The Psychology of Arbitration in the AI Era: Human–AI Interaction, Cognitive Bias, and the Reliability of Arbitral Decision-Making

[Amin Motamedi is an Arbitrator at the Iranian Chamber of Commerce (ACIC), a university lecturer, and holds a PhD in international law]

Artificial intelligence (AI) has increasingly been employed in a wide range of professional contexts, including legal analysis and dispute resolution. Practitioners have relied on AI tools to address legal questions, conduct analytical assessments, and, in certain instances, assist in the drafting of arbitral awards. While human decision-making is inherently complex and shaped by numerous cognitive and psychological biases, these challenges are not confined to human actors alone. AI, despite its technological sophistication, is not immune to analogous forms of bias.

When the psychological limitations and cognitive predispositions of human decision-makers are taken into account, matters that will be examined in greater detail in this essay—it becomes evident that AI, through the data on which it is trained and the algorithms that govern its functioning, may also generate divergent or skewed outcomes. Such outcomes may reflect algorithmic or systemic biases that parallel, and at times amplify, human cognitive biases and mental heuristics.  This essay focuses specifically on the influence of an arbitrator’s psychological and cognitive capacities on human-AI interaction arbitral decision-making and award drafting. Rather than treating human judgment and AI as competing alternatives, it examines arbitration as an emerging human-AI decision-making ecosystem in which arbitrators and AI tools interact throughout legal analyzing and award drafting. The central question is how this interaction should be structured to minimize cognitive and algorithmic biases while ensuring transparency, accountability, and meaningful human oversight.

Who Issues the Arbitral Award: Me, or my Cognitive Biases?

In psychological terms, cognitive bias refers to unconscious patterns of thought that systematically deviate from rationality and impartiality, leading individuals to make judgments that are not entirely objective. In more pronounced cases, these biases may obstruct the parties’ ability to achieve the intended or legally justified outcomes of the arbitral process. In the context of international arbitration, notwithstanding the high level of expertise, technical competence, and professional experience of arbitrators, cognitive biases remain largely unavoidable. These realities have increasingly drawn scholarly and practical attention to the psychology of dispute resolution and the growing relevance of psychological insights within arbitral decision-making.

In contemporary practice, nearly all arbitrators and arbitration practitioners have, at least on one occasion, relied on AI tools for legal research, doctrinal analysis, drafting legal arguments, or even preparing arbitral awards. Human decision-makers, however, differ significantly in terms of emotional thresholds, cognitive capacities, and analytical frameworks. These variations inevitably produce divergent outcomes, including differences in the reasoning, structure, and substance of arbitral awards. From a psychological standpoint, human adjudication differs fundamentally from decision-making assisted or generated by artificial intelligence. Mental and cognitive biases, together with an arbitrator’s professional background and prior experience, play a decisive role in shaping nature and direction of arbitral judgment. The existence of cognitive biases in international arbitration underscores a critical reality: even highly professional decision-makers remain human, operating within cognitively bounded systems influenced by personal and experiential contexts. Acknowledging this limitation does not undermine arbitration; rather, it contributes to a more realistic, equitable, and effective dispute-resolution mechanism.

Among the most significant cognitive biases observed in international arbitration are Role-Induced Bias, Cultural and National Bias, Authority Bias, Anchoring Bias, and Confirmation Bias. Psychological factors may become so influential that not only the substantive merits of a claim, but also the manner of its presentation—particularly in oral submissions—can materially affect an arbitrator’s perception and evaluation of the dispute. Within broader debates on subjectivity and objectivity in psychological studies, it is widely recognized that human analysis and interpretation inevitably contain subjective elements and personal perspective. This observation is equally applicable in legal reasoning and arbitral adjudication. The cognitive processes occurring within the arbitrator’s mind directly influence both decision-making and the ultimate issuance of the arbitral award.

Invisible Arbitrator: What We Hold on to in AI-based Judgment

The key issue facing the arbitral community is not whether AI will become a part of arbitration, AI-assisted arbitration should be understood as a sociotechnical mechanism where human arbitrators and AI tools interact rather than the process of technological replacement. Its legitimacy depends on how AI recommendations are generated, assessed, and incorporated into arbitral decisions. While AI may reduce some human cognitive biases, it can also create new risks, such as automation bias and reliance on flawed algorithmic outputs. Therefore, AI governance in arbitration should emphasize human oversight, accountability, and procedural safeguards alongside technological efficiency. But how can its application be aligned with the core principles of party autonomy, due process, and arbitral independence. Arbitral tribunals gain legitimacy through human reasoning and procedural fairness; therefore, the delegation of substantive or reasoning tasks to AI introduces significant challenges related to accountability, transparency, and enforceability. The New York Convention (1958) and the UNCITRAL Model Law (1985, as amended) assume that human adjudicators will exercise independent judgment. These frameworks do not contemplate the role of algorithmic co-decision-makers or AI-assisted awards. As AI tools become increasingly integrated into the arbitration drafting process, the need for consistent regulation becomes more pressing. Recent scholarship has similarly argued that AI-assisted legal decision-making in international humanitarian law (IHL) context should be viewed as a sociotechnical system in which human and technological actors continuously shape one another, rather than as a substitution of machine judgment for human expertise.

