In a recent US case, the defendant expert’s grave misuse of artificial intelligence breached every element of their duty to the court. Experts are advised to play close attention to this case as the principles are likely to apply to those instructed in England and Wales.
The facts of the case
On 24 January 2020, propylene gas leaked from a worn rubber hose at a Houston manufacturer, Watson Grinding. Three people died and hundreds of homes and businesses were damaged.
Residents, families of the deceased and businesses sued Watson Grinding and 3M, who they claimed had not properly serviced that gas-detection system.
3M hired an engineer expert to provide an opinion on whether its work met the standard of care (at a cost of $90,000).
The case outcome
Before trial, the plaintiffs’ solicitors located the citation overlay document as part of the discovery process and asked to see all AI prompts. This resulted in 365 pages of prompts conducted by the expert and model answers generated by ChatGPT.
A jury found Watson Grinding 70% liable and 3M 30% liable. It awarded $61.5 million to 24 residents and business owners.
What did 3M’s expert Asked ChatGPT to do?
The chat record shows a 16-minute, 57-second exchange between the 3M expert and ChatGPT, which resulted in a draft report with 14 sections.
This is a non-exhaustive list of the key prompts that the 3M expert gave ChatGPT:
- He stated that he had been retained a professional expert witness and uploaded his resume and hundreds of case files. Then he gave the AI bot the outcome: “I need to show that 3M met the standard of care and why.”
- That same day he asked ChatGPT to “create an exceptional expert witness report defending the standard of care of 3M”.
- More specifically he asked it to “show how 3M is 0% at fault for the explosion at watson grinding and how my background and experience is well suited to render this professional opinion”.
- Other prompts went to the specifics of the role. For example, uploading a photo of a gas detector of the type at issue in the case and asking: “What am i looking at in terms of gas detection system.”
- He had the model review his drafts, and the assess the quality of his final report. The last score it gave him was 97 out of 100.
- The night before his deposition, he asked ChatGPT: “What am I basing my standard of care on as an expert witness?”
At the time of his deposition, the plaintiff’s counsel stated that 85-90% of the expert’s 30-page report had been generated by the model. The expert did not agree to a number, but then said, “if you say so”.
What can experts learn from this case?
There is a place for AI in expert witness work. For example, in summarising documents and records, finessing writing style or checking basic calculations.
However, this case provides the utmost grave example of an expert using AI to generate a report, which was entirely favourable to his client. If this had been a UK case, it would have been in breach of the fundamental role of an expert witness – to help the court on matters within their expertise, which supersedes any obligation or loyalty to the person who instructed or is paying them.
Experts must be mindful of the fact that there is a risk that their prompts (and thus thought processes) could be examined in court if the judge of opposing counsel spots that AI has been used.
If an expert is unsure whether their proposed use of AI in a case might infringe their duty or undermine their evidence, they would be best placed to speak directly to their instructing solicitors and counsel for guidance.
How Can We Help?
Bond Solon provides training covering all the core skills and knowledge expert witnesses require to fulfil their role compliantly and effectively.
We also offer university certified training programmes which are widely regarded as the industry gold standard by instructing parties, giving holders a recognised badge that enables them to evidence they can fulfil their expert witness duties compliantly and effectively.