ISDS & Artificial Intelligence: A Concise Review of AI’s Impact on Investor-State Dispute Settlement-Mahdi Ghasemi

Mahdi Ghasemi - Public International Law Student at Allameh Tabatabaei University
2026/01/03

As of today, there is scarcely any person, society or aspect of real life that remains untouched by Artificial Intelligence (hereinafter “AI”), which may be regarded as one of the most significant phenomena of the 21st century, owing to AI’s capacity to store immense volumes of data, process information at high speed, address complex problems, and approximate human capabilities (here, p. 271). Among the many sectors influenced by AI is the field of arbitration, most notably, Investor-State Dispute Settlement (“ISDS”) within the framework of International Investment Law. While a considerable number of books, articles and blogs have been written on AI and arbitration (As examples, see here, here, and here), the intention, here, is to focus specifically on the specialized area concerning AI’s impact on ISDS.

After reviewing the existing literature on AI and arbitration, this post will explore some ideas aimed at addressing the question: What is the current situation of AI in the lex specialis of ISDS?

 Literature Review: Pros

Arbitration, whether institutional or ad hoc, is a swift mechanism chosen by enterprises and individuals as an alternative to judicial proceedings, either due to a lack of confidence in the courts (here) or because of the time-consuming nature of judicial processes (here). A process in which arbitrator(s) selected by the parties must often examine large volumes of evidence and documents, review the memorials, consider previous awards (if relevant), and consult doctrines, before finally rendering a binding award; typically after a lengthy process, though still shorter than judicial proceedings (here).

However, the professional legal world keeps changing, and so does the field of arbitration. Now, imagine a scenario in which an arbitrator could leverage AI, even well-known tools like ChatGPT, though their use is not recommended for “total” legal process since these instruments could start “creating” information that do not originally exist (see here). In the aforementioned scenario, the arbitrator could make more efficient use of their time by reviewing the documents provided by the parties. In other words, as noted by the author of this blog, AI has played [and plays] a significant role for practitioners in tasks requiring minimal professional intervention, such as pre-proceeding document review or legal research. Hence, the first advantage to consider is the reduction of research/review time. AI could streamline the daunting process of analyzing hundreds of pages just to find, for example, a specific quotation or a line that matters.

Another advantage the comes into mind, is the ability to predict various dispute scenarios and anticipate potential outcomes of an arbitration process. This is a benefit that could be envisioned by both arbitrators and the parties. AIs work is based on the input and data provided to them. By developing algorithms, training the models, and tokenizing the input, they can generate hypothetical outcomes of a litigation. Interestingly, two experiments have been conducted on this matter: one on the jurisprudence of the European Court of Human Rights, and the other on the USA Supreme Court (for more info see here, p. 35).

Finally, reduction of costs is another factor for which AIs could be counted on. As one author states (here, p. 319): “Rising costs in international dispute settlement have been linked to access to justice issues, especially affecting developing States and leading to the promotion of cost- saving strategies in litigation and arbitration.” Simply put, activities requiring significant human effort, and consequently financial costs, could be transformed into machine-driven processes that demand nothing more than data. Being able to manage the ‘data hunger’ dilemma means there is no longer a need to bear high costs in you arbitration process (here, p. 15).

 Literature Review: Cons

No phenomenon is free from risks, challenges, or flaws; and the same holds true for AI. Regardless of the specific field of arbitration (commercial or ISDS) where AI could be used, there are real challenges that must be taken into consideration by users; whether they are clients or arbitrators.

First, and perhaps the most important challenge for all users, is the so-called ‘black box’ problem. Longman Dictionary defines Black Box as: “a machine that works in a secret way that is not normally explained to people.” This is, in my view, the most concise and effective definition of the problem at hand. Or, as one author wrote: “black box refers to a model or system where the internal logic is not visible to the user, making it opaque and challenging to interpret.”

An essential element of every judgment or award, whether rendered by a court, tribunal, or arbitral process, is the flow of logic and reasoning. The ability to explain how, and based on which sources and principles, the decision-maker reached a conclusion on the merits of the case.

Can the arbitrator clearly articulate the reasoning behind the outcome? Are the arguments transparent and understandable? These are precisely the questions that an AI, when acting as a decision-maker, cannot answer. No full justification can be provided, at least not completely, even when considering features such as ChatGPT’s ‘think longer’ or ‘deep research’ modes.

This lack of explainability gives rise to the problem of transparency, as the readers and audience of the final award cannot fully understand what happened behind the scenes. Such opacity may, in turn, undermine the legitimacy of the entire arbitration process. Black Box problem omits the dialogue and interaction (here), and makes it difficult to predict next step by the AI in the process (here).

