[Editor’s note: Antonia Birt, a partner with Reed Smith based in Dubai and Abu Dhabi, is a partner in the firm’s Global Commercial Disputes Group with over 14 years of experience in the MENA region. She advises on the full cycle of complex, high-value commercial and construction disputes, from settlement negotiations, mediations and other alternative dispute resolution methods to formal dispute resolution proceedings and enforcement. Rebeca Mosquera is a senior associate at Reed Smith’s New York office and the current president of ArbitralWomen. She is a dual-qualified attorney in Panama and New York, with nearly 15 years of experience in investor-state disputes and international commercial arbitrations. Her experience spans various industries, including renewable energy, oil and gas, infrastructure, construction, technology, real estate and telecommunications.]
Mealey’s International Arbitration Report spoke with Antonia Birt and Rebeca Mosquera about their experiences using AI and perspective on how AI tools can be used to improve the efficiency of international arbitrations.
Mealey’s: How did your career paths lead to your involvement in international arbitration?
Antonia Birt: For me, it was very much by chance. I did not have a plan to end up in international arbitration at all. I wanted something that would allow me to work across borders, and arbitration naturally fit into that. I had the opportunity to work on a couple of arbitration cases, and then I realized how that perfectly suited my international ambitions, how much I enjoyed the work and that each case was different from the last, and it very much grew from there.
Rebeca Mosquera: I have a similar story in the sense that it was coincidence. It came to me by way of Alaska, not a place typically associated with global legal disputes. After graduating law school in Panama in 2005, I moved to the United States and found myself in Alaska, working for Shell Oil. After a few years of working for Shell, I decided to return to private practice. It was then that I had the opportunity to work on one of my first investor-state disputes, the case piqued my interest and opened my eyes to the complexities and nuances of international arbitration. That experience fueled my desire to delve deeper, and I moved to New York to get an LLM in international arbitration at NYU. Since then, I have been involved in over two dozen arbitrations and several international disputes under different arbitration rules. It took a journey from Panama to Alaska to New York, to finally end up on the global stage of international arbitration.
Mealey’s: What has your experience in general been with utilizing AI to improve efficiency in handling arbitrations?
Mosquera: My experience with AI in arbitration has been transformative. AI tools have streamlined many aspects of the arbitration process, particularly in document review and case management. By automating more tedious tasks, AI allows us to focus on the substantive elements of a case like strategy and argument development. For instance, AI can sift through vast amounts of documents, identifying relevant documents with incomparable speed and accuracy.
I am using AI in cases where clients have authorized its use. Additionally, at Reed Smith, we are testing a tool in international arbitration, Jus AI, which enhances summarizing cases, especially awards. Awards in international arbitration can be so extensive, of around 300 pages, and to have a tool that summarizes cases within minutes, while it might take an associate a day to read, and summarize, makes a difference. It goes without saying that a human being is invaluable in ensuring that the final result is accurate. But I like to say that it’s been a game changer in the way I distribute my time and the time it frees up for other more substantial tasks in the case.
Mealey’s: In how many cases have you been able to use AI, and can you quantify the time or amount of money saved by using these tools?
Birt: I use AI for most cases that I’m working on, probably since around earlier this year. We’ve used document production tools for some time, so I will not address those right now. Focusing on the generative AI tools that have come out recently, Rebecca has already mentioned some of those arbitration-specific tools we use within Reed Smith. We also use Harvey, which is a more generic legal AI. I was on the pilot for Harvey last year, and we signed up to this AI tool earlier this year. In appropriate circumstances, we also use more generic generative AI such as ChatGPT.
There are various tools that I use for my work that save me significant time on a daily basis. I am not aware of any studies, at least at this stage, that look at how much time is saved on a statistical basis. Generative AI in legal work is comparatively nascent. From my own experience, I’d say there are days on which I’ve saved at least a couple of hours using AI.
AI can also be very useful for translations. I work across several languages, and while we would not rely on a translation that has been prepared by AI, it is useful to have an immediate translation which allows you to zoom into a particular area.
As another example, when we are reviewing a large selection of documents, AI can be really useful at summarizing those documents or identifying certain clauses that have been referred to in those documents. Again, we cannot rely on it; we still have to review the documents, but it is a significant time-saving measure in terms of identifying the relevant information quickly, subject to priorities.
The other useful AI tool is drafting. Depending on the particular matter and the agreement with the client, there are various time-saving tools where AI can draft certain sections, summaries or emails on the basis of information we provide. Again, any draft is subject to review by the legal team, but it can provide significant time savings in terms of having an initial first draft.
Overall, I’d say there’s no precise number, but the efficiency gains are very significant.
Mealey’s: How reliable are the AI tools you work with? Is Relativity the best option for discovery, for example, or are there other AI tools that could be helpful to attorneys?
Mosquera: AI tools can be remarkably reliable, but I believe part of the skill is to know what to input to receive accurate and relevant output. At the end of the day, AI output always needs to be confirmed and reviewed by someone, in our case, an attorney.
With respect to Relativity or e-discovery more generally, there are different types of AI being used. One type of AI that has been around for decades and nearly everyone used is predictive coding. It learns from human decisions and can then replace manual review into coding. I’ve used Relativity for this — it’s effective, but obviously it’s not the only option on the market.
Another type of AI tool I have used is Brainspace, which is a predictive coding tool integrated into Reveal, a leading competitor to Relativity. I think it’s really important to use the AI tool that best fits the specific needs of the case. Now, with generative AI, Relativity has its gen-AI tool called aiR. Though I haven’t used it yet, I’ve heard the tool performs relevance categorization with even higher recall than traditional predictive coding. At this point, I know that Reed Smith is testing Relativity aiR and, in general, testing other tools.
