Customizing Dispute Settlement: The AI Arbitrator & Optional Appeals at AAA-ICDR

Setting the Stage: AI Arbitrator and Optional Appeals at AAA-ICDR Welcome dear listeners to Confabular, the podcast where we confabulate about international arbitration. Confabular takes you beyond case files and clauses into the heart of international arbitration innovation and insight. I'm your host, Piyush Pandit and get ready for another deep dive with a specialist shaping the field. Today, I'm thrilled to continue our second season with another extraordinary guest. Today we have with us Miss Tara Gopalan who is Singapore based Vice President of AA ICDR for Asia, Africa and the Middle East. Bridging the gap between traditional dispute resolution and the frontiers of the AI driven justice. Based in Singapore, Tara holds a massive portfolio. She leads ICDR strategic operations and case management not just for Asia, but for Africa, for Middle East and the Oceania. Tara is a formidable legal mind. She holds LLB with honors from University of Bristol and NLLM from University College London. Before taking the helm at the ICDR, she spent years in the trenches of dispute resolution as is the Associate Director at the top tier TSMP Law Corporation. As an admitted advocate and solicitor of the Supreme Court of Singapore, she brings deep practical experience in complex commercial litigation and international arbitration to her current leadership role. She joins us fresh of the use of Icdr's India conference where we last met. She was the key voice on the intersection of technology and law at the conference, not only just the organizer. And she's here to discuss how the organization has radically been rethinking the menu of dispute resolution. So, Tara, welcome to Confabula. Thank you so much for joining us. Speaker 2 Thanks for having me Piyush. Speaker 1 Now, before we dive in, let's set the stage. In the world of arbitration, we are often torn between two competing desires. We want it faster, but we also want it right. Today, we are exploring 2 bold initiatives by ICDR that pull in exactly these opposite directions. First, we'll break down the AI Arbitrator, a tool launch recently specifically for documents only construction disputes. It promises unprecedented using a human in the loop system, but does it sacrifice the human touch? That's what we are going to discuss today. Then we'll pivot. 2 optional appellate arbitration rules of the ICDR. This mechanism challenges the sacred cow of arbitration finality by allowing parties to appeal material and prejudicial errors of law. Now it is a fascinating conversation about trade-offs between algorithmic efficiency and the safety net of a second look. The Human-in-the-Loop Design of the AI Arbitrator Now starting my first segment about the AI arbitration. Tara, the launch of AI Arbitrator recently has made headlines and you have spoken very much about it. And I remember you speaking briefly about it during the Icdr's India conference as well. Now we know it's an opt in and documents only constructions dispute arbitration, but can you walk us through the human in the loop architecture as I mentioned in the introduction? And how does AAA ensure that human arbitrator isn't just rubber stamping the AI's draft, which I think is the most difficult conundrum with an AI arbitrator? Speaker 2 Great questions, Piyush, as always. Let me kick off by maybe explaining the structure in which the AI arbitrator operates because the rubber stamping concern may be somewhat A mediated once you understand the procedural architecture. So the AI arbitrator operates under a separate rule regime called the AI LED arbitration rules, and this is administered exclusively through the Digital Dispute Resolution Centre platform, which is for us a global first true AI native case management platform. So when the parties opt in, they're really selecting a very defined procedural track of the AI LED arbitration. At launch, as you pointed out, that track is intentionally limited to two party documents only. US domestic construction disputes a whole bunch of different caveats there. Essentially no evidentiary hearings, no live witnesses. That constraint is deliberate because these disputes are predominantly structured around documentary reasoning exercises under the AI LED arbitration rules. The case progresses through a very tightly sequenced set of submission stages, basically claimant files. The respondent has 10 business days to agree to the use of the AI arbitrator as well as to respond submissions, then proceed and fixed 10 business day intervals. Generally, after the submissions close, there is a mandatory party's feedback stage. He's review and comment on the AI summaries of their claims, evidence, exhibits and legal authorities. That stage exists because fairness is procedural as well as substantive. Parties must see their case accurately reflected before the analytical reasoning proceeds. And I would argue that's a step that's actually superior to current arbitration and litigation because no arbitrator or a judge is generally going to ask you have I quite understood your case in exactly the way that you framed it. But it's only after this step of the feedback exercise that the AI performs its reasoning tasks. And let me be