Vibe Justice

The file arrives already warm, and on the screen sit a category, similar cases, an anomaly flag, and a line that recommends inspection, while the person in uniform still reads, can still set it aside, still signs in their own name, and the picture has been assembled before they entered, so independent inquiry begins to feel like extra friction: everything is already clear. Vibe is almost a technical word here and names the sense of a task already set, which the file carries as a cluster of weak signals, classifications, and recommendations before any encounter with a particular event, and the encounter begins as confirmation.

I call vibe justice the condition in which a formally human legal decision is produced as the result of inference after Inferentia, inside a computationally prepared field of classifications, recommendations, and variations of a path toward a task already set, where a person signs in their own name as the machine’s proxy in reality, remaining the last subject of the decision, while the space in which that decision looks reasonable has already been assembled upstream.

Antidigitism named digitism as an admission rule by which machine-readable representation becomes a condition of a person’s recognition by an institution, a workplace, a service. Here the person is already inside, so the question moves to where the inference on which the next action depends takes place. Digitization can still only record, but when the centre of inquiry shifts into the infrastructure, the system starts deciding what is suspicious, what resembles what, what deserves inspection first, and in what order to show this to a person with authority. Freedom to choose among presented options remains, while freedom to determine which options belong to the case at all narrows.

A recommended inspection

In 2013, in La Crosse, Wisconsin, Eric Loomis was charged in connection with a nighttime drive-by shooting. He denied taking part in the shooting and admitted that he had later that night driven the same car; he pleaded guilty to two of the less severe counts. A COMPAS risk assessment was attached to the presentence investigation, comparing information about a person with group data and returning bars for the likelihood of reoffending; the method remains a trade secret, and the court receives the scores. The circuit judge, ruling out probation, named among the grounds the seriousness of the crime, the supervision history, and the suggestion from risk-assessment tools that Loomis was extremely high risk to reoffend. The COMPAS report showed high bars for violence, recidivism, and pretrial risk. In 2016 the Wisconsin Supreme Court held that considering such an assessment at sentencing does not, of itself, violate due process if the judge observes the limits under which the scores do not determine whether to incarcerate or how severe a sentence to impose, and do not become the sole deciding factor for community supervision (State v. Loomis, 2016), so formally the judge decides, and the language of ready-made bars was already in the room.

On 5 February 2020 the District Court of The Hague held that the legislation governing SyRI had no binding effect. SyRI linked large datasets in order to detect signs of fraud in benefits, allowances, and taxes, and directed attention where the model saw risk. The court found that the Netherlands had failed the fair-balance test under Article 8 of the European Convention on Human Rights: the application was insufficiently transparent and verifiable, and deployment in “problem neighbourhoods” could stigmatize residents in ways the court could not even assess (Rechtbank Den Haag, 2020). After the system had already decided who was worth beginning to inspect, a person could still inspect later.

Order No. 621 of the Prosecutor General of Russia, dated 11 September 2025, approves a concept of digital transformation for prosecutor’s offices through 2030. Among its aims are artificial-intelligence technologies for detecting signs of violations from digitised data on supervised objects, work with citizen appeals, forecasting of promising directions of activity, and flexible, precise focusing. Tasks under “high-technology supervision” include automated quality assessment of law-enforcement work and automated detection of characteristic patterns corresponding to violations, with the preparation of recommendations for inspection. The key result named is an end-to-end official process for ensuring legality, with artificial-intelligence technologies built into the prosecutor’s information systems (Order No. 621, §2). Qualification of an offence, compilation of a case, a finding of guilt, and the imposition of a sentence remain with the person in the concept. Trial operation of digital services with artificial intelligence belongs to the 2029–2030 stage (Order No. 621, §8). The analytic force of the order lies in how an institution’s attention can begin to move from data to classification and anomaly, to recommendation and priority, to prepared context and a human decision.

European guidance on Article 22 of the General Data Protection Regulation requires that a person be able to refuse an automatically generated profile (Article 29 Working Party, 2018); the guidance is a European reading and does not bind the Russian prosecutor’s office.

The investigator, the prosecutor, the judge, and the officer can remain in their places while the origin of the picture they work with changes, and the person who inquires becomes an operator, then a reviewer, then the one who approves, and at each step it is true that a person takes the final decision, though the question moves to who produced the space of possible decisions.

When inspection follows where the model has already pointed, it produces new data, because a place left uninspected stays silent, while a place inspected intensively fills with findings, and the model receives confirmation of its own attention. Kristian Lum and William Isaac showed this in a PredPol simulation, where the algorithm begins to predict future policing more strongly than future crime (Lum and Isaac, 2016). Ensign and colleagues formalised the runaway of that loop when training receives discovered incidents that arose only where the system had already been sent (Ensign et al., 2018). Order No. 621 does not itself describe this risk, yet a pattern corresponding to a violation, together with a recommendation to inspect, opens the same form of selective observation.

Action as information

Inferentia & Probabilisticum named Inferentia as the mechanism of continuous assembly, ranking, and regeneration of a symbolic surround, and Probabilisticum as the name of an era in which reality is mediated by autonomous inference, and a harder operation now attaches to this mechanism, in which inference can be intervention and a system can act on a person in order to know them.

