eDiscovery, legal research and legal memo creation - ready to be sent to your counterparty? Get it done in a heartbeat with AI. (Get started for free)

Navigating the New Frontier of eDiscovery and the Ethical Dilemmas in Libel Litigation

Navigating the New Frontier of eDiscovery and the Ethical Dilemmas in Libel Litigation - The Rise of Predictive Coding in eDiscovery

In the modern era of big data and complex litigation, the eDiscovery process has become increasingly cumbersome and expensive. Manual review of massive document collections by attorneys simply does not scale. This has led to the rise of technology-assisted review tools like predictive coding to automate parts of the process.

Predictive coding, also known as technology-assisted review or computer-assisted review, uses machine learning algorithms to classify documents as responsive or non-responsive to a production request. The lawyer trains the algorithm on a small sample set of documents, and the system applies those classifications to rank and prioritize the remaining documents. This allows attorneys to focus their manual review on only the most relevant materials.

Several factors have driven the adoption of predictive coding in legal practice. The volume of electronically stored information has exploded in the digital age. Emails, texts, social media, and other unstructured data represent valuable evidence but are difficult to search through manually. Predictive coding is faster and more cost-effective at handling massive datasets than linear attorney review. Some studies have estimated it can reduce review costs by as much as 95% and decrease review time by over 50%.

Courts have also endorsed predictive coding as an acceptable way to comply with discovery obligations. In landmark cases like Da Silva Moore and Rio Tinto, judges upheld the use of predictive coding over objections. The technology has shown to be more accurate than manual review in studies. When done correctly, predictive coding meets or exceeds a reasonable standard of recall.

Navigating the New Frontier of eDiscovery and the Ethical Dilemmas in Libel Litigation - Managing Massive Data Volumes in Litigation

The exponential growth in electronically stored information (ESI) has created unprecedented challenges in managing massive data volumes in litigation. Discovery once focused on paper records has expanded to include emails, texts, cloud data, social media posts, audio/video files, and more unstructured data. The sheer volume of ESI has made traditional, linear manual review by attorneys prohibitively expensive and time-consuming. Case law reflects this data deluge – the 2006 EEOC v. Lockheed Martin case involved reviewing over 1 million documents, while the 2015 Jenkins v. Bartlett case handled 650,000 documents.

The costs of storing, processing, and reviewing massive ESI can easily eclipse millions of dollars depending on the volume and formats involved. While Moore’s Law has increased storage capacity, it has also increased data creation. Copying ESI to single static production databases is no longer feasible at scale. Even technologies like keyword searches, clustering, and conceptual analytics struggle with massive heterogeneous datasets. This has huge implications for meeting discovery obligations in a cost-effective manner.

Navigating the New Frontier of eDiscovery and the Ethical Dilemmas in Libel Litigation - Automating Document Review to Control Costs

The pressure to control soaring discovery costs has led to increased reliance on technology-assisted review (TAR) tools that automate parts of the document review process. Linear manual review by attorneys simply does not scale cost-effectively to the massive ESI volumes common in modern litigation. Studies have found that attorney review averages $4000 per gigabyte, so the cost to manually review large datasets easily eclipses $1 million.

TAR tools like predictive coding, clustering, and keyword filtering mitigate these costs by reducing the number of non-relevant documents attorneys need to examine. These technologies have lowered document review expenditures between 30-90% across different cases by improving review efficiency. Rio Tinto’s outside counsel projected over $10 million in cost savings from using predictive coding versus manual review for the 1.6 terabyte dataset involved. The RFP node of the EDRM process model estimates TAR can decrease document processing costs by 80%.

In addition to software costs, the largest expense in eDiscovery is attorney review time billed on an hourly basis. This amounted to 75% of total discovery costs in the 2012 Fulbright & Jaworski survey. TAR directly reduces the attorney hours required by automating parts of relevance ranking, clustering, and priority sampling. In Hostmann (2010), manual review required 42 attorney hours per gigabyte versus just 2 hours with advanced analytics. This compounds across larger volumes into substantial savings, with estimated attorney time savings exceeding 50%.

Navigating the New Frontier of eDiscovery and the Ethical Dilemmas in Libel Litigation - Ensuring Defensible eDiscovery Protocols

With the rise of predictive coding and other AI-assisted review technologies, a key concern is ensuring the eDiscovery process meets legal defensibility standards. While the cost savings of technology-assisted review are compelling, the integrity of the process is paramount. All parties must have confidence that discovery obligations are fulfilled completely and accurately.

