Predictive Coding: Seed Sets and Litigation Privilege

By | January 8, 2020

The use of predictive coding technology in litigation document review has increasingly been welcomed by courts in different jurisdictions. That notwithstanding, there is still legal uncertainty as to whether litigation privilege applies to seed sets. A seed set (also referred to as training set) is a sample of documents selected from a larger document set and used to train the predictive coding algorithm to enable it identify responsive documents in a larger database. Seed set could be developed using random sampling, judgmental sampling or a combination of both (hybrid).

A resolution of the issue as to whether privilege applies to seed set is important as it is relevant to the determination of any obligation by counsel to disclose to the opposing counsel seed sets used in training predictive coding algorithm. In many cases, the courts have interpreted a party’s willingness to disclose its seed set as evidence of ‘cooperation and transparency’ often required by the court before authorising the use of predictive coding in document review. If seed sets are subject to privilege, then compelling or coercing a party to disclose privileged documents to opposing party would not only amount to an affront on the bedrock of civil litigation, but could also impede the acceptability of predictive coding technology in civil litigation.

The purpose and scope of litigation privilege

The primary purpose of litigation privilege also referred to as solicitor-work privilege is to create a ‘zone of privacy’ for counsel to prepare and present their case to the court in civil litigation. It is a vital privilege in modern adversarial system and its objective is to create the professional space needed by counsel to prepare their case without the risk of exposing their litigation strategy to opposing counsel. In the US jurisdiction, this privilege was formally endorsed by the United States Supreme Court in Hickman v. Taylor. The Court in Hickman noted the vital importance of litigation privilege in affording counsel “a certain degree of privacy, free from unnecessary intrusion by opposing parties and their counsel.” The Supreme Court went further to observe the need for counsel to be able to “assemble information, sift what he considers to be the relevant from the irrelevant facts, prepare his legal theories and plan his strategy without undue and needless interference.” Such private space for the lawyer’s preparation of client’s case is vital to the litigation system. Without such space, the court in Hickman noted that:

… much of what is now put down in writing would remain unwritten. An attorney’s thoughts, heretofore inviolate, would not be his own. Inefficiency, unfairness, and sharp practices would inevitably develop in the giving of legal advice and in the preparation of cases for trial. The effect on the legal profession would be demoralizing. And the interests of the clients and the cause of justice would be poorly served.

The Supreme Court of Canada adopted a similar view in Blank v. Canada wherein the court observed that the object of litigation privilege is “to ensure the efficacy of the adversarial process”. To achieve this objective the Supreme Court stated that parties to litigation “must be left to prepare their contending positions in private, without adversarial interference and without fear of premature disclosure.” Prior to the decision in Blank,  Jackett P. of the former Exchequer Court of Canada in Susan Hosiery Ltd. v. MNR, explained the purpose of litigation privilege (once referred to as lawyer’s brief rule) as follows:

Turning to the “lawyer’s brief” rule, the reason for the rule is, obviously, that, under our adversary system of litigation, a lawyer’s preparation of his client’s case must not be inhibited by the possibility that the materials that he prepares can be taken out of his file and presented to the court in a manner other than that contemplated when they were prepared … If lawyers were entitled to dip into each other’s briefs by means of the discovery process, the straightforward preparation of cases for trial would develop into a most unsatisfactory travesty of our present system.

The scope of litigation privilege is very specific. It covers all documents prepared by counsel for the dominant purpose of litigation either during the course of litigation or when litigation was reasonably contemplated or anticipated. The Supreme Court of Canada in Lizotte v. Aviva noted that the privilege “protects against the compulsory disclosure of … documents whose dominant purpose is preparation for litigation.”

Facciola and Favro were of the view that since the decision of the United States Supreme Court in Hickman, and its subsequent codification in Rule 26(b)(3) of the Federal Rules of Civil Procedure, parties to litigation in the United State are precluded from discovery of documents that are prepared by opposing parties in anticipation of litigation. Hence all documents that has the potential to reveal a lawyer’s case strategy, assessment of the strength and weaknesses of the case, as well as its line of proof or defence (as the case may be) is almost absolutely immune from discovery. Facciola and Favro also noted that this protection will apply to counsel’s selection of document where such selection might provide an indication into the lawyer’s mental processes. Thus, the judicial approach to determination of litigation privilege in United States and Canada seems to suggest that litigation privilege could apply to seed sets developed by counsel for use in training predictive coding software in e-discovery.

Application of litigation privilege to seed sets

To date, the use of predictive coding or issues relating to its use in civil litigation has yet to be properly litigated before any Canadian court. Hence, we are left with dearth of jurisprudence in the Canadian jurisdiction to provide any clear guide to judicial view or opinion on the applicability of litigation privilege to seed set. In this situation, it may be proper to turn attention to the US jurisdiction which seems to have a body of guiding case law. Addressing the legal basis for the applicability of litigation privilege to seed set would require a consideration of whether selection of documents from a larger population of documents by counsel for use in the preparation of a case gives rise to privilege. This issue was adequately addressed by the United States Court of Appeals for the Third Circuit in Sporck v. Peil.