Another challenge posed by the use of AI in arbitration is its heavy reliance on information and the quality of the input it receives. While confidentiality is one of arbitration’s key advantages, it also becomes a challenge when AI is involved. The closed-door nature of arbitral proceedings, combined with the confidentiality clauses found in many investment instruments, has traditionally restricted non-disputing parties from accessing important information about arbitrations (here, p. 13-15), including programmers and AI developers. As noted above, for an AI to begin functioning and making decisions, it must first have algorithms established. However, the confidential nature of awards and the sensitivity of the parties involved (here, p. 7) present a significant challenge for this technology, preventing it from assembling the necessary pieces of the puzzle and rendering a judgment.

Lastly, a key disadvantage to consider in this section of the post, is the issue of bias. Needless to mention again that the input data changes the course of the answering process for AI. Imagine a situation in which the data provided to the AI, including case law and doctrinal sources, contains a hidden bias favoring a particular group, such as investors. Some studies signify that arbitrators often tend to rule in favor of foreign investors, especially those from major Western capital-exporting countries, and may even show a preference for the USA when it acts as a respondent state (here). This bias is often attributed to the structure of investment treaty arbitration, which can lack truly independent and impartial adjudication, raising concerns about the protection of sovereign authority and public resources (here). The problem of bias, namely an AI favoring investors, raises two opposing solutions: either AI should be completely excluded from the arbitration process, or, at least, not used in high-profile, precedent-setting cases; alternatively, a strict and concrete review must be applied to all data fed into this powerful tool in ISDS and other types of arbitration. Neither of these solutions currently exists.

 AI in ISDS?

In this section, the current benefits and challenges of using AI in the field of ISDS will be briefly analyzed. It goes without saying that the pros and cons discussed above also apply here; however, due to the unique nature and specific conditions of this field, a concise yet independent analysis is required.

Concerning the advantages of applying AI in ISDS, promoting objective and fair decision-making comes to mind. Despite the potential for investor-favoring bias, AI systems have the capacity to significantly enhance objectivity and fairness in dispute resolution (here). By eliminating emotional influence, remaining unaffected by external pressures, and being shielded from monetary or self-interest considerations, AI could serve as an impartial and independent tool, provided that its inputs are free from hidden biases. They are capable of handling even the most complex cases laden with hundreds and sometimes, thousands of pieces of evidence without any preference problem or other issues such as ‘double hatting.’

AI is able to analyze large volumes of data, including case histories and legal precedents, to generate insights that help inform decision-making. For example, as noted by one author, IBM’s AI Ross has been widely adopted by law firms around the world as a tool to streamline and accelerate legal research.

One more benefit worth highlighting, before finalizing the post with the cons, is smart dispute prediction. AI tools, with their extensive access to national laws, case law, jurisprudence, legal doctrines, books, articles, and emerging trends, and if sufficiently trained, the established practice in the field, can anticipate potential future problems arising from an investment treaty.

This early detection of disputes enables stakeholders and interested parties to take preventive measures, promoting proactive communication and collaboration to reduce the likelihood of conflicts arising in the first place (here). This tool could even prove helpful when parties are, before the emergence of dispute, interpreting a controversial clause in a treaty, by providing access to the travaux préparatoires (if access granted and confidentiality issue solved).

To conclude this section, I would also like to discuss the dilemma that has not been mentioned above yet carries significant implications and has consistently appeared throughout the ups and downs of ISDS: The Inconsistency Dilemma.

A long-standing challenge in the field of ISDS is the lack of clear and consistent definitions for its key standards, such as Fair and Equitable Treatment (FET), National Treatment, Due Process or Non-discrimination (see here and here for more info). Different tribunals interpret these standards differently. For example, regarding the due process requirement under BITs, the ELSI case offers one definition, while tribunals in Lemire v. Ukraine or Alghanim v. Jordan provides alternative interpretations. This lack of consistency makes it challenging to base final decisions on previous awards and rulings. On the other hand, even if an AI could identify the commonalities among previous rulings on investment standards, such as FET, a risk would still remain: being conservative and not having the initiative for each case. In other words, AI’s potential in legal applications is limited by the absence of well-established metrics for assessing the quality of individual cases, which hinders its ability to ensure justice on a case-by-case basis (here). Simply put, the data-dependent AI could not use the ‘creative decision making’ process for each case, since it gets its information and establishes its algorithm based on the previous rulings. A situation akin to the ‘l’épée de Damoclès’ in French.

Conclusion

At last, as Pierre Drai, former President of the Cour de Cassation française, once said: “To judge is to enjoy listening, to try to understand, and to want to decide” (Juger, c’est aimer écouter, essayer de comprendre et vouloir décider).

Delivering judgment on the merits of a case is therefore not merely a technical exercise; it also requires an appreciation and comprehension of human beings, which must be taken into account. An invaluable prerogative that, at least for now, remains beyond the reach of AI, and none of the previously mentioned advantages give the carte blache to use AI unconditionally or mechanically. AI has its own pros and cons, but it is still raw when it comes to human-centric legal processes.