By Leslie King O’Neal
AI-Generated Evidence is Here
Arbitrators (and judges) depend on the evidence presented to decide disputes. Until recently, most evidence was documentary (paper or photographs) or testamentary (witness testimony, either live or recorded). Some construction lawyers remember digging through bankers’ boxes of project records and tagging them with sticky notes for copying. Those days are over.
Design and construction projects routinely create data using AI tools, which may become part of the evidence in a dispute. One of the first questions counsel should ask at the project team at the beginning of a case is what AI tools were used on the project and where the data is stored.
Design and Construction Are Paperless
Today design and construction projects are mostly paperless. Designers and constructors collaborate virtually, create BIM models, and use project management software to connect project stakeholders and manage projects.[i] Artificial intelligence analyzes data from different technologies (drones, sensors, cameras, IoT devices), to assist project teams in assessing risk, tracking progress, productivity, safety and recognizing potential delays.[ii] Experts use AI tools in developing reports on delays, construction defects, and accidents.
New Types of Evidence Create New Challenges
The combination of electronic data and AI analysis has transformed the evidence used in design and construction cases into electronically stored information (ESI) or even virtual reality. But, these new types of evidence create new challenges for attorneys, arbitrators (and judges).
Evaluating admissibility of AI-generated evidence challenges judges and arbitrators. When a party offers such evidence, it raises numerous questions for the arbitrator (or judge) to consider.
Preliminary Questions for AI-Generated Evidence:
- Is the AI-generated evidence relevant?
- Do factors such as potential prejudice, confusion, or delay outweigh its probative value?
- Have opposing counsel and the arbitrator had an opportunity to review underlying data showing the AI-generated evidence is valid and reliable?
- Is there testimony from qualified lay or expert witnesses to authenticate the evidence?
Sedona Conference Issues Decision Tree for Evaluating AI-Generated Evidence
Fortunately, The Sedona Conference Journal recently published a decision tree, outlining the steps for evaluating AI-generated evidence.[iii] The authors, Hon. Paul W. Grimm (ret.), Maura R. Grossman, and Kevin F. Brady, are recognized experts in the fields of evidence and artificial intelligence and its impact on the law.[iv]
The decision tree leads the reader step by step through the questions for evaluating AI-generated data before it can be admitted into evidence in a federal court. Of course, arbitrators generally aren’t bound to follow the Federal Rules of Evidence (or state evidence rules) regarding admitting evidence. However, the federal rules provide useful guidance in determining whether evidence meets basic admission requirements.
Arbitrators Have Discretion to Admit or Exclude Evidence
Arbitrators often are wary of excluding evidence, since one of the grounds for vacatur under the Federal Arbitration Act is “refusing to hear evidence pertinent and material to the controversy . . . .”[v] But arbitrators have discretion in admitting or excluding evidence. The AAA Construction Arbitration Rules provide arbitrators may reject evidence that is “unreliable.” [vi] Courts have upheld awards where arbitrators excluded evidence, if the hearing was still fundamentally fair to both parties.[vii]
Practice Pointers
In addition to the decision tree, The Sedona Conference Journal article offers practice pointers for lawyers and for arbitrators related to AI-Generated Evidence:
- Identify and disclose early: Identify any AI-generated evidence at the start of the case. Consider whether the opposing party will need discovery or if the arbitrator needs to approve its use.
- Raise the issue immediately with opposing counsel and the arbitrator: Discuss plans to use AI-generated evidence early (at the preliminary hearing). Ask the arbitrator to set deadlines for disclosure, discovery and challenges related to the AI-generated evidence.
- Schedule a hearing to determine validity, relevance and whether the AI-generated evidence is appropriate.
- Explain the AI-generated evidence clearly: Describe the problem the AI was designed to solve and how it works so the arbitrator and opposing counsel can understand it. Show how the AI output ties to the issues in the case.
- Protect sensitive information: Consider the need for protective orders to address trade secrets, proprietary date or privacy concerns related to the AI-generated evidence.
- Use qualified experts: Hire experts who can clearly and simply explain the AI’s development training, testing, and operation. Ensure that expert reports address validity, reliability, error rates and potential biases.
- Demonstrate validity and reliability: Show that the AI has been independently tested for validity and reliability. If no independent testing exits, be prepared to explain why and what other steps were taken to demonstrate validity and reliability.
- Prepare for admissibility challenges: If anticipating a challenged to admissibility of AI-generated evidence, file detailed pretrial motions either in support of or opposing admission, covering validity, reliability, error rates, bias, and provenance (where the evidence came from and chain of custody) so the arbitrator can make an informed ruling.
[i] For more detailed information about and descriptions of new technology being used in design and construction and tips on its discovery and its use as evidence in hearings or trials, See, Technology in Construction Law, Leslie King O’Neal, editor (ABA Press 2023).
https://www.americanbar.org/products/inv/book/433512440/?login
[ii][ii] Daniel Fontana, How Artificial Intelligence is Transforming the Construction Industry: Past, Present and Future (Associated General Contractors of America, April 29, 2025) https://www.agc.org/sites/default/files/users/user50797/Fontana_Daniel.pdf
[iii] P. Grimm, M. Grossman, Kevin Brady, Decision Tree for Evaluating AI-Generated Evidence, 27 The Sedona Conference Journal ____ (forthcoming 2026). https://thesedonaconference.org/publication/Decision_Tree_for_Evaluating_AI-Generated_Evidence
The Sedona Conference (TSC) is a nonpartisan, nonprofit charitable 501(c)(3) research and educational institute dedicated to the advanced study of law and policy in the areas of antitrust law, complex litigation, intellectual property rights, and data security and privacy law. Its mission is to move the law forward in a reasoned and just way.
[iv] See, e.g. Maura Grossman, et all, Navigating AI in the Judiciary: New Guidelines for Judges and Their Chambers, https://thesedonaconference.org/civicrm/mailing/view?id=3064&reset=1;
Paul Grimm, Cary Coglianese, Maura Grossman, AI in the Courts: How Worried Should We Be? 107 Judicature No. 3 (2024)https://judicature.duke.edu/articles/ai-in-the-courts-how-worried-should-we-be/#:~:text=The%20U.S.%20Constitution%20vests%20decision%2Dmaking%20authority%20in,room%20for%20algorithmic%20hallucinations%20in%20judicial%20opinions.
[v] 9 U.S.C. §10(a)(3).
[vi]Rule 35(b), AAA Construction Arbitration and Mediation Rules and Procedures.
[vii] LJL 33rd Street Associates v. Pitcairn Properties, 725 F.3d 184 (2d Cir. 2013).
Leave a Reply