clear, it's not a single AI. There are multiple agents that we've pulled together to be able to perform the work. So broadly speaking, the AI identifies issues in dispute, evaluates evidence against those issues, applies the governing legal framework, and prepares a draft award with brief reasoning. So this is a very US style arbitration award, right? It's trained on actual AAA construction awards about 1500 or so, and it's calibrated by experienced arbitrators. So after the AI generates that draft award, the human arbitrator reviews the reasoning chain as well as the draft award. Under the rules, the arbitrator may then approve the award, modify the award, rewrite any part of the award or the whole award before signing off on the actual award. The award does not exist as a legally enforceable instrument until the arbitrator signs it. So there are a couple of additional safeguards on top of this broad architecture. First of all, the human arbitrator who is in the loop is selected from our roster, right? That someone who is a AAA trained arbitrator has administered AAA cases gone AAA training, it's requirements of AAA annual education requirements. Most importantly, you still have all of the standard AAA disclosures that have to be made when an arbitrator is appointed, as well as signing up to the oath, which is basically an oath to comply with our code of ethics that we impose on all of the arbitrators. After this phase, again, same as with any AAA case, the parties have a right to object to the appointment of the human arbitrator and the AAA may remove and replace the arbitrator for bias or lack of independence. Same structure as any traditional arbitration. Same thing with a AAA case. The award must be issued within a fixed period of time, right? So I know other institutions you possibly get extension on extension for for whatever time. That's in the rules. AAA cases you'll scarcely see an extension. You probably have to be in the hospital for me to give you some sort of extension on the production of the award. So in the AI LED arbitrator cases, the award is issued in five business days after the feedback stage closes. Absent unusual and exceptional circumstances. There is limited correction for clerical or calculation errors, same as a standard arbitration process. The platform. One thing to note is that it operates in a closed environment. So case data that you input into the system to be able to perform your case, it's not ingested into the foundation models and performance of the awards etcetera. It's benchmarked against human awarded construction matters for us to be able to understand how the AI, you know, with the human in the loop is really performing vis A vis actual data that we have on a century of construction arbitrations. So governance oversight, bias testing, all of this is ongoing. And I'd actually tell you bias testing is something that we're super obsessed with, right? You think about AAA domestic construction in the US. Construction rules are basically used in every construction project in the US because it is the standard form rules for the American Institute of Architects, right? So imagine you have super large construction company building, you know, 10 towers of condos, right? And super large construction company maybe has used AAA in 200 other projects that they've done over the last 20 years. The one thing that you want to be careful about is that just because they might, you know, win more regularly that they don't, that the AI doesn't think that this person, so this person must therefore win this case because in all of the data that I've been given, they've won the vast majority of the time, right? So bias testing is something that's incredibly complex, beyond my comprehension as a lay lawyer, but I'm telling you, this is wildly fascinating to me to understand the metrics that are involved, but I digress. Beyond Rubber Stamping: AI Arbitrator's Ethical Future and Access to Justice So let me come back to your original question on rubber stamping. Now, I tell you, Piyush, it's not a new question, right? For at least a century that we've been doing this, parties have been asking, courts have been asking, the institution has been asking, are arbitrators rubber stamping even party submissions right. That has been the constant conundrum. But I'd say this, the arbitrators that we put and that we allow to sit within the AI LED arbitrator, they've been specifically trained to handle this caseload. They are hand picked arbitrators. We're not going to let any Tom, Dick or Harry use it because this is a product that we actually believe in and that we actually think has the potential to change the face of dispute resolution. But in the same way that arbitrators over the last century of the Triple E have analyzed the submissions that are before them, these arbitrators that we place as humans in the loop, we'll continue to analyse submissions that are placed before them. The architecture that we employ structures the analysis, but it preserves human adjudicative authority. And I think early testing that we've done in some of these documents, only construction disputes do show this cost and time saving, but do also tell us that the arbitrators are aware of things like algorithmic bias. You know, and there's a tendency