of resources, including spiritual motivation and physical coercion, to be used invariably interlaces means and ends. The choice of means is a moral choice, not a neutral selection of instruments after the moral work has been done at the level of ends. Sheer wielding of force is not power. Power begins with a specification of purpose. The expression of power thereby entails a moral choice.4 The moment the operator specifies a purpose, the moral evaluation has already begun. The amoralist frame assumes the specification has happened in some neutral pre-moral moment, and the assumption is wrong.
The second move is the structural one Siu compresses into the means-expressed sentence. In the modern technological environment, the very existence of means capable of achieving certain ends creates a disposition to use those means toward those ends. Means are not docile. Means produce ends. The amoralist who tells himself the operator can choose ends independently of which means are available has not noticed that the available means are themselves the operator's effective option set, and the choice of which means to develop is the choice of which ends will become realizable.
Bertrand de Jouvenal pushed the analysis to its limit. All power, whether for purposes good or ill, is corrupt.5 The corruption is not contingent on the operator's character or intent. The corruption is structural. Once the operator has acquired means at scale, the operator's moral options have already been reduced by the existence of those means. The choice of not using the means becomes itself a moral position that the operator has to actively maintain against the structural disposition Siu names. Most operators do not maintain it. The means get used. The ends follow.
Siu does not endorse Jouvenal in full. He cites him to mark the position. The position deserves to be marked. Many thinkers like Bertrand de Jouvenal have been led to the conclusion that all power, whether for purposes good or ill, is corrupt.6 Siu's own reading is more granular — power is not uniformly corrupt; it is effective, defective, or deceptive morality depending on how the operator runs it. But the granularity does not save the operator from the moral question. The granularity is the moral question. Every act of power is some specific kind of morality being expressed; none is morally neutral.
Siu reaches for a piece of contemporary television to ground the argument in a concrete exchange. A 1969 debate on student riots. The moderate student takes a firm stand against coercive tactics while expressing belief in the movement's objectives. The militant student responds:
"This sounds so much like so many things I know we in the black community have heard a long time. I remember, you know, very interestingly, people used to say to Martin Luther King: 'But is it right to break a law simply because you disagree with it? Do two wrongs make a right?' Now what you wind up saying is 'Well, I agree with the issues you're raising, but I don't necessarily agree with your means,' which means then you don't do anything about the issues . . . Poor people come into Washington, D.C. They said: 'I'm against poverty, but don't mess up the grass and landscape.'"7
The militant student is making Siu's argument from the activist side. Agreeing with the ends while rejecting all available means is operationally identical to opposing the ends. The means are the ends in operational form. The selector who chooses no means is selecting for no ends. The moderate's position is internally inconsistent in the same way the amoralist's position is. Both treat means and ends as separable when they are not.
Siu's positive position is more interesting than either pole he is rejecting. Power is not amoral. Power is also not uniformly corrupt. Power is effective, defective, or deceptive morality.
Effective morality is the rare case where the means deployed produce the ends the operator stated, and both are sustainable at the scale of operation. The Manhattan Project ended a world war; effective morality at that scale would require the means deployed to be proportionate to the ends achieved with no compounding moral costs the operator did not anticipate. Most operators do not achieve this.
Defective morality is the more common case. The operator deploys means with one set of ends in mind and produces additional ends he did not intend. The Manhattan Project produced the Cold War. The Cold War produced the nuclear arms race. The nuclear arms race produced the architecture of mutually assured destruction the world has lived under for eight decades. The means expressed ends the operator could not see at the time of deployment, and the ends were not all the ones the operator chose.
Deceptive morality is the operationally cynical case. The operator publicly states ends he is not pursuing while privately pursuing different ones, deploying means that serve the private ends while claiming they serve the public ones. I'm against poverty, but don't mess up the grass and landscape is the comic-grade version of this. The serious version operates at the scale of foreign policy, monetary policy, and institutional design. The deceptive operator is running a moral system; he is just running it at his own benefit while presenting it as a public good.
The trichotomy refuses both the amoralist and the uniform-corruption frames. Power is a thoroughly moral phenomenon is not a virtue claim. It is a structural claim. Every act of power is some specific kind of morality being expressed; the question is which kind, not whether.