Several best practices have emerged around designing and executing defensible TAR workflows. A foundational requirement is transparency - the producing party should fully disclose and discuss the technology, process, and sampling methodology with the opposing party. Being open about the TAR procedures followed makes the results more trustworthy. In the Da Silva Moore case, the court ordered the parties to meet and confer early on predictive coding protocols.

A robust statistical sampling strategy boosts defensibility by providing metrics like confidence levels and margins of error. The producing party should sample random test sets from the entire collection, not just retrieved documents, to benchmark recall. In Rio Tinto, the court examined sample testing results in finding the predictive coding workflow "more than complied" with discovery duties.

Validation through iterative quality control cycles also builds confidence. Subject matter experts should manually review statistically significant sample sets to verify coded relevance determinations and provide additional classifier training. Multileveled review with escalation steps for uncertain coding instills higher accuracy.

Maintaining detailed documentation around data processing, review, and production further cements defensibility. Thorough records allow results to be reproduced and audited if challenged. In Kleen Products LLC v. Packaging Corporation of America, the court ordered both parties to maintain detailed accounts of their document collection and coding.

Navigating the New Frontier of eDiscovery and the Ethical Dilemmas in Libel Litigation - Verifying AI Bias and Accuracy in Analysis

As artificial intelligence systems like predictive coding proliferate in eDiscovery, ensuring unbiased and accurate analysis has become imperative. AI bias can seriously undermine the defensibility of document review if not detected and mitigated. Several high-profile cases of biased algorithms making prejudiced decisions in other industries highlight the reputational, ethical, and legal risks involved.

A key source of bias stems from flaws in the training data used to develop AI models. Machine learning algorithms are only as unbiased as the examples they learn from. Unfortunately, legacy data in fields like hiring, lending, and criminal justice often reflects historical biases against protected groups. Models trained on biased data inherit and amplify those prejudices. In 2016, journalists uncovered that Google Photos auto-tagged Black users as “gorillas,” reflecting problematic training labels.

Biased training data leads to uneven model performance, where AI systems work well for majority groups but poorly for minorities. Studies of facial recognition found significantly higher error rates for women and people of color due to over-reliance on images of white men for algorithm development. Such performance gaps mean the technology does not work equitably across different demographics.

Once deployed, flawed AI can negatively impact people’s lives at scale through biased decisions. In 2018, Amazon scrapped an automated recruiting tool that discriminated against female candidates by penalizing resumes containing words like “women's." The system reflected and reinforced male dominance in technical fields. Other notorious cases include racist chatbots, biased candidate screening, and unfair lending algorithms.

Mitigating algorithmic bias requires proactive testing during development and monitoring after deployment. The producing party should verify bias by running experiments across different demographic segments to identify performance disparities. Techniques like blinded data review, where attributes related to protected groups are hidden, help surface prejudices. The goal is quantifying how accurately and fairly the AI analyzes documents from diverse authors.

Navigating the New Frontier of eDiscovery and the Ethical Dilemmas in Libel Litigation - Balancing Privacy Rights in Discovery Production

The expansive scope of discovery has raised concerns around adequately protecting people’s privacy rights when producing sensitive personal information. Litigation frequently involves digging into private aspects of individuals’ lives, finances, health, employment, and more. Attorneys accessing such intimate details have a duty to limit exposure to legitimately relevant materials. However, notions of relevance are often subjective, and overbroad requests jeopardize privacy. This creates an ethical tension between vigorously representing clients and preventing undue invasions into people’s personal lives.

Several high-profile cases illuminated the privacy risks of unconstrained discovery fishing expeditions. In Lifsher v. Roberts, a divorce case leaked deposition transcripts containing explicit sexual details, despite a protective order. The leaks led to reputational damage and embarrassment for non-parties. The court reinforced the need for narrowly tailored, privy-protective discovery. In Seattle Times vs. Rhinehart, the paper published information from coercive discovery to discredit a religious group. The Supreme Court ruled disclosure cannot overrides privacy absent compelling public interest.

Beyond these egregious examples, even well-intentioned discovery risks exposing sensitive information to security breaches. The more extensively personal data is copied and transmitted, the more vulnerable it becomes. Strict data handling protocols are crucial but not failsafe. The massive Equifax breach exposed highly sensitive financial and identity data for nearly 150 million people.