The issue in Sporck v. Peil arose from pre-trial discovery. In response to discovery request, the defendant produced hundreds of thousands of documents from which the plaintiff’s lawyers selected about one hundred thousand for copying. Prior to deposition, the defendants’ lawyers selected some documents (from the larger population of documents already produced to the plaintiffs) to prepare the witness for deposition. The documents were said to represent the lawyer’s opinion on relevancy and possible legal defenses. The plaintiffs sought discovery and production of all the documents that were used to prepare the witness for the deposition. The defense counsel refused to identify the specific documents arguing that the documents had already been disclosed as part of the broader discovery process, and that the group of documents used to prepare the witness was subject to attorney work privilege and immune from disclosure. The trial judge, while acknowledging that the documents constituted attorney work product, ordered production of the documents on the basis that “it was not “opinion” work product entitled to absolute protection.”

On Appeal to the United States Court of Appeals for the Third Circuit, the appellate court ruled that the selection and compilation of documents by counsel in this case for pre-trial discovery qualifies as opinion work product and hence immune from discovery. The court based the rationale for its decision on a quote from James Julian, Inc. v. Raytheon Co. [93 FRD 138 (D.Del.1982)] wherein it was stated:

In selecting and ordering a few documents out of thousands counsel could not help but reveal important aspects of his understanding of the case. Indeed, in a case such as this, involving extensive document discovery, the process of selection and distillation is often more critical than pure legal research. There can be no doubt that at least in the first instance the binders were entitled to protection as work product. [emphasis supplied]

Facciola and Favro identified four principles emerging from the decision in Sporck v. Peil. First, assertion of privilege over group of documents selected from a larger population requires a proof that the identification of the selected documents would reveal counsel’s mental impression. Second, the privilege will only apply to identification of the selected documents. It does not protect the documents from disclosure as part of a broader document disclosure process. In this way, the group of documents becomes like a needle in a haystack. While counsel is not obligated to identify the needle, he is obligated to turn it over with the haystack. Thirdly, compilation of documents by counsel may not be privileged where counsel has no reasonable expectation that the mental impression likely to be revealed by the selected documents would remain private. Fourthly, the court may not be inclined to protect counsel’s compilation of documents where the number of document is voluminous. This is based on the assumption that the more voluminous the selected documents, the more difficult it becomes to discern the lawyer’s mental process from a review of the selected documents. This fourth principle though must be approached with caution in an era where the volume of documents involved in electronic discovery (hence in creation of seed set) is becoming bigger in size than ever.

What is evident from the discussion so far is that litigation privilege may apply to a lawyer’s selection of a group of documents from a larger population of documents. This is especially the case where such selection is done in the course of preparation for litigation, and where the lawyer’s mental impression about the case or legal strategy could be gleaned from an examination of the selected documents.

Applying this principle to seed set, it might be reasonable to argue that litigation privilege may apply to seed sets used in training predictive coding algorithm in litigation document review. The applicability though will be dependent on the method or technique used by counsel in creating the seed set. Seed sets generated using random sampling technique are unqualified for litigation privilege because of the absence of any exercise of skill, judgment, and reasoning by counsel in selecting or generating documents. Documents generated through this process is incapable of revealing a lawyer’s litigation strategy, or the lawyer’s opinion on the strength and weaknesses of the case.

However, seed sets generated using judgmental sampling are deserving of litigation privilege because the technique involves extensive use of human intelligence to select a group of documents that meet predetermined criteria set by the lawyer. Thus, development of seed set using this technique requires exercise of independent legal judgment by the lawyer. The documents are meticulously selected by a lawyer based on lawyer’s exercise of skill, judgment, and reasoning. Evident in this method of generating seed set are the basic elements necessary for assertion of litigation privilege. This group of documents should be entitled to litigation privilege for many reasons. First, the documents were selected by the lawyer in the preparation of litigation. Secondly, the documents were meticulously selected by the lawyer and hence potentially reflects the lawyer’s understanding of the case, the lawyer’s opinion and legal theory. Thirdly, disclosure of such documents may compromise or reveal the lawyer’s mental impression of the case and litigation strategy.

With regards to seed set developed using the hybrid technique, the application of litigation privilege will depend on the extent to which the two techniques above are utilized in developing the set. Where the dominant technique utilized is the judgmental technique, it is reasonable to argue that litigation privilege should apply. The contrary would be the case where random sampling is dominant.

Leave a Reply

Your email address will not be published. Required fields are marked *