to therefore then militate against this by being a little bit more vigilant in these types of cases. And we've been able to do that because the arbitrators that we deploy to these cases are intentionally selected, right? If you're someone that doesn't understand the pitfalls of the use of artificial intelligence, if you're someone that doesn't live and breathe the AAA code of ethics, you're unlikely to be an arbitrator that I can safely deploy to the AI Arbitrator platform. But these guys that we put on it, these ladies that we put on it, given that it's International Women's Day today, are truly best of the best. And we intend to do that as we roll out the AI arbitrator into other case loads moving forward. Speaker 1 I think it's it's a very long road ahead and we've only treaded like AAA has only traded very fraction of it so far. And I think it can go really far in terms of its capabilities because I think it is really limitless. Speaker 2 I, I think that the one big thing is thinking about how this tool elevates access to justice with the cost savings in particular that are involved. Regular people have disputes of these kinds of sizes, right? This is going to be dramatically like such a major shift in their ability to get justice for these types of cases. And that's exactly what it is that we want to be able to see, right? Some tools are built for the large cases. Some tools are built for the regular human being, for you and I, to be able to deal with disputes that we have on a day-to-day basis. Speaker 1 And, you know, I think it's also a great opportunity because my listeners are usually lawyers or people who are getting into law. But this particular episode, I think would interest engineers also who are working in the field of AI, that possible possibilities are limitless if you intersect law and AI and see where it goes, right? Speaker 2 There's this quote, and I can't remember quite who it's from, but it it's something that's ingrained in the triple AAI town halls. And so we keep repeating this every quarter or so, which is that the bounds of AI are really limited by the bounds of your imagination. And as we slowly start to expand the way that we think and expand our imagination, we're going to see it scale in different industries. It's going to end up doing different things. The difficulty is, though, understanding that it's a tool that's in development and that it'll get better as it goes along. But more often than not, the first couple of iterations aren't the best and you'd have to learn to be able to work with it, to let it give it time essentially to develop, to get to a point where it's actually useful. And also, I guess the the separate issue is understanding what it should not touch as opposed to what it cannot, right? The ethics and the morality of that conversation, which is deeply fascinating. Speaker 1 Yeah, and I wonder how far is it when AI acquires some sort of conscience, when we have to actually think about these questions of morality and ethics of things that, you know, AI starts doing. Recalibrating Finality: Structure and Data of Optional Appellate Rules So shifting gears, Tara, now let's talk about optional applet rules. The finality of awards has always been arbitrations, double edged sword I feel and these rules allow for to a review of material errors of law are clearly erroneous facts. In your experience leading the Asia division, are you seeing a shift where parties perhaps in high value infrastructure or state contracts are willing to create speed for that extra safety net of appeal? Speaker 2 Great question again. So we've spent some time looking at the optional appellate cases to actually understand what industries they touch, what are the claim values involved. And the data is actually quite interesting. So let me maybe just organize this in, in structure first, which is the optional appellate rules actually just equence, finality and arbitration. They don't replace it in any big way, right? If a notice of appeal is filed within 30 days, the underlying award is not treated as a final award for the purposes of enforcement under the New York Convention or otherwise. Enforcement timelines are told. Any already initiated enforcement proceedings are state. What then happens is a separate appellate tribunal is appointed, typically A3 member panel drawn from a designated appellate roster. The grounds of review, as you pointed out, very limited, very defined material and prejudicial errors of law, clearly erroneous determinations of fact, essentially what you'd see at very high levels of appeal. In most jurisdictions, the appellate record is confined to what's before the original tribunal, so it's unlikely that you're going to have new evidence or new arguments. There is also no rehearing and no remand back to the original tribunal. What the appellate tribunal can do is adopt the award substituted in whole or in part, or request some sort of very limited clarification from the parties. It must issue a written decision with an explanation. When that appellate decision is issued, that appellate decision becomes the final award for the purposes of enforcement. So structurally finality has been recalibrated, isn't removed, it's just allocated to a second stage. So coming then into the data