A survey of 557 arbitration professionals in the United States, conducted jointly by the American Arbitration Association (AAA) and Jus Mundi, has produced sharply conflicting findings on the use of artificial intelligence in arbitration. The results show a profession that embraces AI for early-stage, low-stakes work but firmly rejects it for the tasks that define the arbitrator's role.

What arbitrators will use AI for — and what they won't

The survey data reveals a clear hierarchy of trust. Document review and organisation leads the list of current or potential AI use cases, followed by legal research, and then case management and timeline tracking. But even the most popular of these three use cases does not exceed 51% adoption.

The resistance hardens sharply when AI approaches the core adjudicative function. A majority of 58% said they would not primarily use AI to help with award drafting — the final 'judgment' in an arbitration. Another 49% would not primarily use AI in an evidentiary hearing. Only 14% of respondents said they would primarily use AI for drafting awards.

One figure stands out for its counterintuitive nature: just 11% said they would use AI for translation. This comes despite a market that now includes companies focused exclusively on legal translation, alongside several others offering high-quality general translation. The likely explanation, the survey suggests, is risk aversion in a context where every word could matter.

The time-savings paradox

On the question of benefits, the results split almost down the middle. While 43% of respondents report significant time savings from AI use, 37% see no major benefits at all.

Several explanations are possible. Practitioners who distrust AI outputs — perhaps following a poor early experience — may simply not use the technology enough to generate savings. Others may use AI sparingly and see modest gains that disappear within the broader scope of their manual work. A third possibility, raised by commentators on the data, is that some arbitrators fear being replaced by AI and wish to downplay its utility.

Why arbitration resists automation

The overall finding is that arbitrators remain conservative in their deployment of AI, trusting the technology primarily at the early stages of a matter — review and research — and withholding it from the critical stages. The logic is structural. The final award document is meant to represent the personal judgment of the professional arbitrators. They actively want that document to reflect their own reasoning, not an AI-generated rewrite of the key arguments in the dispute.

Arbitration, of all legal work streams, is one rooted in human judgment: parties ask respected experts for their personal view in order to resolve often complex and previously intractable disputes. Yet the process still involves substantial 'paperwork' and 'donkey work', particularly at the early stages — which may explain the split character of the survey results.

What the sponsors said

Commenting on the report, Linda Beyea, AAA group vice president, said: 'The more legal professionals work with AI tools, the better they understand their capabilities, where they can deliver value, and what is needed to use them effectively and responsibly. These findings reinforce what we have seen in our AI governance research: organizations that use AI extensively grow more confident in their ability to use the technology responsibly.'

Annie Lespérance, Head of Americas at Jus Mundi, added: 'This research gives the arbitration field something it hasn't had before: an empirical starting point. The data doesn't tell a simple story of adoption or resistance; it reveals a community thinking carefully about where AI belongs in practice and asking for the guidance and tools to move forward responsibly.'

Practical consequence for practitioners

For arbitration counsel and institutions, the message is one of calibrated adoption. AI tools directed at document review, legal research and case management have a receptive audience; tools marketed for award drafting or hearing support face near-majority or outright majority resistance. Providers and law firms building AI workflows for disputes should concentrate on the front end of the arbitration lifecycle, where usage already approaches half of the professional population, and treat the adjudicative core as a boundary that the market itself has drawn. The near-even split on perceived benefits also suggests that demonstrable, measurable efficiency gains — rather than general capability claims — will drive the next phase of adoption.