James Fearon in 1995 posed the puzzle of rationalist war: if war is costly for both sides, why do they not find in advance a settlement both would prefer to a fight? The coherent answers are private information about capabilities and resolve together with an incentive to hide or distort it, commitment problems, and, more weakly, indivisibility of the issue (Fearon, 1995). Where information about the opponent is incomplete, fighting can reveal capabilities and costs, after which the range of settlement changes, and Dan Reiter follows that bargain through the initiation, conduct, and termination of war (Reiter, 2003).

A platform and a user rarely sit at one bargaining table where an agreement could be reached before the experiment begins. In Antidigitism a person is already inside the trial once a system can reach them with a display, a notification, or a change of rank and take a response; the reach of the channel is enough to start. One operation transfers from bargaining theory, in which action itself produces information. An algorithm chooses what to show this person now, sees whether a response arrives, and updates its estimate of what will work next. Li, Chu, Langford, and Schapire applied that scheme to Yahoo’s personalised front page, naming advertisements beside news (Li et al., 2010). Trying something new and keeping what worked is already a computational paradigm, even where a system ranks by other means, and a display, a notification, a change of rank, a proposed price, or a generated image can do the work of a small probe.

Post-Simulacrum Self-Service Without Contact already described the loop in which a system observes, infers susceptibility, selects or generates a stimulus, exposes, measures, and updates, while refusal also enters the training. Post-simulacrum sharpens one step further here, when a produced object can exist solely to elicit a reaction, and the classical chain of reality, representation, audience gives way to another, where traces lead to inference, a generated stimulus, a reaction, and new traces, and representation drops out. A thousand variants of a headline or an image need not have an artistic or even an advertising existence in the older sense if their function consists only in discriminating reactions, and media ceases to depict and begins to probe, while in probabilisticum representation becomes experimentation.

A person is then included as part of an experimental apparatus, where reaction is enough and the loop does not wait for publication, and an unknown vulnerability meets a probabilistic probe, the reaction updates a belief, and the next attempt arrives closer. Advertising, interface, A/B trial, and generative variation converge on one grammar, where preference becomes experimental data.

Uncertainty is expensive because it asks for direct action and the creation of situations in order to obtain information, and that costs time, awkwardness, incomplete knowledge, and the right to be wrong inside a living encounter. The Manifesto for Collaborative Concurrent Extreme already describes knowledge growing through the materialisation of situations and events, and digital inference offers a cheaper path, where the unknown becomes a prediction at once and the encounter is already read as a profile with a score and a ranked list. Delegating inference is often locally rational, and in the aggregate each such transfer removes one place where direct experience could have modified the model of the world. In direct experience, when a person walks, notices resistance, repairs a misunderstanding, and the hypothesis changes because reality pushes back, the experience itself is already inference, and when inference occurs before perception, experience begins after prediction, and music is met where the recommender let it through, a route is driven along the computed line, and a case begins with an anomaly flag.

When a machine-readable state receives the authority to decide what counts, digitism gives legitimacy to digital intervention. To touch the world, a computer needs a person, because it has no other interface to reality, and people become its proxies. From traces, without direct knowledge of the object, Inferentia already gathers probabilities that Probabilisticum sends into iterative action, while post-simulacrum supplies disposable sign-stimuli without an originating referent, and a fifth operation binds these axes, action as the acquisition of information.

In vibe coding a person hands the model their sense of the task, and the assembly follows that sense, while the model runs generative iterations and yields variations, and when the result of inference after Inferentia meets the present need and solves the task, the work is accepted, even if it has already left the earlier intention, and the code can remain unread. Vibe justice promises the same move at another scale, where anyone can set a task and formulate it however they like, down to a banal “convict citizen N,” and after inference the system produces variations of a path toward that outcome, and people take the one that now solves the task. In the Russian political system as I observe it, target indicators and target court cases already belong to the ordinary work of institutions, and now the setting of the task stands farther from the executors of the law. The institution places the specialist in a convenient cockpit of inference, where generative variations have already been fitted to the assigned task, and the more completely the feeling that everything is clear takes shape there, the more expensive independent investigation of the particular event becomes.

Judging by vibes is dangerous, because the ground of a decision becomes the formulated task and the variations fitted to it, while a human name still stands on it, a name through which the result of inference enters reality. Once the unknown has already been turned into a score and a ranked list, an institution closes the encounter with a rule, and the signature is placed inside that readiness. Digital inference gathers an institution’s attention before the signature: the loop reaches each person, people gather by a signal, and classifications and recommendations arrive before a living encounter can change the picture.

The danger of combining vibes and digitism is that the transition into post-simulacrum and Probabilisticum leaves even the judicial system detached from reality, and people work through their own conclusion already from inside the infrastructure where recognition of a person and the formulated task belong to one field, and these moves remain fragments of one new era.

Antidigitism contests machine-readable recognition as the sole condition of admission. Here the person is already inside, and what this essay contests is the right to use a life as the medium of a computational experiment, a right that appears once a channel can reach a person and take a response, while the signature is placed inside a field already assembled.