between 20 T-21 through to 2025 T 7% of appellate matters proceeded to a decision under the OAA rules. Of those decisions T 9% adopted the underlying award in full and 41% substituted in whole or in part. So it's a functioning internal correction. It's not going to be the case that they rubber stamp the original award. There's a decent prospect of success. What surprised me most is that it's cross sector financial services, technology, commercial real estate and franchise matters were sort of the top four types of cases that we saw between 2021 through to 2025. Some really cool other data points that I will share with you very briefly is that appellate proceedings will also split between decisions rendered on the written record alone. So 41% documents only and those involving an oral hearing, which is essentially 59%. So this again kind of shows you that there's a lot of flexibility in terms of how you structure that appellate process. Why Optional Appeals Enhance Arbitration: Speed, Safety, and Complex Cases One major sort of thing that's happened with the optional appellate arbitration, we finally had a public court decision that has enforced the appellate arbitration award. So that's in the case of FIH Hong Kong Limited and Emblem Solutions LLC, which is a 2025 U.S. Federal Court decision that confirmed an appellate arbitration award between a Hong Kong based manufacturer and AUS counterparty, essentially enforcing under the New York Convention. The dispute concerned allegedly defective mobile devices that were supplied to Emblem Solutions, giving rise to indemnity and commercial loss claims. The ICDR Tribunal issued a final award and pursuant to the party's arbitration agreement, there was an appeal under the OAA rules and Appellate Toilet Tribunal that was seated in Dallas, TX, reviewed the original award and following that review, adopted the underlying award in part and substituted in part, issuing a final award on appeal in September 2025, and when Emblem sought confirmation of the final award, the US District Court for the Northern District of Texas. Treated the appellate award as opposed to the underlying award as the operative arbitral award for enforcement purposes. So that's something that we're, we're really, really, really excited about. The question that I often get in relation to appellate arbitration is OK. I mean, I'm, I'm glad that I have this opportunity to, to, you know, get some finality. Maybe, maybe not, that's something I might want. But isn't this just going to drag out the proceedings, right? Isn't this going to be essentially 2 arbitrations? You're going to go into like multiple years of processes. And that's where I think coming back to the data really helps, right? So looking at just ICDR cases, median time to an actual award from the time you file your arbitration is 25 months for large value cleats, and the appellate phase when you tack that on top of the ICDR arbitration is about 3 months long. That's it. 20.5 + 3 takes you to 23.5, which I'm telling you is faster than the vast majority of institutions that I compete with in my market. So I would say the choice isn't really just speed versus safety and thinking about whether or not you want an optional appellate arbitration. Because ultimately I think the safety and the time that's taken to ensure that safety, that three months is less time that you would spend attempting A backdoor argument of the merits of the dispute through some sort of, you know, long drawn natural justice type element. Chances are that's not even going to be successful because the grounds for review under the FAA, the New York Convention, the Model Law, you know, the Indian Arbitration and Consolation Act, all very limited grounds for review. I think the question is really, do you want arbitration to have this internal correction mechanism in the same way that litigation has that internal correction mechanism? And the reality as to why you might want that is because these cases are getting more and more complex. You have these construction arbitrations that have 250 causes of action. And it's not just about legal complexity, but sometimes it's a precedent case. It might be on your standard terms and conditions, and you've then got a decision through the enforcement process. That decision has become public. If you are a giant business and you've got millions of customers, that's then going to implicate, you know, your existing contracts going forward, your current terms of conditions, unless you're able to do some sort of retroactive amendment, which is controversial, right? So these are the points at which you might want that internal correction mechanism. It does also allow you to customize your process in slightly different ways, right? So if you think about it, you've got 3 member tribunals, you've got single member tribunals. Maybe you think initially, you know, I don't want the complexity of a three man tribunal at the first instance. I just want a sole arbitrator to determine the dispute at first instance. But then you realize, OK, there might be discrete issues within the 250 causes of action that could benefit from a collaborative delivery process between a member of between three members. And that then allows you to make your costs a little bit more