The granularity vs. Jouvenal problem. Siu rejects Jouvenal's uniform-corruption claim and substitutes the trichotomy. The trichotomy is more accurate descriptively. It is also less operationally constraining: an operator who can claim effective morality has more permission to act than an operator who has to admit all power is corrupt. Whether the trichotomy is honest analytical refinement or motivated hedging is not fully resolvable from the spec alone. The honest reading is that both readings have merit and the operator has to decide which one applies to a specific case.
The amoralist position is not actually defeated. Siu's argument requires that means specification entails moral choice. A skeptical amoralist will reply that the moral choice is exhausted by the choice of ends, and the means specification is determined by efficiency considerations downstream of the moral work. The reply is wrong, but Siu does not address it directly. His move is to deny the separability rather than to argue against the amoralist's reduction. The argument is more rhetorically persuasive than logically airtight, and operators who have committed to amoralism are not displaced by it.
The technological-imperative claim has counter-examples. Siu argues that means existence drives end realization — given a nuclear bomb, an annihilated city is certain to follow. The historical record contains counter-examples. Many states have acquired chemical weapons and not used them. Many states have acquired nuclear weapons and used them only twice. The disposition Siu names is real and important but not deterministic, and operators who treat it as deterministic become defeatist about the possibility of restraint.
Psychology — How the Means Stop Looking Like Choices: Power and Psychological Permission — The Milgram studies document a specific psychological mechanism that explains why the means-ends collapse Siu describes is so reliable. When authority structures specify the role, the role-occupant's moral threshold for harming others shifts dramatically. The means becomes available because the role makes it available, and the role-occupant's evaluation of the means happens inside the frame the role has already established. The amoralist position Siu rejects is, in psychological terms, the lived experience of operators who have completed role-capture: the moral evaluation of the means is no longer felt as the operator's evaluation; it is felt as an institutional given that the operator simply executes.
The handshake produces what neither page states alone. The operator who runs the amoralist position is not defending it consciously; he is reporting his lived experience. The means feel separable from the ends because the operator's moral apparatus has been pre-installed by the role, and the apparatus does not ask the question Siu is asking. Siu's argument cannot persuade the role-captured operator because the operator is no longer in a position to hear it. The reach of Op#74 is structurally limited to operators who have not yet completed role-capture, and the operators who most need to hear the argument are precisely the ones the argument cannot reach. This is the silent finding behind the unsettling fact that Siu's normative claim — power is a thoroughly moral phenomenon — has been available since 1978 and has not been operationally absorbed by the institutional power class he wrote it for. The class cannot absorb it because the class has been pre-formatted by the role to not see the question.
History — When Means Specification Becomes Inevitable End: Stalin's Redefinition of Leninism — Stalin's career is a clean demonstration of Siu's means-expressed thesis at state scale. Lenin had specified ends — the worker's revolution, the dictatorship of the proletariat, the transition to communism. Stalin inherited the means — the Cheka, the party apparatus, the centralized economic instruments, the ideological vocabulary that licensed the deployment of all of these. The means expressed the ends. What Stalin produced — the gulag, the famines, the purges, the terror — was not a deviation from the Leninist project. It was what the means expressed once the means were available at scale and the operator was willing to use them. Stalin's redefinition of Leninism was not primarily a doctrinal move; it was the linguistic adjustment required to describe what the means were producing as if it had been the intended end all along.
The handshake produces a finding the orthodox debate over Stalinism mostly avoids. The question of whether Stalin betrayed Lenin or fulfilled Lenin is the wrong question. The means specified by Lenin produced ends Lenin could not have specified, and the operator who inherited the means produced the ends the means contained. Lenin would not have approved of every specific outcome; the disapproval is irrelevant to the structural question. Once the Cheka existed, an apparatus capable of mass terror existed, and the apparatus produced mass terror under any operator who was willing to deploy it. Siu's principle predicts the outcome at the level of the apparatus, regardless of who occupies the seat. The orthodox debate (was Stalin a betrayer or a fulfiller?) is a distraction from the architectural question (does the apparatus produce these outcomes structurally, regardless of operator?). Reading the two pages together moves the debate from the personality of Stalin to the design of the means, and the design of the means is the actual moral question Siu insists is inseparable from the ends. The operator who inherits a means built for purposes he claims to oppose has, in Siu's framing, already inherited the ends those means express, and his disavowal does not prevent the expression.