One of the downsides of generative AI for e-discovery is the cost of it. As mentioned, it becomes crucial to choose the best tool that fits the needs in a case. As for international arbitration, there is usually very limited document production. So it boils down to which AI tool is best suited. If a case requires the review of approximately 100 documents of more than 20 pages each, and only one of those documents is 300 pages, Harvey may be an appropriate tool to interact with the document depending on the information needed. In construction arbitration, where parties tend to have more documents, and the documents are very technical, Relativity may be more appropriate. It really depends on a case-by-case base.
At Reed Smith we have the RED team, Records and E-Discovery team. This team is responsible for testing several generative AI tools before we present them to our clients. My personal experience with Relativity has been good so far, but it’s important for attorneys to stay informed about all relevant tools in order to determine which to best use for a specific need within a case.
Mealey’s: Are there any ethical concerns posed by using AI for discovery, correspondence or other tasks?
Mosquera: If I had to group them, I would categorize them into three main ethical issues when using generative AI on a case.
The first category is confidentiality. Some free or open-circuit AI tools do not offer automatic protection for confidentiality of information that is input into the system. They use all the information to train their tool, and there is a potential that the input used, which may contain confidential information, can be used for future responses the tool furnishes to others. As we have a duty to protect client sensitive information, we need to be cautious of the information we input in open-circuit AI tools.
The second category is accuracy, which is why the output needs to be checked and confirmed by a human. Generative AI can hallucinate. It can provide you with inaccurate information that doesn’t exist because the tool wants and is trained to give you an answer. Several attorneys in the U.S. have already been sanctioned for including AI-generated wrong information and fake citations. The Mata v. Avianca case [F. Supp. 3d, No. 22-1461 (PKC), 2023 U.S. Dist. LEXIS 108263 (S.D.N.Y. June 22, 2023)] became the poster child of what not to do when using generative AI. Two other duties were infringed in that case, the duty of candor to the court and to opposing counsel and the reasons were simple. The attorneys provided wrong information and also insisted that the AI-hallucinated cases did indeed exist, causing the court to waste its time dealing with those. This case, however, says more about the ethical standard of the attorneys than any flaws of an AI tool.
The third category are the certain restrictions that are imposed. Many clients, law firms and courts impose restrictions on using AI. In the wake of Mata v. Avianca, some courts issued local rules requiring that everyone appearing before them disclose whether they used AI. These turned out to become more of blanket statements.
For international arbitration, in particular, the Silicon Valley Arbitration & Mediation Center (SVAMC) recently launched the Guidelines on the Use of AI in Arbitration. I believe that as the use of AI becomes more predominant in arbitration, we will see these guidelines more frequently reflected in procedural orders, precisely for accountability purposes, to address ethical issues and other arising concerns that may become concerns to clients, firms, courts and arbitral tribunals.
Birt: I agree with Rebeca, and because of these ethical issues, transparency is very important, particularly with respect to clients. Lawyers should make sure that our clients are aware when and how AI is being used. This is something that I think is still being worked out as part of the industry, but it’s certainly something our firm takes very seriously. It’s essential that clients understand how information is being used in AI tools and what kind of risks are acceptable or not and also that clients are aware human judgment remains critically important in the legal practice and we are reviewing any AI work product.
Mealey’s: Can you walk me through how predictive legal analysis can help attorneys, and do you see such analysis as possibly contributing to efficiency in arbitrations?
Birt: To avoid confusion, all tools we’ve discussed so far are available and we’re using them. With respect to predictive legal analysis, I’m not familiar with any such tools yet. I have heard that some tools have been developed; I have also heard that some tools have been tested, but I have not yet seen that any tools are widely available.
Mosquera: That’s my understanding as well, which is why I call predictive legal analysis the crystal ball of AI. It’s forward looking, like Antonia said. What we anticipate is that these tools will potentially predict, for example, what arguments can be best put forward. Maybe we didn’t put forward a good argument and the tool suggests that, based on the documents provided, a particular argument would be very good, or based on documents you provided, there may be a 20% chance of winning, so it could suggest beginning settlement conversations. I haven’t used AI-generated predictive legal analysis yet; it’s more about what I have heard that this tool could potentially do. AI is a hot market, and we see a rise in legal tech companies, such as Harvey, JusMundi and LexArb. It will be interesting to see what the future has to bring.
Birt: From what I understand, it’s in development, and there should be tools at some point in the future, which will be really helpful. We expect to see tools that can analyze a particular fact pattern and provide analysis as to the chance of success of a particular claim or argument, taking into account data points from previous cases, including judgments that are publicly available, and awards, some of which are publicly available.
We expect to have tools in the future that can give us statistics about how particular claims and scenarios have played out. Obviously, any type of tool like that would be invaluable for settlement or strategic purposes, considering whether a particular claim or argument should be run and how it’s likely to play out. I think that’s all very exciting, but we will have to wait and see how such tools will operate.
One aspect of predictive analysis, although not legal analysis, that we are already using is related to the issue of predicting legal costs. Firms including ours are working on this — looking at costs of arbitrations, taking into account data points regarding previous arbitrations and analyzing estimated future legal costs. Obviously, that’s very useful because clients need information about costs in order to have control over their budgets. Having empirical data from previous cases is very useful and allows AI tools to assist with pricing.
Mealey’s: Would any predictive legal analysis be able to access sufficient data points in arbitration cases when most of the awards are not made public?
Birt: A problem in the world of arbitration is the lack of transparency because the vast majority of arbitral awards are not published. AI will be able to rely on public court judgments and limited publicly available arbitral awards. But we need to be conscious that we’re operating in an environment where full transparency does not exist, and that will hamper any AI efforts.
This interview has been lightly edited for clarity.