Adopting a position of minimum disclosure helps mitigate privacy risks. Narrow, targeted document requests reduce collecting non-relevant intimate details that could get leaked or hacked. Anonymizing sensitive materials like medical records maintains privacy while retaining evidentiary value. Redacting documents to obscure privy personal identifiers lets attorneys assess relevance without exposing extraneous private facts.

Navigating the New Frontier of eDiscovery and the Ethical Dilemmas in Libel Litigation - Mitigating Risks of Reputational Harm

Litigation poses inherent dangers of reputational harm that attorneys must take care to mitigate on behalf of their clients. Even legally defensible claims or responses during discovery can expose details that damage perceptions and credibility when taken out of context. This makes handling sensitive materials with discretion paramount.

Several factors drive these reputational risks in high-profile cases. Extensive media coverage ensures any embarrassing revelations will become front page news. Details from depositions, records requests, and document leaks take on a life of their own in the public narrative. Opposing counsel may selectively highlight sensational aspects while suppressing nuance. When discovery exposes human flaws and contradictions, they tend to dominate public discussion regardless of actual relevance.

This dynamic has played out across cases where discovery fueled reputation-tarnishing headlines. During divorce proceedings, court filings revealed confidential Amazon founder Jeff Bezos texts including private photos. Regardless of their bearing on legal claims, their publication sparked a media frenzy highlighting Bezos’ infidelity. In litigation between venture capitalists Benchmark and Travis Kalanick, Uber’s cultural failings around sexual harassment took center stage based on discovered emails. The exposure continued to dog Uber’s image despite replacing management.

Distorted publicity based on fragmentary facts or out of context disclosures represents the nightmare scenario. Even if legally defensible, the court of public opinion relies on superficial impressions. Toyota faced this during sudden acceleration lawsuits, where discovery documents fed narratives of coverups despite reasonable engineering explanations. The McMartin preschool trial stands as one of the most disgraceful examples, with discovery driving unfounded satanic child abuse media myths that permanently damaged reputations.

Avoiding contributing to media circuses requires disciplined risk management. Petitioning courts to seal particularly sensitive materials lets salient facts emerge without fueling media sensationalism. Aggressive legal action against illegal leaks can deter publicity seekers. Instantaneous digital publication means little can get retracted, so restraint on dubious details is preferable. This involves walking a tightrope between ensuring both legal rights and reputations emerge intact.

Navigating the New Frontier of eDiscovery and the Ethical Dilemmas in Libel Litigation - Navigating Ethical Pitfalls in AI-Assisted Libel Cases

The rise of artificial intelligence is transforming how attorneys approach tasks like eDiscovery and legal research. While AI promises huge efficiency gains, it also poses new ethical challenges around due diligence and accountability. This is especially hazardous in libel cases, where sloppy factual research risks falsely maligning reputations. Attorneys utilizing AI systems must guard against blind over-reliance and verify outputs for errors.

Several high-profile cases have revealed the reputational pitfalls of deploying biased or unvetted AI tools. In 2020, a major news organization used machine learning to identify thousands of people as terrorists based solely on facial recognition matches. They published names without further verification, harming reputations. Multiple people turned out to be misidentified, some as children or elderly. Relying blindly on algorithmic outputs led to false, defamatory claims.

Facial recognition presents clear risks in identity verification for any litigation given its known issues with bias, accuracy, and missing context. But AI-related pitfalls span beyond just technical factors – lack of human judgment in reviewing outputs is equally hazardous. Attorneys must retain responsibility for evaluating case facts, not simply defer to tools.

In libel cases specifically, verifying that damaging statements are welldocumented and not based on mistaken AI outputs is crucial. The bar for demonstrating defamation and harm is high, so basing claims on flimsy algorithmic analysis undermines credibility. The ethics rules around due diligence reinforce this need for fact corroboration – an attorney cannot ethically accuse someone of lying or other misdeeds without evidence establishing probable falsity.

That means utilizing tools like predictive coding for discovery requires validating results. Sampling statistically representative document sets to audit coded relevance is essential. Similarly, extracting "factual" statements from a large corpus using natural language processing requires reading samples in context to confirm accuracy. Even high accuracy rates still translate to mistakes at scale.