Author: Rana Sajjad*
This blog post is the final part of a three-part series on “Multi-Intelligence Justice”, a term coined by the author while conceiving the idea of Multi-Intelligence Justice which he has developed into a jurisprudential theory. In part I, the author introduced the idea of Multi-Intelligence Justice and explained the rationale behind it while comparing the strengths of both human and artificial intelligence and discussing how dispensation of justice involved more than logical reasoning and an application of the law to the facts of the case. In Part II, the broad parameters, key considerations and the practical and normative implications of the Multi-Intelligence Justice framework were discussed for the first phase of the adjudication process, the case file review.
| Jurisdiction: |
Topics: |
Part III: Human and Artificial Intelligence Through the Remaining Phases of Adjudication
As proposed in Part I and Part II of this blog post series, the goal of Multi-Intelligence Justice (MIJ) is to have an optimal combination of both Human Intelligence (HI) and Artificial Intelligence (AI) whereby AI amplifies HI to make dispensation of justice in international arbitration both efficient and efficacious. The background and rationale for MIJ was given in Part I and in Part II, the MIJ framework’s broad parameters, key considerations and the practical and normative implications were discussed with a focus on the first phase of the adjudication process, the case file review. In this final part, I will resume the phase-by-phase analysis of the interplay between HI and AI in the remaining phases of the adjudication process of international arbitration.
The first phase of case file review is followed by the second phase that involves framing of the issues and identifying the core legal and factual questions that need to be answered to dispense justice. This, in my opinion, should be a solely human endeavour without any AI involvement. This is because there is no substitute for a human’s nuanced approach, a contextual understanding of the case as well as an understanding of parties’ underlying anxieties and motivations and of course a practical understanding and real-world experience in connection with market realities and industry-specific customs/practices. Broadly, these strengths and abilities relate to a human’s emotional intelligence which is why a potential counter-argument here is how emotional intelligence is not considered the sole preserve of humans anymore because AI can mimic a nuanced approach to the case and a contextual understanding of it. But even mimicking these human abilities may not be sufficient as a lack of real, lived experience, still the sole preserve of humans, undermines AI’s ability for causal reasoning that entails figuring out the cause-and-effect relationship, an indispensable part of dispensation of justice. So while AI may be very quick and effective in identifying correlations, it may not grasp the real-world causes involved in the case or the true intent of the parties. Therefore, this second phase of framing the issues should be handled solely by the Tribunal without any assistance from AI. The MIJ framework would, therefore, specify that in this phase, there will be no use or involvement of AI by the Tribunal and hence no interplay between HI and AI.
The third phase of the adjudication process that involves conducting legal research to identify the applicable laws, rules, principles, conventions, treaties, and any relevant case laws, whether analogous or distinguishable, would be entrusted to AI to find the relevant laws in the shortest time possible. Therefore, as a first step, the Tribunal would provide all the relevant facts and articulate the fundamental issue(s) in the case to give AI the background and context before instructing/prompting it to research the applicable laws. This raises a couple of questions though:
- Whether the prompt should be a concise one, perhaps just one long sentence or a long-winded one providing the factual background in detail.
- Whether the Tribunal should articulate the prompt itself or it should have the power to assign this task to a tribunal secretary.
The MIJ framework would govern the length and wording of the research prompt/question as well as who is drafting the prompt since these questions can potentially have far-reaching implications for the research outcomes and, in turn, the final decision/award. Addressing these questions in the MIJ framework would also help ensure uniformity in the scope of the Tribunal’s role in connection with prompting AI for legal research as well as the limits on the exercise of the Tribunal’s discretion in this regard. Instead of curtailing the Tribunal’s powers and discretion, having uniform parameters would give the Tribunal clarity on the fundamental steps to be followed while exercising these powers. In addition to saving time, it will enable the Tribunal to use its human creativity to reframe and adapt these prompts to meet the specific research needs of each case without worrying about breaching any fundamental guardrails and limits on its discretion for using the research prompts/questions. The overriding consideration is optimizing the use of the Tribunal’s discretion during the research phase by striking a balance between preventing the potential abuse of any unfettered discretion on the one hand and avoiding unnecessary constraints on the full exercise of that discretion on the other.
In line with the MIJ framework’s guidelines discussed above, the Tribunal would prompt AI to conduct the research on the entire array of applicable laws of, in certain cases multiple jurisdictions, and also cover their interplay, overlap and potential conflict. AI would essentially be doing the work of a tribunal secretary, only orders of magnitude faster. One feature of the MIJ framework is that the Tribunal would not be bound by AI’s research results from the first prompt. For an optimal use of HI, if required, the human traits of curiosity and creativity should be brought to bear by the Tribunal to go beyond the initial results churned out by AI if they do not resonate with the Tribunal and do not “feel” right. The Tribunal can then explore another dimension for the research that it may either not have considered earlier or takes into consideration after reviewing AI’s initial research results. In view of AI’s speed, the Tribunal would have more freedom – and time – to do so having AI as its quick and efficient research partner. Therefore, under the MIJ framework, all the instances in which the Tribunal is not fully satisfied with the outcome of AI’s initial research results because it did not find the relevant law, the Tribunal will have the option of conducting additional rounds of research. While uniformity and consistency might dictate having a limit on the number of times this research can be conducted, considering the diversity of the cases and the unique requirements of each case, this decision should be left to the judgment of the Tribunal. However, for purposes of transparency, the MIJ framework would require the Tribunal to document the number of times the research was conducted and the reasons why AI’s initial research results were not deemed satisfactory.