proportional, right? You're not going to have three members across the full gamut of the initial arbitration or even if you use one of the AAA's streamline panel options. But you might just have them on really discreet issues that you take up on appeal if you believe that the arbitrator at the initial state has committed some sort of material and prejudicial error. So I think that error correction mechanism is something that we are going to see more adoption of. And what I really like about the rules, and I encourage everyone to read them and give me your thoughts about it, is really the elegance in which we've moved finality from underlying award to that appellate award. The drafting of that continues to blow my mind. Speaker 1 You know, as I, as I understand it, it does not only add another layer of checks and balances and error corrections as you said, but it is also counter intuitively faster instead of going in loops with probably traditional courts and it takes off the burden of the local courts in the national jurisdictions as well. So it adds efficiency without compromising the time duration of the dispute. Speaker 2 I think that's absolutely right. You know, if you think about how long challenges take to be determined in national courts that tax on such a significant period of time, in some jurisdictions actually at a domestic arbitration level, you might even be able to get an appeal on a point of law to your national court. More often than not, they try not to interfere with the facts. But you might have a peel on a point of law, as you do in Singapore, for example, under domestic arbitration legislation. So this just takes that error correction, keeps it within the arbitration, which is really where the parties always wanted to be, right? The purpose of the courts was really to be more limited as a supervisory body, as a body that helps you with some sort of enforcement. It was the whole reason why we all decided to get out of the courts and into arbitration in the 1st place. And so I think this is just a way of safeguarding that original intention of keeping the arbitration, the substantive part of that process with very limited cheerial intervention by national courts. Extreme Customization: The New Standard for International Arbitration So Tara, in the very beginning I mentioned how ICDR is trying to make the dispute resolution menu based where the parties can be very flexible and you know, customize the justice that they want. And that's exactly what my next segment is that if you look at these two initiatives together for speed and cost or applet rules for correctness and safety, it seems ICDR is moving towards a menu based arbitration world or a model. Now, do you think the future of international arbitration lies in this extreme customization where standard arbitration becomes a thing of past? How do you how do you see it? Where is this headed? Speaker 2 So I think one of the things to sort of bear in mind with the AAAI Cdr is that we already administer a very differentiated system, right? That is by sheer virtue of the scale at which we operate. In 2025, AAA administered over 580,000 fresh cases. The ICDR alone received 725 new international arbitrations and 73 new international mediations, and that's involving parties from 96 countries. And if you look at adr.org, which is our website, you're not going to find a standard set of arbitration rules. You'll find commercial arbitration rules, construction industry arbitration rules, international dispute resolution procedures, consumer rules, employment rules, labor rules, healthcare rules, mass arbitration rules, expedited procedure, emergency measures. My favorite example is construction, because within construction alone you've got construction industry arbitration and mediation rules, dispute avoidance and resolution board procedures, initial decision maker procedures, fast track procedures, supplementary fixed time and cost tracks. There's so many different menu options, as you put it. All of these caseloads reveal sort of sector specific patterns and that's data that we continuously publish on the AAA case, though some of these are specialist sectors in terms of the fact that there are highly technical or nuanced issues that arise, whether that's construction, energy, technology. But then technology you've also got different subsets because like crypto for example is highly different subset, right? And we believe that parties have always demanded that sort of sector specific differentiation. It's why we have specialist panels, for example, where these arbitrators only take on certain types of cases and that other arbitrators who are not part of the panel may not be eligible to receive an appointment from the AAA if they're not part of that specialist panel. So what we're really seeing is that there is now greater understanding of the fact that arbitration and dispute resolution broadly cannot be A1 sized fits all model, whether that's domestic or international, right. You do have to spend some time thinking about what is the right institution within the right institution, what are the right rules? Is your process one that should be a multi tiered dispute resolution process? How should that multi tier system work? Does it need to be a condition, Preston to move