🤖 When the Tail Wags the Dog: AI and the Quiet Reconfiguration of ADR

by Pierfrancesco C. Fasano

The contemporary debate surrounding artificial intelligence in the legal domain has predominantly focused on State adjudication, predictive justice, and algorithmic systems supporting judicial decision-making. Yet the same epistemic concerns are increasingly emerging — perhaps in an even more acute form — within the universe of Alternative Dispute Resolution (ADR), where confidentiality, procedural flexibility, and the privatisation of dispute-management infrastructures render algorithmic influence considerably less visible, and therefore more difficult to scrutinise.

Artificial intelligence within ADR does not merely operate as an administrative aid or a tool of procedural efficiency. Increasingly, it functions as an epistemic infrastructure of dispute resolution itself: shaping what parties perceive as reasonable claims, which precedents appear relevant, which outcomes seem probable, and which settlements are regarded as fair, efficient, or commercially rational.

In other words, the algorithm no longer merely assists the management of disputes; it progressively contributes to constructing the very cognitive framework within which disputes are understood and resolved.

This transformation manifests differently depending upon the nature of the ADR mechanism involved — whether mediative, adjudicative, evaluative, or hybrid — yet the underlying dynamic remains consistent: the gradual displacement of argumentative rationality by statistical, predictive, and classificatory logics.

Mediation: From Human Facilitation to the Algorithmic Conformation of Consensus

Mediation constitutes perhaps the most sensitive terrain for the application of artificial intelligence.

Traditionally, mediation rests upon:

• listening;

• mutual recognition;

• relational conflict management;

• creative exploration of interests;

• and the consensual construction of solutions.

Artificial intelligence, by contrast, tends structurally to privilege:

• statistical recurrence;

• economic benchmarking;

• predictive modelling;

• pattern recognition;

• and probabilistic settlement analysis.

An increasing number of Online Dispute Resolution (ODR) platforms and AI-assisted mediation systems are capable of:

• estimating likely settlement zones;

• proposing compensation ranges;

• predicting litigation outcomes;

• classifying negotiation strategies;

• and recommending standardised conciliatory clauses.

Such instruments undoubtedly enhance:

• efficiency;

• accessibility;

• procedural speed;

• and cost reduction.

Yet they simultaneously introduce a structural epistemic risk: the transformation of consent into statistical conformity.

Where parties are repeatedly exposed to:

• algorithmic benchmarks;

• probabilistic success rates;

• average settlement values;

• and “recommended” outcomes;

their autonomy may progressively become confined within a cognitive perimeter designed by the system itself.

The settlement ceases to emerge solely from dialogue between the parties and increasingly reflects the implicit normative pressure exerted by predictive rationality.

Over time, even mediators themselves may unconsciously begin to perceive statistically dominant solutions as inherently “reasonable”, thereby allowing the logic of the dataset gradually to replace the autonomy and creativity traditionally associated with consensual dispute resolution.

Arbitration: Predictive Analytics and the Algorithmic Hierarchisation of Awards

Within arbitration, artificial intelligence primarily manifests through:

• legal analytics;

• predictive arbitration systems;

• intelligent precedent research;

• behavioural profiling;

• and outcome forecasting.

Contemporary platforms already permit users to:

• analyse arbitral tendencies;

• estimate procedural duration;

• predict success probabilities;

• evaluate interpretative inclinations of particular arbitrators;

• and reconstruct citation networks between awards and decisions.

The epistemic problem becomes especially pronounced within systems characterised by:

• confidentiality;

• limited publication of awards;

• concentrated reputational authority;

• and proprietary datasets controlled by a small number of institutional or commercial actors.

In such an environment, whoever controls the dataset progressively controls the predictability of arbitral law itself.

This generates a phenomenon of algorithmic self-reinforcement:

• the most accessible awards become the most cited;

• the most cited become the most authoritative;

• the most authoritative generate further statistical relevance;

• whilst innovative or minority approaches become progressively marginalised.

The consequence may be the emergence of a form of private algorithmic nomophylaxis, lacking the constitutional safeguards ordinarily associated with public adjudication.

Within intellectual property and technology disputes, for example, predictive systems may increasingly influence:

• FRAND determinations;

• SEP valuation methodologies;

• damages quantification;

• UPC and PMAC litigation strategies;

• and assessments concerning injunctive relief.

The resulting transformation would not merely be technical. It would shape the evolution of the contemporary lex mercatoria itself.

Expert Determination: The Illusion of Technical Neutrality

Expert determination represents one of the ADR mechanisms most vulnerable to algorithmic naturalisation.

Because such procedures rely heavily upon technical expertise — including:

• royalty calculations;

• asset valuation;

• scientific assessments;

• software compliance;

• industrial performance metrics;

• and economic modelling —

algorithmic outputs are particularly likely to be perceived as objective, scientific, and therefore incontestable.

Yet even technical algorithms inevitably incorporate:

• selective criteria;

• methodological assumptions;

• embedded economic preferences;

• weighting systems;

• and implicit normative choices.