1. List the means you currently command. It is Tuesday morning. Write down the major capacities your role gives you — budget authority, hiring authority, surveillance access, regulatory leverage, communications channels, networks of influence. Look at the list. Each item is not just a tool; each item is a disposition toward specific ends that your role makes likely. Notice what ends those means tend to produce when used and when not used.
2. Distinguish the ends you claim from the ends your means produce. For each means on your list, ask: what does this means actually produce when deployed at scale? The honest answer is often different from the formal end the means is associated with. The budget authority is associated with allocating resources; what it actually produces is institutional dependencies that constrain future budget decisions. The surveillance access is associated with security; what it actually produces is information asymmetries that change every relationship in the institution. The means express ends that exceed the formal end the means is pointed at.
3. Identify the means-expressed end you would not endorse if it were stated as the goal. If your surveillance access produces, as a side effect, a culture of self-censorship in the institution, ask whether you would endorse creating a culture of self-censorship if it were stated as the goal. If you would not, you are deploying a means whose expressed end you do not endorse, and Siu's principle implies that the deployment is morally non-neutral whether you intended the side effect or not. The means-expressed end is part of what you are doing.
4. Decide whether to keep, modify, or retire the means. The honest options for a means whose expressed end you do not endorse are three. Keep it and own the expressed end. This is the operationally honest amoralism. Modify it to express different ends. This is the operationally engaged moralism — change the architecture of the means so the ends it expresses are ones you would endorse. Retire it. This is the rare case where the means cannot be modified into ends-expression you can defend, and the only honest move is to give up the means.
5. Avoid the I'm against poverty, but don't mess up the grass pattern. When you find yourself defending an outcome by claiming opposition to the means that produce it, you are running the pattern Siu's militant student named. The pattern is operationally identical to opposing the outcome. If you genuinely want the outcome, you have to engage the means; if you do not engage the means, you do not get to claim the outcome.
6. Run the audit annually. Means change. Roles change. Ends-expression patterns change with both. The annual audit is whether the means you currently command are expressing ends you would still endorse if the ends were stated explicitly. If they are not, the corrective action is one of the three options in step 4. Most operators run the audit once at the start of a role and never again, and the means slowly start expressing ends the operator would not have endorsed at the start.
The institutional class spends most of its energy on the means it commands and very little of its energy on the ends those means express. This is not laziness; it is the structural consequence of role-capture, and Siu's analysis predicts it. The operator who has internalized the role no longer sees the means as morally evaluable; he sees them as the institutional environment within which his moral evaluation happens. The implication is that the most morally consequential decisions of the operator's career are made implicitly, by the architecture of the role he accepted, and not explicitly, by the choices he makes within the role. The interview at which he was hired, the position he applied for, the institution he joined — these are the moments at which his future moral exposure was set, and most operators do not realize this until decades later, after the means they have been running have expressed ends they would not have endorsed if asked at the start. The honest implication is uncomfortable: the moral life of the institutional operator is determined more by which institution he joined than by what he did once inside it. The means were chosen at the door.
Siu's normative claim — power is a thoroughly moral phenomenon — has been available since 1978 and has not produced visible institutional change in the operator class he wrote it for. If the psychology-of-permission handshake is right, the claim cannot reach role-captured operators because they are no longer in a position to hear it. Is there a developmental or structural intervention that produces operators who can still hear the claim after role-capture? Or is the answer that the claim only reaches pre-captured operators (graduate students, junior staff) and the institutional class is structurally beyond its reach?
The trichotomy (effective / defective / deceptive morality) is descriptively rich but operationally vague — most actual cases involve all three at once. Is there a way to quantify the proportions, or does the categorization remain a heuristic that helps the operator notice the question without resolving it?
The technological-imperative claim — that means existence drives end realization — has counter-examples (chemical weapons not deployed at scale, nuclear weapons used only twice). What are the structural features of the counter-examples? Is restraint an architectural design choice, a personal-character feature of specific operators, or an emergent property of the institutional substrate? The operator who wants to build means that resist their own deployment needs to know the answer.