Letting the machine decide: using AI as a decision maker in construction disputes

Introduction

In a landmark move for dispute resolution, the American Arbitration Association – International Centre for Dispute Resolution (AAA-ICDR) announced this month that it is offering an AI arbitrator for two party, document-only construction disputes.

The potential time and cost savings associated with this approach will be compelling for construction businesses, whose projects are driven by tight schedules and profit margins. However, whilst there are obvious immediate benefits, these will need to be weighed against some of the risk factors including risk of bias, lack of transparency and loss of human judgment.

In this article, we consider the impact of using AI as a decision maker in construction disputes, in particular how it might shape other forms of ADR and whether the potential challenges outweigh the advantages.

The AAA-ICDR model

The AAA-ICDR is offering a “human in the loop” model, where the AI arbitrator will evaluate the merits of claims and prepare draft awards, but a human arbitrator will review it to test the reasoning and evaluate the decision, with the aim of safeguarding trust, transparency and due process. At present, it will be limited to construction disputes although, if successful, the aim is to roll it out to other industries in 2026.

The AI arbitrator has been trained on actual arbitrator reasoning from over 1500 AAA-ICDR construction cases and calibrated and trained with human arbitrator input in order to emulate human judgment. Participation is voluntary and fully opt-in.

Obviously it remains to be seen how successful this initiative will be, but with AAA-ICDR estimating that the use of an AI arbitrator will deliver cost savings of 30-50% and time savings of 25-30%, it is likely to be a popular option in the construction sector where time and money are at a premium.

AI as a decision maker in construction arbitrations

Arbitration is a popular form of dispute resolution in the construction industry, with parties often preferring it to litigation because it is a private process, generally delivering more flexible and quicker results, particularly in disputes with clearly defined technical issues, and especially for international projects.

Construction disputes are often technically complex, multi-party, multi-contract and high value, which can lead to costly disputes that are heavily reliant on a large data set. AI is already being used to help analyse this large amount of documentation, support experts, and streamline case preparation – so the idea of using it to review submissions and decide the result of a dispute might seem like the next logical step.

Saving time and cost is an obvious benefit, but an AI decision maker could also mean that there is greater access to justice for smaller businesses who might struggle to afford arbitration.

Letting a dispassionate machine decide may mean the dispute resolution process is less adversarial, and possibly more consistent, reducing tensions and potentially preserving relationships. This is a key factor in an industry where projects may form part of a wider development or offer future opportunities for the parties.

Could an AI decision maker be used in other forms of ADR?

Given that construction disputes typically seem a good fit for the use of an AI arbitrator, there is no reason why it could not be used in other forms of Alternative Dispute Resolution (ADR), which are commonly used in the industry.

It is easy to envisage that highly technical disputes referred to expert determination could be prime candidates for an automated decision maker, where the AI model could learn the specific technical data relevant to the dispute in order to reach a conclusion. However, given that the decision of the expert is usually binding on the parties with fairly limited grounds for appeal (unless otherwise agreed), there are the obvious concerns around inaccuracies and lack of transparency in reaching the decision.

In the UK, statutory adjudication applies to ‘construction contracts’ falling within the remit of Housing Grants, Construction and Regeneration Act 1996 (as amended)). Disputes are resolved in short timescale (28-42 days) and in a relatively cost effective manner (often referred to in the industry as delivering a form of ‘rough justice’) with decisions being binding, unless and until they are finally determined by litigation, arbitration or agreement.

Adjudication is a common and well thought of method of resolving disputes, which was originally introduced to try to resolve cash flow issues and payment disputes in supply chains. However, there have been questions over recent years as to whether it remains fit for purpose for larger and more complicated disputes, which have a significant data set and often require detailed witness and expert evidence.

It could be that the speed and efficiency offered by an AI decision maker could help resolve this issue. Further, as many disputes referred to adjudication have recurring themes – the validity of payment notices, failure to serve notices, variation principles etc. - an automated decision maker could give parties the benefit of more consistent and predictable results.