An AI system employed to determine:

• SEP royalties;

• intellectual property valuations;

• reputational damage;

• or technological compliance;

does not simply calculate objective truth.

Rather, it constructs normative outcomes under the appearance of technical inevitability.

The deepest epistemic risk lies precisely in this transformation of contestable methodological choices into apparently neutral computational facts.

Within expert determination, the danger is therefore not automation itself, but the replacement of critical technical scrutiny by deferential reliance upon algorithmic authority.

Early Neutral Evaluation: When Prediction Becomes Prescription

Early Neutral Evaluation (ENE) arguably represents the clearest point of intersection between ADR and predictive justice.

ENE was historically designed to provide:

• preliminary legal assessment;

• neutral risk analysis;

• realistic procedural evaluation;

• and guidance capable of facilitating settlement discussions.

Artificial intelligence now enables:

• automated comparative analysis;

• probabilistic outcome estimation;

• scenario simulation;

• and strategic litigation modelling.

Yet it is precisely here that the most profound cognitive risk emerges: prediction tends psychologically to become prescription.

When an algorithm:

• attributes a 78% likelihood of defeat;

• recommends an “optimal” settlement range;

• identifies statistically dominant precedents;

parties naturally tend to adjust their behaviour accordingly.

The evaluative process risks thereby becoming a mechanism of algorithmic conformity rather than informed autonomous judgement.

Predictive rationality ceases merely to inform the dispute and instead begins subtly to govern it.

Domain Name Disputes: Algorithmic Bad Faith and the Standardisation of Cybersquatting Analysis

Domain name disputes constitute one of the most advanced laboratories for AI-assisted ADR.

Procedures such as:

UDRP;

URS;

ccTLD dispute mechanisms;

and national domain ADR systems;

already rely extensively upon:

automated precedent research;

semantic analysis;

pattern recognition;

bad faith detection;

and trademark similarity assessment.

Artificial intelligence is increasingly capable of:

• identifying suspicious registration clusters;

• detecting serial cybersquatting behaviour;

• analysing linguistic structures;

• correlating WHOIS and DNS information;

and recommending relevant precedents to examiners or panellists.

However, the principal risk lies in the gradual automation of the concept of bad faith itself.

Historically, bad faith under UDRP and URS jurisprudence has remained a contextual and argumentative category, dependent upon factual nuance and interpretative assessment.

Algorithms, by contrast, tend structurally to:

• convert bad faith into a statistical pattern;

• standardise behavioural indicators;

• privilege correlations;

• and penalise atypical factual circumstances.

Over time, cybersquatting risks becoming interpreted not through argumentative legal analysis, but through probabilistic categorisation derived from historical datasets.

The dispute resolution process itself may thereby evolve into a semi-automated system of reputational classification.

The Digital Services Act and Article 21: Double Algorithmic Mediation

The out-of-court dispute settlement mechanisms established under Article 21 of the Digital Services Act represent perhaps the most sophisticated — and potentially most problematic — context for contemporary algorithmic mediation.

Certified dispute settlement bodies under the DSA operate within ecosystems already profoundly shaped by platform algorithms governing:

• ranking;

• moderation;

• recommender systems;

• demonetisation;

• account suspension;

• and content removal.

Disputes arising under Article 21 therefore already originate within an algorithmically mediated reality.

The ADR body itself may additionally employ:

• AI-assisted triage systems;

• automated complaint classification;

• clustering analysis;

• standardised decision templates;

• and predictive benchmarking.

A form of double algorithmic mediation consequently emerges:

• first through the platform;

• then through the dispute resolution mechanism itself.

The systemic implications are considerable.

Concepts such as:

• harmful content;

• disinformation;

• illegality;

• manipulation;

• or proportional moderation;

may progressively become stabilised through statistical correlation rather than genuine argumentative scrutiny.

Within the DSA framework, the governance of AI in ADR is therefore not merely procedural.

It is constitutional, democratic, and epistemic in nature.

The Need for an Epistemic Governance of AI in ADR

ADR historically emerged as a space characterised by:

• flexibility;

• autonomy;

• pragmatism;

• confidentiality;

• and relational problem-solving.

Yet precisely these characteristics now render it especially vulnerable to algorithmic naturalisation.

The absence of:

• systematic publication;

• transparent datasets;

• public reasoning obligations;

• appellate nomophylactic structures;

• and independent algorithmic audits;

facilitates the silent transformation of legal rationality into statistical rationality.

For this reason, the future development of European ADR will likely require:

• algorithmic transparency registries;

• mandatory disclosure obligations concerning AI usage;

• independent auditing mechanisms;

• meaningful human oversight;

• procedural rights to contest algorithmic outputs;

• and European ethical standards for AI-assisted dispute resolution.

The challenge is not to prohibit artificial intelligence within ADR.

The real challenge is to prevent consensual dispute resolution from being quietly reconfigured by opaque systems that transform statistical probability into legal truth and computational convergence into the appearance of genuine consensus.