Institutional responses to AI in arbitration reveal a converging but still largely soft-law approach focused on transparency, human oversight, and ethical responsibility. The Chartered Institute of Arbitrators’ Guideline on the Use of AI in Arbitration (2025) requires disclosure of “material” AI use, empowers tribunals to regulate AI via procedural orders, and cautions arbitrators against delegating substantive reasoning to AI, though it relies on voluntary uptake and leaves open questions about AI-assisted drafting’s effect on arbitral reasoning. The Silicon Valley Arbitration & Mediation Center (SVAMC) Guidelines (2024) complement this by emphasizing technological literacy, data security, vendor accountability, and clearly identifying AI-generated content, while discouraging generative AI for dispositive reasoning. The International Bar Association’s report “The Future Is Now: Artificial Intelligence and the Legal Profession” (2024) extends professional ethics to AI use by underscoring lawyers’ duties of competence and supervision before submitting AI outputs, principles that logically extend to arbitrators. IBA At the multilateral level, UNCITRAL Academy’s 2024 Report signals AI as a priority for harmonization, pointing toward future model legislative work on algorithmic transparency and human oversight. 

Delegating reasoning to AI risks undermining arbitrator independence and accountability; the CIArb AI Guidelines caution that AI should assist but not replace human judgment. Confidentiality and data protection are also critical: the SVAMC Guidelines recommend strict data-security measures, safeguarding privileged communications, and maintaining audit trails. Additionally, cross-border data-transfer regulations (e.g., the EU GDPR) must be considered. The European Union Artificial Intelligence Act (2024) provides a benchmark for this approach by classifying certain AI applications, such as those assisting in legal reasoning, as “high risk,” thereby imposing mandatory standards of oversight, auditability, and traceability. Finally, AI systems are prone to bias, so explainability is essential, with outputs verified by human arbitrators to ensure fairness and enforceability across jurisdictions demanding transparency. 

Bridging Human Cognition and AI Computation in Arbitration Decisions: Harmony or Tension?

The relationship between cognitive bias and AI in arbitration is best understood through the interaction between human and technological actors rather than through a comparison of their individual strengths. Human expertise guides the design, selection, and evaluation of AI outputs, while AI increasingly shapes legal research, drafting, and analytical reasoning. Consequently, reliability depends not only on the capabilities of each actor individually but also on how their interaction is structured and governed. The topic of cognitive bias and psychological factors in the context of human beings and AI is fascinating and multi-layered, as both involve processes of decision-making and information processing, albeit through different mechanisms. Cognitive biases in humans arise from mental limitations, past experiences, emotions, and the need to simplify a complex world. In contrast, in AI—at least as of the time of writing—biases typically originate from the data, algorithmic design, and the objectives set by humans. In essence, the types of biases differ between the two: AI can, in fact, replicate human biases without involving emotions.

Altering biases in humans requires self-awareness (with or without the assistance of a psychological professional) and sustained practice, which is often slow and uncertain. AI, however, cannot “sense” its biases consciously; any corrections must always be applied externally by human operators. From another perspective, human biases are generally rooted in mental architecture and individual experience, whereas AI reflects a mirror of human-designed data and systems. This means that AI possesses neither free will nor emotions, yet it can amplify or reinforce human biases.

In the domain of judicial decision-making, both human judges and AI exhibit distinct strengths and weaknesses, which in turn affect the reliability of issued rulings. A human judge can consider not only the letter of the law but also procedural practices, the social spirit of justice, the specific circumstances of each case, and the intent of the parties, thereby making flexible and ethically informed decisions. Such capabilities are particularly crucial in complex and sensitive cases. However, human decisions can be influenced by cognitive biases, personal prejudices, or social pressures, potentially leading to inconsistencies in similar cases.

AI can analyse legal materials rapidly, consistently, and across large datasets, while human arbitrators contribute contextual judgment, ethical reasoning, procedural fairness, and an appreciation of the parties’ intentions. Neither human judgment nor AI should therefore be regarded as independently superior. Rather, AI-assisted arbitration functions as a sociotechnical system in which reliability depends upon the quality of interaction between human decision-makers and technological tools. Effective governance should consequently prioritize transparency, meaningful human oversight, clear allocation of responsibility, and mechanisms for identifying both cognitive and automation-related biases. 

In the final analysis cognitive biases are dynamic and cannot be eliminated. Nevertheless, as this essay has demonstrated, increasing awareness of both human cognitive biases and the algorithmic biases embedded in AI systems enhances the effectiveness, balance, and overall reliability of the arbitral process.