At present it is not clear whether AI could be an ‘adjudicator’ under the Construction Act or whether a decision would be enforced by the Court. There obviously needs to be further development in the industry and some form of guidance from the Government and adjudicator nominating bodies – but if it proves popular in arbitration, there is no reason to think it could not be utilised in other forms of ADR.

The use of standing dispute adjudication boards in major projects is increasingly popular and it is not difficult to see the benefit that AI would bring if the parties agreed to input project documentation into the AI to allow for a faster analysis of a dispute and potentially a first pass at a decision, to be reviewed by those on the board. This would follow the AAA-ICDR’s safety net for a ‘human in the loop’, but may drastically reduce the cost of the DABs and scope for disputes on the project.

What are the drawbacks?

Whilst the benefits to using AI to help make decisions in construction disputes are clear, there are also a number of concerns and considerations to be taken into account by the parties involved and also in the wider legal community.

• Accuracy and reliability – AI is reliant on the data set it has been trained on. If the data is historical, incomplete or incorrect, this could result in a biased or flawed decision, or one based on hallucinated facts or law. Construction disputes are also often fact dependent and nuanced, requiring interpretation of bespoke clauses, complex legal principles, consideration of the intention of the parties and awareness of site conditions and numerous parties. Will AI be capable of looking at all of these often conflicting factors in context and analysing them as a human would, or will it miss critical context?
• Lack of transparency - unlike traditional software, where specific rules govern the output, AI models learn patterns and make predictions based on the vast amounts of data that have been reviewed, making it difficult to explain or evaluate the conclusion reached. This ‘black box’ reasoning raises concerns where arbitral decisions (and, indeed, other forms of dispute resolution like adjudication) need to demonstrate legal reasoning, evaluation and transparency. Such automated decisions could therefore be more open to challenge and difficult to enforce, particularly on the grounds of natural justice in adjudication.
• Accountability and liability – there are also certain legal and ethical considerations raised by using AI in this capacity, particularly where there are limited grounds to challenge decisions. What if AI makes a decision that is incorrect or negligent – is the arbitral body responsible? Or the developer of the AI model? Will the AI model be permitted to learn from the information it has been given and the decision it has made? Data privacy and confidentiality are key (and often the reason that parties select arbitration over litigation); there may be a greater risk of a data breach where AI is used, leading to a lack of trust in the system and potential for further disputes.

Current AI models are not capable of replicating human judgment and evaluation – maybe they never will. And in an industry where disputes often require the decision maker to analyse credibility, intention and conduct, it is easy to see where AI might fall short. Whilst this might be partly addressed by the AAA-ICDR’s “human in the loop” or hybrid model, it remains to be seen how successful this will be and if/how decisions will be challenged. Hopefully the AAA-ICDR will publish some statistics and data on this down the line.

However, this is also an industry that has embraced the interim resolution of disputes via adjudication, even after the courts enforce decisions where the adjudicator gets the law wrong. If parties to disputes are turning more to adjudication (together with other forms of ADR) due to cost and time saving benefits, it is not difficult to see a situation where they embrace AI decision makers for the same reasons.

Conclusion

Fundamentally, an AI decision maker will offer huge benefits to the resolution of disputes in the construction industry, particularly offering speed and efficiency which in turn will cut costs and give greater access to the justice system.

However, does quick or brief necessarily mean better? Will some of the cost and time savings potentially be eroded by an increase in challenges and further disputes down the line regarding how the decision was reached?

Notwithstanding this, the adoption of AI as a decision maker does seem like the next logical step in dispute resolution because of its obvious benefits, provided that rigorous governance is put in place to ensure that there sufficient protections and transparency, and that the nuances specific to construction are capable of being addressed.

It is also remains to be seen how reliable the AI's outputs will be in practice, and whether there will be significant adoption of the tool in the industry – time will tell.