The Unseeing Gaze III: Can AI make a friend-enemy distinction?

 The consequences of an individual human's action are rarely soverign even if in a phenomenological (first person) sense they are made based on an assumption of sovereignty. 

Really the Schmittian idea that the Soveriegn decides on the state of exception, on the conditions that merit war in society, itself are undermined when we consider that a lot of decisions today in war or for war are made by automated decision making systems, preciely because the strategy employed in modern warfare exceeds tactical considerations that can be encompassed by human intelligence. Particularly from the point of view of nations with established armies, fighting nations with irregular armies that create asymmetric situations for warfare. 

Tim Blackmore in his history of military technology used by tactically advanced military units, describes how the only way to combat asymmetric warfare for a conventional army is through technology that provides a panoptic vision of the landscape. These military packs or swarms combat the knowledge of the so called terrorist with several view points on the landscape: the viewpoint of a completely interconnected sensorial apparatus, everyone in the group is constantly aware of the perspective of all oters in the group, and various levels of transmitted Satellite imagery, along with the binocular sight of a gun, the infra-red goggles etc, create a kind of panotic, if fragmented view of the landscape. It is precisely in this very base sense that AI appears to have 'understanding' to produce a kind of synthesis to allow this fragmented image to turn into one that is 'perceived'. 

Carl Schmitt, a German political and legal theorist wrote what has become in the years following its publication, a book (The Concept of the Political), in 1932 following which he joined the German Nazi Party. However concepts in Schmitt have been critical to political and social theory since, to many philosophers including Arendt, Derrida and Agamben. Really what Schmitt provides is an outline of political subjectivity, that for many is an outline of subjectivity itself. In this sense Schmitt's sense of the subject is perhaps the most base level real political understanding of a subject one can have. 

Schmitt starts his book with the question 'What is the state?'. Its hard, however to define a state: a territorial unit, a people, a government, a procedural order? Schmitt argues that in fact we need to start with politics which comes before the state. He defines politics by what he calls the frien-enemy distinction. This distinction comes to define then the bounds of the modern state as a territory within which the enemy has been banished, and the enemy precisely as a state one might be willing to engage in war. The enemy is thus the outside to the state. 

Schmitt who has some concern with catholicism in the text, wonders if it is alright to fight one's enemy when one is supposed, in Christian terms to 'Love one's enemy'. He argues in fact that the Christian term (enemicous as opposed to hostace) for enemy implies a very different sense. The friend-enemy distinction he is discussing defines politics and the state. It is worth quoting a lengthy passage Schmitt writes on the Hegelian conception of the enemy:

'Hegel also finally established a definition of the enemy that is otherwise usually avoided by modern philosophers: it is the ethical difference (not in the moral sense, but in the sense of the “absolute life” in the “eternal of the people”) manifested as the Foreign in its living totality that is to be negated. “Such a difference is the enemy, and difference, correspondingly, is simultaneously its counterpart, the existence of opposites, yet also the negation of the enemy; and this negation on both sides equally is the danger of the conflict. This enemy can ethically only be an enemy of the people and itself only another people. Because here the particularity appears, so it is for the people that the individual puts himself in danger of death.” “This war is not a war of families against families, but of peoples against peoples, and thus hatred itself is undifferentiated, separate from all personality.” '

War is an exception to peace. But war has to be waged to ensure peace. The ruler or soverign decides on the exception. Soverignity is the ability to decide on the exception. 

                                                                                            -x-

The Minab School Attack on 28th February 2026 begs many questions of this distinction, most of all if autonomous warfare can produce this distinction. However we can start with an even more basic case. 

IFF (Identifier Friend or Foe) systems are used conventionally by ground aviation staff to check if an aiborne veihichle is a friend or a foe. The IFF can only identify if an airborne vehichle is a friend, and cannot positively identify a foe. An aircraft might be just not able to respond to the message form the IFF. 

This isn’t just a technical quirk—it’s a bigger philosophical problem. The system doesn’t actually pick out enemies. What it does is fail to recognize friends. The “enemy” is just what’s left over when there’s no sign of identification. So, it’s not really about recognizing hostility. It’s about the absence of recognition. If you follow Schmitt’s logic, this presents a problem right away. For him, the friend-enemy line is an existential choice, something decided in the heat of conflict, where groups name each other as enemies. But that’s not what’s happening with IFF systems. The distinction isn’t existential, and it isn’t mutual; it’s practical. Here, “enemy” isn’t a person, or even a subject—it’s just a missing signal.

