AI in Mediation

AI in Mediation

 

Private • Impartial • Affordable 
because no argument should cost $200 per hour to resolve.

Skip the shouting, we’ll reach out to them and mediate.
TheMediator.AI guides both sides privately, finds common ground,
and proposes a non-binding resolution.

themediator.ai

 

Introduction

The above quote is from the landing page for the online AI mediator, albeit beta.

Should we be worried?  Provided the legal community, particularly those involved in dispute resolution, adhere to the current guidance on the use of AI and professional standards, I think not.

There is much in the press recently about AI generally, with The Economist discussing Anthropic’s warning about “recursive self-improvement”; Pope Leo XIV’s wider Magnifica Humanitas; the EU’s Artificial Intelligence Act; CIArb’s draft Guideline for the Use of AI in Arbitration (discussed in my earlier article on the Use of AI in Arbitration); and the IBA’s recently announced AI Committee and the Guidelines on the Use of Generative Artificial Intelligence in Mediation.  Most of it is cautionary, recognising the benefits of AI but recommending that all AI generated output should be thoroughly verified to avoid “hallucinations”, confidentiality protected and, under no circumstances, should a Large Language Module (an LLM) agent be permitted to make decision or exercise judgement.

What is required is for human judgment to be preserved, and even enhanced.

AI is moving incredibly quickly, and it is clear that some regulatory intervention, limiting where AI may take us, is needed; though, Anthropic’s proposed “pause” on development seems unlikely.  At this stage, no jurisdiction has enacted legislation specifically regulating the use of AI in mediation. 

The IBA  Guidelines on the Use of Generative Artificial Intelligence in Mediation (a “living document”) provides some assistance, noting increasing efficiency, reducing costs and broadening access to justice are all for the better, but recommending that appropriate safeguards are required. 

AI in Mediation – the IBA Guide

For parties and their representatives, the IBA Guide identifies six applications – risk analysis; mediator selection; preparing position and interest papers; preparing for mediation sessions; support during those sessions; and drafting settlement agreements.

For mediators, eight potential uses are discussed – case assessments; customising mediation procedures; enhancing understanding of disputes; enhancing analysis of disputes; preparing for mediation sessions; managing those sessions; generating options for resolution; and promoting balanced treatment of the parties.

The Guide also provides assistance for mediation institutions.

The safeguards proposed are as to be expected – protecting party autonomy; maintaining confidentiality; ensuring mediator neutrality; and providing for a balanced process, all predicated on disclosure of the proposed use of AI tools.

At times, when compared to the more general statements, like the SCC Arbitration Institute Guide, many of these guidelines read like the output from a drafting committee of enthusiasts fuelled by too much caffeine.  To be fair, the IBA Guide is better than most, but they all have a whiff of overthinking risks that largely don’t exist; or at least, are no worse than what we already manage.

In practice, to what extent can AI assist with mediation, in the context where human judgment is not completely surrendered; that is to say, when the benefits of AI are maximised to improve critical thinking, quality of output, issue identification and negotiation, without compromising confidentiality, acquiring fictions and dumbing down the process by surrendering human critical thought.

Mediation process

Mediating a complex commercial dispute raises significant interlinked issues, each of which will need to be identified, discussed and settled.  This process will usually benefit from more than the commonly accepted one-day mediation horse-trade (often stretching into the wee small hours of the following day as the mediator weaves magic).

The goal has to be a durable settlement agreement, which properly covers all the issues on an informed basis.

Before the Mediation

For the parties and their representatives, as noted above, the factual background needs to be fully understood, case law researched and advice formalised by legal advisors as a first step.  That advice would normally include a consideration of whether mediation or arbitration would be the preferred forum, and the most appropriate mediator for the dispute agreed.

Thorough preparation by clients and counsel for the mediation session is always critical to a successful outcome.  That will include:

• preparing position summaries and opening statements
• consideration of expert opinion
• assessment of each issue, along with likelihood of success
• preparing quantum assessment for losses on each issue, linked to the likelihood of success
• consideration of the best alternative to a negotiated agreement (BATNA), the worst alternative to a negotiated agreement (WATNA), and the most likely alternative to a negotiated agreement (MLATNA) for each issue, including the impact of each on quantum assessment
• modelling option, so that the settlement outcomes can be updated in real time
• testing both the reasonableness of the clients’ positions, any gaps in either fact presentation or legal position
• establishing negotiation mandate and delegated settlement authority
• preparation of a draft settlement agreement, for the most likely acceptable outcome

Transparency

Much is made of transparency in a number of the guides mentioned above, including disclosing the use of AI, which platform has been used, the prompts adopted and the extent of AI generated output.  To my mind, this level of disclosure goes too far. 

