The AI Mediator Problem: Can a Machine Be Neutral?

The AI Mediator Problem: Can a Machine Be Neutral?

Revisiting Party Autonomy and Procedural Fairness under India’s Mediation Act, 2023 in
the Age of Online Dispute Resolution

I. Introduction

Bot Mediation, a company from California, debuted its AI-enabled mediator system in 2024.¹ In Singapore, the nation's highest court has been experimenting with using AI-enabled technology to support pre-trial mediations.² India, on the other hand, which recently enacted its first legislation exclusively dealing with mediation, the Mediation Act 2023³ , is now standing at a critical crossroads where it must decide whether or not to incorporate the use of artificial intelligence within its online dispute resolution ("ODR") system and if yes, then how?

The dilemma that faces the country no longer seems far-fetched. Indeed, according to the
Mediation Act, online mediation is recognized as a valid form of dispute resolution. Since there is a multitude of ODR platforms cropping up in India, the reality is that with time artificial intelligence is playing a greater role in dispute resolution, interaction, and even resolution of disputes without mediation at all. It now begs the legal question: can an algorithm be neutral? Is the Indian legal system, with its emphasis on party autonomy and fairness, equipped to regulate the use of artificial intelligence in mediation?

This blog presents the case that the implementation of AI within ODR-mediated proceedings, while offering potential efficiency, raises structural conflicts with the fundamental tenets of impartiality, party autonomy and procedural justice outlined under the Mediation Act, 2023. The blog further argues that the Mediation Act 2023, as currently in force, lacks legislative preparedness when confronted with the issue of AI mediators.

II. The Rise of AI in ODR-Based Mediation: A Global and Indian Snapshot

Online Dispute Resolution originally emerged in the 1990s as a response to e-commerce
disputes, borrowing ADR mechanisms and transplanting them onto digital platforms. Today,
ODR has evolved far beyond asynchronous email-based negotiation. Platforms like Modria,
Smartsettle and CADRE ODR now incorporate machine learning to analyse dispute patterns,
predict settlement ranges and optimise negotiation strategies. These are functions which
scholars describe as the emergence of a “fourth party”: the technology that supplements or replaces the human third-party neutral.

AI operates in mediation across a spectrum. Support systems for AI help perform certain
tasks such as creating summaries of the documents provided, translating messages and
detecting emotional outbursts or making a decision regarding the range for negotiation from past trends. On the opposite side are substitutive AI systems, which involve software that completely replaces the human mediator role. The American Arbitration Association’s “AI Arbitrator” tool, launched for low-value construction disputes, exemplifies this substitutive model: the AI assesses evidence, applies legal reasoning and drafts a recommended award for human review.⁴

In India, the Mediation Act, 2023 provides the statutory foundation for ODR through Section 30⁵ , which permits mediation to be conducted “online” provided parties agree in writing. The Act, however, does not define “online mediation”, nor does it specify the technological standards, platform requirements or AI-governance frameworks applicable to such proceedings. This legislative silence, as argued below, is constitutionally and procedurally consequential.⁶

III. Can a Machine Be Neutral? The Algorithmic Bias Problem

Neutrality is the sine qua non of mediation.⁷ The Mediation Act, 2023 mandates that a
mediator “shall be neutral and impartial” and discloses conflicts of interest that may affect independence.⁸ Further it provides grounds for challenging a mediator on the basis of justifiable doubt as to independence or impartiality. The question is whether these provisions, designed for human mediators, can be meaningfully applied to an AI system.

The answer, at present, is deeply problematic. AI systems are trained on historical datasets that reflect the socio-economic biases, power asymmetries and settlement patterns of past disputes. A machine learning model trained predominantly on commercial disputes between sophisticated corporate parties will carry structural assumptions that are ill-suited to consumer disputes, labour conflicts or family mediations involving vulnerable parties. This is not a theoretical concern: studies on algorithmic decision-making in legal contexts have documented systematic bias against low-income parties, non-native language speakers and minorities in predictive outcomes.

