Artificial intelligence can play a supporting role in arbitrator work and, depending on technological and legal developments, could even replace arbiters in decision-making. This article focuses on the supportive role.
Where AI can help
AI-based solutions can assist arbitrators at three levels: case administration and procedural management, data collection and analysis, and the decision-making itself. This piece concentrates on the latter two.
International commercial arbitration often involves processing and analysing vast volumes of filings, evidence and related decisions. Data collection and analysis matter for two reasons:
- first, AI can help process submissions, evidence, relevant court decisions and scholarly commentary more efficiently and accurately;
- second, AI can directly aid drafting of arbitral awards by producing boilerplate text and standardised elements relating to the case’s legal and factual background.
At the level of reaching a specific award, AI can contribute by estimating the likely outcome of a new case based on many prior decisions.
Why these tools are often developed in common-law systems
Many AI tools have been developed in common-law (anglosaxon) jurisdictions because those systems emphasise case law and an inductive method: resolving a case frequently involves analogising to previous cases. Machine learning algorithms trained on large collections of past cases and decisions can therefore reach faster and often more accurate conclusions for individual disputes. By contrast, the deductive approach typical of continental legal systems makes direct transplantation of such case-based AI models more challenging.
The problem of reasoning and justification
Probability-based outputs from algorithms are insufficient by themselves for parties. Both statutory or regulatory requirements and parties’ expectations in dispute resolution demand proper reasoning. At present, AI solutions cannot reliably produce the detailed legal rationale required for an arbitral award — and it is uncertain whether they will ever reach that capability.
Views differ:
- Some argue the issue is technical and solvable: AI will eventually be able to generate adequate reasoning.
- Others stress the indispensable role of human judgment: the arbitrator’s synthesising thinking, professional experience and nuanced argumentation cannot be fully replaced by an AI arbitrator.
Training data quality and bias
At every level, AI conclusions are only as good and as trustworthy as the training data used in supervised learning. The quantity and quality of data processed by these systems are critically important. Training datasets can themselves be biased, meaning an AI may reproduce human prejudices embedded in historical decisions. This is a sensitive matter in dispute resolution where a final, enforceable decision is issued and can only be challenged in limited circumstances.
Ensuring impartiality and independence in the adjudicating arbitrator is therefore crucial, and those involved in curating and training AI systems bear significant responsibility.
Conclusion and next steps
In summary, AI offers promising support for arbitrators—from data analysis to drafting award elements—especially where extensive prior case law is available. However, its current limitations in producing robust legal reasoning and the risks posed by biased training data mean it cannot yet fully replace human arbitrators.
The next article in this series will examine situations in which AI-based tools do not only support but actually replace human decision-making. The earlier instalment of this series is available on the authors’ website.
