Musing about Discovery and Synthetic Intelligence

A few of us are collaborating in beta testing of generative synthetic intelligence (“AI”) for authorized functions within the regulation agency setting.  To date the decision is – associates can breathe straightforward, at the least for now.  Nothing we’ve seen is able to replicating authorized analysis even at a first-year stage of high quality.

However that doesn’t imply that AI gained’t influence prescription medical product legal responsibility litigation.  Particularly, we’re not stunned to be taught that AI is getting used within the context of FDA-required antagonistic occasion reporting, purported issues with which have grow to be one of many different facet’s go-to preemption dodges.  Only a few examples from a easy Google search:

Antagonistic occasion instances endure scientific evaluation.  Case analysis contains assessing the opportunity of a causal relationship between the drug and antagonistic occasion, in addition to assessing the result of the case.  An AI mannequin was developed primarily based on related options utilized in causality assessments; it was educated, validated, and examined to categorise instances by the likelihood of a causal relationship between the drug and antagonistic occasion.  AI/ML has additionally been utilized to find out seriousness of the result of ICSRs [Individual Case Safety Reports], which not solely helps case analysis, but additionally the timeliness of particular person case submissions that require expedited reporting.

FDA, “Utilizing Synthetic Intelligence & Machine Studying within the Growth of Drug & Organic Merchandise,” at 10 (2022).  “We conclude that AI can usefully be utilized to some features of ICSR processing and analysis, however the efficiency of present AI algorithms requires a ‘human-in-the-loop’ to make sure good high quality.”  Ball & Dal Pan, “Synthetic Intelligence” for Pharmacovigilance: Prepared for Prime Time?,” 45 Drug Security 429, at summary (2022).

Early detection of ADRs and drug-induced toxicity is a necessary indicator of a drug’s viability and security profile.  The introduction of synthetic intelligence (AI) and machine studying (ML) approaches has resulted in a paradigm shift within the subject of early ADR and toxicity detection.  The appliance of those fashionable computational strategies permits for the speedy, thorough, and exact prediction of possible ADRs and toxicity.

Yang & Kar, “Utility of Synthetic Intelligence & Machine Studying in Early Detection of Antagonistic Drug Reactions (ADRs) & Drug-Induced Toxicity,” 1 Synthetic Intelligence Chemistry, at summary (2023)

What this tells us, as litigators in MDLs and different mass torts, is that plaintiffs’ efforts at taking “discovery” of AI algorithms employed in FDA-mandated antagonistic occasion reporting gained’t be far behind.  Significantly with AI, nonetheless, there’s a superb line between what has already been created and what AI can create going ahead. The bottom line is to restrict such discovery to what “discovery” is meant to be, as outlined by Fed. R. Civ. P. 34.  Within the case of digital data, Rule 34(a)(1) permits a requesting celebration “to examine, copy, take a look at, or pattern . . . electronically saved data” (emphasis added).  Thus, requestors are restricted to discovering “knowledge . . . saved in any medium.”  Id.

The 2006 Advisory Committee notes specify that “Rule 34 applies to data that’s mounted in a tangible type and to data that’s saved in a medium from which it may be retrieved and examined.”  Different key language within the feedback is:

The addition of testing and sampling to Rule 34(a) with regard to paperwork and electronically saved data shouldn’t be meant to create a routine proper of direct entry to a celebration’s digital data system, though such entry is likely to be justified in some circumstances.

(Emphasis added).

We emphasize these factors as a result of what we don’t need to occur is for the opposite facet to transcend entry to “saved” data allowed beneath Rule 34, and as a substitute attempt to manipulate AI applications to create new outputs that – the opposite facet will contend – display hypothetical inaccuracies or shortcomings that will by no means have occurred within the real-world operation of such AI.

The authorized proposition is solely this:  “Plaintiff could not require Defendants to create proof that doesn’t presently exist.”  Brown v. Clark, 2013 WL 1087499, at *5 (E.D. Cal. March 14, 2013).  “Defendants haven’t any obligation beneath the invention guidelines to create proof to assist Plaintiff’s claims.”  Warner v. Cate, 2016 WL 7210111, at *9 (E.D. Cal. Dec. 12, 2016).

Whereas Plaintiff is entitled to hunt related proof from the Defendants in discovery and to file a movement to compel if essential, Plaintiff could solely search proof that already exists.  The principles of discovery don’t enable Plaintiff to compel Defendants to conduct an investigation to create proof for Plaintiff.

Rider v. Yates, 2010 WL 503061, at *1 (E.D. Cal. Feb. 5, 2010).  Events “aren’t required to create proof that doesn’t presently exist with a view to adjust to their discovery obligations.”  Bratton v. Shinette, 2018 WL 4929736, at *5 (E.D. Cal. Oct. 11, 2018).  “If no such [evidence] exists, as [the producer] purports, [requestors] can’t depend on Rule 34 to require [them] to create a doc assembly their request.”  Abouelenein v. Kansas Metropolis Kansas Group School, 2020 WL 1124396, at *4 (D. Kan. March 6, 2020).  A “[p]laintiff shouldn’t be entitled to play-by-plays of ever-changing knowledge.”  Moriarty v. American Basic Life Insurance coverage Co., 2021 WL 6197289, at *4 (S.D. Cal. Dec. 31, 2021).

That’s what permitting plaintiffs to govern a defendant’s AI reporting system quantities to.  They’d be going past merely accessing “saved” data and as a substitute could be demanding to make one thing new – corresponding to a intentionally incomplete antagonistic occasion report – that didn’t exist when such “discovery” was sought.  We have to anticipate plaintiffs trying this interference with our shopper’s AI methods, with antagonistic occasion reporting representing a very probably early stress level.