The parties are currently not expected to disclose the use of juniors, research assistants, online legal search platforms, texts, economists, cost consultants or any other experts or tools used to establish the parties’ positions.  As it currently stands, AI has become embedded in so many of the search engines, computer programmes and APP’s that it would be impossible to disclose where an AI agent has assisted. 

Adopting AI in the preparation and presentation of a party’s case adds no greater risk than current processes, provided the existing rules on professional behaviour and not engaging in misleading or deceptive conduct are applied.[1]  Any client, lawyer or other representative has no excuse for presenting falsehood in any disputes process.[2]

The more critical issue is, to what extent does AI create inequality in the process and how can the mediator benefit from its use?

AI Enhancement

The mere fact of dispute carries with it a disagreement of fact, or legal entitlement – in the construction industry, that will invariably mean the interpretation of contracts and the application of the contract to those facts.

AI clearly provides tremendous advantage in easing the tedium of much of that process.  The most valuable output is collating the relevant facts, identifying the critical issues and modelling the quantum, or similar rights, against the likelihood of success.

However, providing a cold analytical summary (without the wider factual background) and drafting an agreement based on probable outcomes won’t address the emotional investment in the matters in dispute, or manage the parties’ wider interests.  The parties need to be heard.

If we accept that the parties’ positions can be well and efficiently organised, collated and presented using AI, without sacrificing human input from the clients themselves and their legal advisors, then where is the issue?

Disclosure

For a mediator, the first issue is disclosure.  Without delving into the detail of how AI is to be or has been used following appointment, it is helpful in the current formative stage of the use of AI, to have the discussion.  Are the parties using AI either in preparation or during the mediation session.  The extent of disclosure is up to the parties, but to at least establish that AI may have been, or will be used in the mediation, can reduce, if not remove, any suggestion of power imbalance.  This discussion need go no further than recognising that AI tools will most likely be used, and may well be used by the mediator in the mediation sessions.

It is not really material for the mediator to know more.  Each of the parties is free to engage in the mediation how it wishes, short of deception or derailing the process itself.

Preparation for the Mediation

In advance of the mediation, the following will be discussed and explored with the parties and/or their counsel:

• the appointment agreement, and costs
• the nature of the dispute
• the issues in contention
• the relief sought
• the mediation process itself, and what to expect
• who is likely to attend
• the distinction between positional and interests based bargaining
• the extent of preparation by the parties for the session, including preparing models assessing likelihood of success and likely settlement options
• consideration of BATNA, WATNA & MLATNA, and the strength of the party’s position on each issue
• the expectations of the opposing parties, and potential for compromise
• AI can be extremely useful for the parties in this preparation.  For mediators themselves, perhaps less so.

Taking a common construction dispute over variations as an example, a number of issues will arise:

1. were the items claimed properly “variations” in terms of the contract, and to what extent?
2. what was the impact of the variation on the critical path?
3. is the contractor entitled to an extension of time?
4. what are the cost implications?

Each is interlinked.

Mediation Session

While much of this can be developed on a whiteboard (and there is much to be said for building each issue, the claim and any agreement with the input from each of the parties as the whiteboard is filled), technology can also help.  If the mediator has developed a model, projected into a screen for each party to review as it is developed, the impact of each agreement or compromise, or simply the gulf between the parties, can be developed in real time.  An AI model can ensure that each amendment is carried through to related issues.

While tempting, preparing such models in more than outline or framework in advance is unlikely to gain traction.    At most, such a model would be a whiteboard on steroids.

Perhaps more prosaically, an AI agent could easily be used productively to draft the settlement agreement, under the instruction and review by counsel.

Conclusion

The AI controlled mediation, proposed in the quote above, falls well short of what is to be expected of mediation, at least for complex disputes and those with high emotional quotient.  It also has the potential to dilute or completely circumvent the application of professional standards expected of lawyers and most other professionals engaged in dispute resolution, when managing the parties’ wider expectations.

It also provides no comfort on the proliferation of hallucinations which seem to plague most LLM agents.

I do not hold to the complex obligations for disclosure contained in most guides on the use of AI in either mediation or arbitration.  The tools and services used by clients and counsel pre-AI were under no such obligation; AI is just another tool.

However, at the initial stage of the mediation process, discussing whether AI can assist in the mediation session strikes me as helpful, and even inevitable, once there is greater understanding of how the parties might benefit.

For mediators, like the whiteboard, modelling the issues, likelihood of success and quantum achieved in real time using an AI agent projected on a screen during the mediation has real benefit in increasing the efficiency of the process and the durability of any resulting agreement.

Perhaps of greater significance is that clients are increasingly expecting arbitrators and mediators to use AI tools to bring down costs and increase efficiency.  That should be the most persuasive point of all.