More fundamentally, an AI system’s “neutrality” is an engineered artefact, a product of
design choices made by its developers, choices that embed particular assumptions about what a fair outcome looks like. As many experts in the ethics of artificial intelligence have observed, lack of disclosure does not mean a lack of potential for bias; rather, it implies that the potential is less obvious and thus more difficult to contest.⁹ According to the Mediation Act, any party can file a request for the termination of a mediation mandate because of “the existence of circumstances raising justified doubts as to the impartiality of the mediator.”¹⁰ But who triggers this challenge against an algorithm? What procedural mechanism does a party use to inspect the training data of an AI mediator for embedded bias? The Act provides no answer.

IV. Party Autonomy Under Algorithmic Pressure

Party autonomy, which means that the parties involved in the dispute will have control over the course, participants and final result of mediation, is a core value of mediation. This runs through the entire mediation process, from start to finish, regardless of whether the mediation ends in a settlement or termination of the process. The Mediation Act, 2023 reflects this in multiple provisions: parties can choose their mediator, withdraw from mediation after two sessions and must give informed consent before an agreement becomes binding.

AI-assisted ODR platforms pose a subtle but serious threat to this autonomy. Predictive
analytics tools, which generate “recommended settlement ranges” based on historical case
outcomes, do not merely inform-they anchor. Behavioural economics has long established the anchoring effect: once a numerical reference point is presented, parties tend to negotiate around it rather than away from it. When an AI system presents a settlement range as “data-driven” and “objective,” it carries an epistemic authority that a human mediator’s suggestion does not. The party who receives a range that undervalues their claim may not recognise the structural assumptions embedded in that figure; they may simply internalise it as the fair outcome.

The phenomenon is especially apparent in cases where one party is a repeat actor, such as an insurer, bank or online service provider and another is a one-time actor, such as a customer or a small enterprise. The former is generally better acquainted with the ODR system, has access to more data for calibrating the AI algorithms’ output and is able to put additional resources into maximizing its utility from interacting with the system. Hence, an AI-mediated dispute resolution process will not balance the playing field but rather tilt it further towards the stronger side.

The Mediation Act’s requirement of “free consent” for the validity of a mediated settlement agreement under Section 27¹¹ is rendered meaningless if the consent is shaped by an algorithm that neither party fully understands and that the less powerful party cannot effectively interrogate.

V. Procedural Fairness: The Transparency and Accountability Deficit

The right to a fair procedure is not merely a domestic legal norm; it is an international standard. The UNCITRAL Technical Notes on Online Dispute Resolution (2016) emphasise that ODR frameworks must ensure transparency, impartiality and due process.¹² Domestically, the Supreme Court of India has read the right to a fair hearing into Article 21 of the Constitution in the context of adjudicatory proceedings.

AI-mediated ODR creates a transparency deficit on at least three levels. First, the decisional logic of AI systems-even when they merely “guide” rather than “decide”is typically opaque. Machine learning models, particularly deep neural networks, operate as black boxes: their outputs cannot be traced to a clear chain of reasoning that a party could interrogate or challenge. Second, the training data used to build the model is proprietary and not disclosed to parties. Third, the platform’s terms of service, rather than a statutory code of conduct,govern the AI mediator’s behaviour-creating a private regulatory regime insulated from public accountability.

The Mediation Act mandates confidentiality of mediation proceedings¹³ ,” which, ironically, may compound the opacity problem in AI-mediated contexts: if the AI’s processing of a party’s data is “mediation communication,” it may be shielded from disclosure even in proceedings challenging the validity of the settlement. This conflation of confidentiality with algorithmic opacity must be addressed.

VI. The Legislative Gap and the Way Forward

Although pioneering in terms of recognising online mediation, the Mediation Act, 2023 does not have an architecture ready for AI. This is seen from the criteria of qualification for the mediator, the conflict of interest rules as well as the procedure for their disclosure and challenge. The forthcoming Mediation Council of India, once constituted, can step in to fill this lacuna with appropriate regulations.