You see this even more in modern automated warfare. Machine learning models scan for movement patterns, heat signatures, odd communication blips, and other statistical oddities. They aren’t naming enemies in any Schmittian sense. They’re assigning threat levels. A person becomes a target—not because someone points and calls them “enemy”—but because they breach a line in a probability chart. The “decision” isn’t a single moment; it’s pieced together from training data, sensors, model tweaking, and on-the-fly algorithms. What feels like a snap judgment is really an accumulation of little factors.

So, the sovereign power—the power to declare exceptions or identify threats—isn’t tied to a single moment of authority anymore. It’s baked right into the technology, continuously recalibrated rather than announced outright. War stops being a break from peace and turns into a steady state, always fluctuating according to shifting probabilities.

Think back to IFF systems. The system can’t say, “There’s the enemy.” All it can do is fail to back someone as a friend. That uncertainty turns into suspicion, and suspicion can escalate into an attack. But the trigger for this escalation isn’t figured out in the moment—it’s set in stone ahead of time: in the rules of engagement, in how much risk is allowed, even in how sensitive the algorithms are.

So where does the “decision” really happen? It’s in how the system is built. And that shift scrambles the question of who’s responsible. If the wrong target—like a civilian plane—gets hit because it replied with the wrong code, who actually made the call? Was it the operator running the system? The engineer who wrote the code? The higher-ups who set policy? Or was it the system itself? It’s a mess, and no answer really satisfies, because sovereignty is scattered across the whole structure.

Still, the Schmitt logic has been twisted, not erased. There’s a friend-enemy distinction, but it’s watered down. It doesn’t depend on direct recognition, mutual identification, or even intention. It works through exclusion—through whatever can’t fit into the network. In this sense, modern warfare doesn’t get rid of Schmitt. It stretches his ideas as far as they can go. The enemy isn’t another nation’s soldier anymore—it’s the missing response, the blip on a radar, the anomaly the data can’t sort out.

So now, politics doesn’t hinge on deciding who’s the enemy. It’s all about how the system keeps producing a space where the lines blur, where “friend” and “enemy” are no longer drawn sharply, but constantly adjusted—and sometimes, with deadly results, miscalculated altogether.

The so-called “Maven” class of systems—named after Project Maven—represent a shift from identification to pattern extraction. Originally developed to assist analysts in parsing drone footage, these systems do not “see” in the human sense. They correlate. What appears as a figure on the ground is decomposed into vectors: movement, heat, clustering, deviation from baseline patterns. The output is not recognition but annotation—bounding boxes, probability scores, flags.

In accounts that attribute systems of this kind to events like the Minab bombings, what is at stake is not whether the system “identified” a target, but how a target came to be constituted within a chain of inferences. A building is not a school or a shelter in itself; it becomes one through classification. When classification is automated, the object appears only as a provisional stabilization within a stream of data.

The intelligence such systems produce is therefore neither fully human nor fully autonomous. It is assistive, but in a peculiar way: it assists by narrowing the field of attention while expanding the field of possible error. The analyst, or operator, does not decide in a vacuum but within a pre-structured perceptual field, already sorted, highlighted, and weighted by the system.

This is where the earlier discussion of IFF becomes relevant. Just as IFF cannot positively identify an enemy, Maven-like systems cannot positively identify a target in any absolute sense. They can only assign likelihood. The “decision” to strike emerges at the point where likelihood is treated as certainty, where probabilistic thresholds harden into operational commands.

If such a system was indeed implicated in the Minab attack, then the question is not whether a machine made the decision, but how decision-making was distributed across the system: across datasets, training regimes, sensor inputs, interface design, and human interpretation. The strike, in this sense, would not be the outcome of a sovereign act, but of a cascade.

And yet, precisely because this cascade culminates in destruction, it retroactively appears as a decision. Sovereignty reappears at the point of impact, even if it was absent in the process.

                                                                                            -x-

What systems like Maven ultimately expose is not the disappearance of sovereignty, but its mutation. For Carl Schmitt, the sovereign is the one who decides on the exception, who draws the line between friend and enemy in a moment of existential clarity. But in the context of automated perception, that moment dissolves. The exception is no longer declared; it is inferred. The enemy is no longer recognized; it is statistically produced. Yet the structure Schmitt describes does not vanish—it is displaced into the architecture of the system itself.

The point being that what is present in these systems is not a consciousness but a 'projected consciousness'. Whether it is the case of  a projection involving friend enemy decisions in real life or the fictional projective consciousness in the case of HAL in the movie '2001 a space odyssey'. I will atempt to understand projection better as a system in the next and final blogpost in this series.