Upon completion of AI’s research, before deciding on the applicable laws, the Tribunal will verify the accuracy of the research results to the extent of the laws that appear relevant to it partly because of the AI hallucinations covered in the earlier parts of this series. Verifying and cross checking these results would be especially important for jurisdictions the Tribunal is not familiar or has much experience with. The Tribunal could accomplish this either by conducting the research itself or asking the tribunal secretary, where there is one, to do so. Under the MIJ framework, the Tribunal would be required to specify how the research was verified and make observations about whether there were any reasons for verification beyond the MIJ framework’s standard requirement of verification. Finally, upon completion of the verification, the Tribunal has to certify that it is fully satisfied with the results.
The fourth phase is an important one in which the relevant law is applied to the facts of the case and the reasoning behind the decision is given. Both the Tribunal and AI will undertake this simultaneously with the caveat that the Tribunal cannot review AI’s analysis before completing its own. This is because all those innate human traits we have been referring to: the emotion, compassion, intuition and a sense of justice, may not come into play in their true spirit if the Tribunal reviews, and possibly gets influenced, by AI’s analysis/reasoning/ before undertaking its own analysis/reasoning. By restricting access to AI’s analysis/reasoning until the Tribunal’s own analysis/reasoning is completed, the MIJ framework would ensure that the HI and AI interplay is optimal and the use of one is not coming at the cost of the other.
Since we have been referring to the value and benefits of all the innate human traits crucial for arriving at a just decision, we also need to take into consideration the potential pitfalls of these very human traits. Specifically, in the realm of international arbitration, in view of the geographic and cultural diversity of the Tribunal’s members, these human traits could also be the cause of an unconscious racial or ethnic bias in decision-making. Although this element of subjectivity is unavoidable and virtually impossible to eliminate, the Tribunal can use AI to research and spot patterns within similar cases decided in the past which are publicly available to ensure that its decision is not an outlier, a potential red flag for any unconscious bias. Of course, AI itself has its own training data bias which is why a heavy reliance on AI to spot patterns to minimize or eliminate human bias might just end up reinforcing the Tribunal’s bias. Nonetheless, this is an important issue and potentially one that is accounted for in the MIJ framework in due course.
Turning back to the Tribunal’s analysis/reasoning, once it is done, it can review AI’s analysis/reasoning and either augment (an apt term from the standpoint of augmented intelligence) or revise its analysis/reasoning. In either case, under the MIJ framework, it should be clearly stated in the final decision/award and this would conclude the final phase of the adjudication.
The final phase relates to arbitral award’s non-substantive aspects such as the procedural history and summary of the proceedings and the award’s general structure and format. AI should assist the Tribunal with these aspects as there are no unique human traits that add any value to them. Therefore, the MIJ framework will provide that the Tribunal should not spend any time on these sections and aspects of the arbitral award as it would undermine the whole objective of the MIJ framework of not just making the interplay of HI and AI optimal but the use of HI and AI in their individual capacities optimal too. The one important requirement/guideline of the MIJ framework would be for the Tribunal to document the specific prompts/instructions given to AI to work on and/or improve these non-substantive aspects of the arbitral award.
In closing this three-part series on MIJ and its framework, it is pertinent to mention that while AI may already be partially undertaking the tasks/functions in the different phases of the adjudication process discussed in this series in an undisclosed and ad hoc manner, the MIJ framework aims to provide uniform guidelines and specific requirements for how, when and to what extent AI is used during the adjudication process in international arbitration and what the mechanics of an optimal interplay between HI and AI should look like. In the absence of clarity on the nature and scope of AI’s role and its interplay with HI, the procedure for arriving at decisions would become arbitrary, a black box of mistakes and missteps, an unfulfilled promise of efficient and efficacious justice without any way to figure out what contributed to the erroneous outcomes and how. The MIJ framework aims to minimize these errors/mistakes/missteps both by HI and AI to ensure that the final award rendered by the Tribunal is a result of an optimal combination of the abilities and strengths of both HI and AI. This optimal combination governed by the MIJ framework would facilitate consistent and transparent adjudication and serve as the bedrock for efficient and efficacious justice in international arbitration.
* Rana Sajjad, a Columbia Law School alum, is a dual-qualified lawyer licensed as an Advocate of the Supreme Court of Pakistan and a Member of the New York Bar. He has over 25 years’ experience of practicing law in Pakistan and the U.S. in practice areas including contracts, cross-border transactions, commercial litigation and domestic and international arbitration. He is the Managing Partner at Triage Law, a Lahore-based commercial and arbitration law firm, and the Founder & President of the Center for International Investment and Commercial Arbitration (CIICA), Pakistan’s first international arbitration center.