Here, three recommendations are made. First, there is need for standards for the disclosure of the training datasets used, explanation of the logic behind decisions made by the machine and the historical outcome rate disaggregated by party categories of ODR platforms using AI in mediation. The requirement made by Singapore to technology providers to explain how AI impacted dispute resolution offers a useful example.

Secondly, the provision for human oversight in all cases involving AI in mediation should be incorporated into the Act. It would require the presence of a qualified human mediator in the mediation process as a mandatory step prior to finalisation of the agreement between parties. This replicates the AAA’s AI Arbitrator model where a human arbitrator approves the award recommended by AI.¹⁴

Thirdly, there is need to establish an additional cause for disputing mediated settlement agreements on account of algorithmic biases and failure to give informed consent under section 28 of the Act.¹⁵ The causes of disputes currently provided by the Act - namely, fraud, corruption, impersonation and mediable subject matter - are insufficient.

VII. Conclusion

AI technology as a part of ODR-mediated disputes is not something for the future, it is
something already happening. The Mediation Act 2023, grounded in the dual principles of
party autonomy and process fairness, faces a danger that the two pillars will quietly crumble due to the introduction of algorithms beyond legal oversight. The question "can a machine be impartial?" doesn't have a definite 'yes' or 'no' answer. Machines can reduce bias but they cannot be free from preconceptions programmed into their code. The legislation has to give a structure to this problem through the means of mandating AI transparency, human involvement and additional grounds for contesting AI-related decisions.

As India becomes a preferred place for resolving international disputes, its ODR system will be evaluated according to its efficiency as well as its capacity to do justice. Therefore, the problem of artificial intelligence mediator is, ultimately, a philosophical question of mediation's purpose - whether or not its voluntary nature will withstand the algorithmization of disputes.

References

1. Bot Mediation, ‘Bot Mediation: AI-Powered Mediation for Efficient Legal Dispute Resolution’ (American Bar Association, 31 March 2025) <https://www.americanbar.org/groups/law_practice/resources/law-technology today/2025/ai-powered-mediation&gt; accessed 14 April 2026. ↩
2. Singapore International Mediation Centre, ‘SIMC Practice Note on the Use of Technology in Mediation Proceedings’ (2023) para 4.2 <https://www.simc.com.sg\> accessed 14 April 2026. ↩
3. Mediation Act 2023. ↩
4. American Arbitration Association, ‘AI Arbitrator: Overview and Procedural Guidelines’ (2025) para 3 <https://www.adr.org\> accessed 14 April 2026. ↩
5. Mediation Act 2023, s 30. ↩
6. Ethan Katsh and Orna Rabinovich-Einy, Digital Justice: Technology and the Internet of Disputes (OUP 2017) 53. ↩
7. ‘Enhancing Mediator Neutrality Through Question-Asking’ (Mediate.com, 10 September 2007)<https://mediate.com/enhancing-mediator-neutrality-through-question-asking/> accessed 15 April 2026. ↩
8. Mediation Act 2023, s 10(1). ↩
9. Sam Sherret, ‘The Role of Artificial Intelligence in Online Dispute Resolution: A Brief and Critical Overview’(2022) 38(2) International Journal of Law and Information Technology 117, 120. ↩
10. Mediation Act 2023, ss 10-11. ↩
11. Mediation Act 2023, s 27. ↩
12. UNCITRAL, Technical Notes on Online Dispute Resolution (United Nations 2017) paras 24–26. ↩
13. Mediation Act 2023, s 22. ↩
14. ‘The AAA-ICDR’s AI Arbitrator: A New Chapter in Dispute Resolution?’ (Herbert Smith Freehills Kramer,10 November 2025) <https://www.hsfkramer.com/notes/arbitration/2025-11/the-aaa-icdr-ai-arbitrator-a-new-chapter-in-dispute-resolution\> accessed 15 April 2026. ↩
15. Mediation Act 2023, s 28. ↩