on to the next stage? Should it not be? Do you want to include or exclude the ability to seek interim relief? Should you only be seeking interim relief from a tribunal? Should you be going to the courts? What does your arbitration architecture provide for? And we recognize that disputes are heading in a direction where this level of thought needs to be put into every type of dispute that you see. You can't go to a rule book and copy a model clause. It doesn't work. It hasn't worked for years and there is no indication that it is going to work moving forward. That's why we set up this thing called close Builder AI, right, which builds on a database of cleaned clauses, different sort of permutations, etcetera for parties to be able to work through what is the right permutation for their case. Obviously because it's a AAA product, it only uses AAAI Cdr rule sets. But that's I think the beauty behind it because not many institutions have differentiated sector specific rule sets. And I think beyond just looking at the different rule sets, beyond just looking about whether or not you want mediation or a dispute avoidance and resolution board involved, you've now then got to think of slightly more complex things like optional appellate arbitration. To what extent do you want AI involved in that decision making process? Right. These are all questions that we now have to structure as part of the thinking, not just in terms of institutional design, but in terms of the actual procedural design that tribunals in discussion with the parties deploy onto actual live cases. So yes, I think international arbitration has to be far more customized. I think it is anathema that we are copying model clauses from the back of rule books without thinking about the context in which that dispute sits, especially as we get far more globalized with far more jurisdictions involved. I'm not just talking about counterparties from disparate jurisdictions. Right? Performance alone can take place in multiple jurisdictions. You've then got regulatory compliance, which is a whole different track. You might not even be within the EU and still have the EU legislation applied to you, right? And then you think about enforcement where the assets actually located. If you're a shell company based in Singapore, you might not have your bank accounts in Singapore. You may be somewhere in the Cayman Islands, they may be in India, right? So you've got to think about all of these things as you're structuring that Dr. architecture to be able to decide what's the right fit for the case that's in front of you. It can no longer be A1 size fits all. Where we sort of think below 5 million, below 250,000, we go to expedite it. It proceeds in one week. Anything above 250,000, it's not expedited. It is standard arbitration. To my mind, I don't think that ought to be such a thing as a standard arbitration. I think the arbitrators who think of the usual procedures or you know, what usually happens in arbitration cases, I'm not entirely sure that you're actively applying your mind to customize a process for that particular set of dispute that is before you. And I think that's something that we need to go back to as the reason why commercial parties chose arbitration, the reason why they fell in love with arbitration, the reason all of us have jobs today. That is something that we need to honor so that that love for arbitration doesn't become an unrequited love. We continue to be in touch with what the commercial users demand and develop processes, improve processes, not for the sake of doing that, but because that brings us more in line with what the market demands. And I'm very proud to work at an institution that puts those actors first to think about processes that work for them and to therefore calibrate in a way that helps the parties get to where they want in as cost and time efficient a way as is possible. Speaker 1 So I think I think that that was really interesting and now that I want to discuss more and I have so many follow up questions, but I think that will have to wait because we are now at the conclusion of this particular episode. So I think if there's one take away from today's conversation with Tara is that arbitration is no longer as you rightly said, a one-size-fits-all product. It is more and more customizable and institutions like AAAI Cdr are pioneering that customizing in the process so that the commercial entities who love the arbitration and we who are working in the in this field have more and more work to do and more and more fun doing the work that we have to do. So whether you are a construction contractor needing a quick AI assisted decision to free up your cash flow, which is stuck, or a multinational needing the appellate safety net for a better company dispute, the future of our field in arbitration is clearly about customization, as you rightly pointed out. Now, a huge thank you to Tara Gopalan for joining us from Singapore and giving us a week behind the curtain of ICDRS working and strategy and rules and everything. And this concludes today's episode of Confabularb. As arbitration evolves, so must we. And thanks for being part of this journey and for tuning in. Keep engaging, keep confabulating. And until next time, I am Biyush Pandit signing off.