TOOLS OF THE TRADE – MAY 2026
CROSSING THE LINE IN ADJUDICATION DECISION DRAFTING
Note from the Writer
I make no apology for acknowledging that I used artificial intelligence to assist in the drafting of this article. Indeed, to ignore such tools would be to ignore the realities of modern professional practice. However, that assistance was exactly that – assistance. The structure, reasoning and conclusions remain my own, having been subjected to careful scrutiny, refinement and, where necessary, correction. I have not adopted outputs uncritically, nor delegated judgment. Crucially, I take full responsibility for the content that follows, both in its analysis and its conclusions.
Use of AI in Adjudication
The increasing use of artificial intelligence (AI) in adjudication is no longer speculative – it is now embedded in daily practice. From summarising submissions to structuring draft determinations, AI tools are rapidly becoming indispensable. The recently issued 2025 guidelines by the Association of Arbitrators (Southern Africa) (AoA) recognise this shift and seek to frame it within principled boundaries. They emphasise that AI, properly used, enhances efficiency, but must never compromise the adjudicator’s independence, reasoning, or ultimate responsibility.
AI’s utility is obvious. It can digest voluminous submissions, identify inconsistencies, map arguments, and even propose structured draft decisions. Comparable developments are already visible in institutional arbitration: AI systems can analyse claims, generate case summaries, and even produce draft awards for review. In construction adjudication – where timelines are compressed and documentation extensive – such tools offer undeniable advantages.
Yet herein lies the tension. The AoA guidelines, echoing broader international standards, stress that AI must remain an assistive tool. Core principles – independence, impartiality, confidentiality, and due process – remain non-negotiable. The adjudicator must understand, interrogate, and ultimately own every conclusion reached. AI outputs must be verified, not adopted.
So when does the use of AI cross the line?
The line is crossed when the adjudicator ceases to exercise independent intellectual judgment and instead becomes a passive curator of AI-generated reasoning. This may occur subtly: reliance on AI summaries without revisiting primary evidence; acceptance of AI-structured reasoning without testing alternative interpretations; or, more critically, allowing AI to shape the outcome rather than merely assist in analysing the inputs. At that point, the adjudicator’s decision-making is displaced rather than supported.
A second boundary concerns transparency and accountability. If an adjudicator cannot explain how a conclusion was reached – because it emerged from opaque AI processes – the integrity of the determination is undermined. The guidelines are clear that AI must be used in a manner that preserves explainability and allows for scrutiny.
This leads to the more provocative question: could parties bypass the adjudicator entirely and refer disputes to an AI system?
Technically, the answer is increasingly, “yes”. Platforms already exist that simulate arbitral reasoning and generate draft outcomes. For straightforward, document-only disputes, AI may soon provide a fast and cost-effective alternative. However, such systems still depend on human oversight, and leading institutions explicitly maintain that only a human decision-maker can issue a binding award.
More fundamentally, adjudication is not merely an exercise in pattern recognition or probabilistic reasoning. It involves judgment: weighing credibility, interpreting contractual nuance, managing procedural fairness, and applying equitable considerations. These are not purely computational tasks. They require accountability, something AI cannot bear.
The future, therefore, is unlikely to be one of replacement, but of recalibration. The adjudicator who ignores AI risks inefficiency; the adjudicator who over-relies on it risks irrelevance – or worse, invalidity. The discipline lies in maintaining the adjudicator’s central role as the author of the decision, not its editor.
Crossing the line occurs not from the use of AI – it is about surrendering judgment to it.
Director
Missed a Previous Arbitrarily Speaking Issue?
Catch up here.
