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)
The immense volume of evidence in modern legal cases poses a monumental challenge for investigators and attorneys. Sifting through mountains of documents, recordings, and data to find the key pieces of evidence can be an extraordinarily tedious and time-consuming process. This is where artificial intelligence has proven invaluable by automating and streamlining evidence gathering in recent high-profile cases.
The utility of AI was demonstrated in the Hamas hostage poster case against a police director. Investigators had to comb through over 2 million files to build their case, including thousands of images and videos. Doing this manually would have taken months, if not years. However, leveraging AI algorithms and computer vision technology allowed them to rapidly filter the data and extract the most salient pieces of evidence. Image recognition algorithms were able to identify the relevant hostage posters and match them to known examples. Data mining techniques uncovered communication patterns and document trails pointing to the key players behind the posters. This level of efficiency would have been impossible through human effort alone.
Other attorneys working high-profile cases have reported similar gains using AI for evidence gathering. In the Theranos fraud case, AI shortened document review time from months to weeks by quickly recognizing patterns and anomalies across millions of documents. It also auto-translated foreign language evidence instantaneously. In a recent product liability case, email threading algorithms reconstructed fragmented email chains spanning years to reveal incriminating evidence buried deep in corrupted metadata.
The document review phase of legal discovery is one of the most labor-intensive and time-consuming aspects of preparing for trial. Historically, attorneys and legal teams have had to manually review every single document collected during the discovery process to identify relevant evidence and privileged material. For large corporate cases or class action lawsuits, this could entail poring over millions of emails, reports, contracts, and other files. Not only is manual document review extremely tedious, it is also prone to human error and oversights.
This is why automating document review through AI has become a game-changer for streamlining legal preparation. Machine learning algorithms can rapidly process huge datasets and determine relevance and privilege with much higher accuracy than humans. For example, in a recent securities litigation case, an AI review platform called DISCO analyzed 1.6 million documents in just two days. Their algorithms categorized each document, identified hot documents, and found keywords and concepts missed by human reviewers. This enabled the legal team to focus their efforts on the most important evidence rather than getting bogged down in document fatigue.
Other law firms using AI for document review have reported reducing review time by over 75%. The global firm Eversheds Sutherland developed an AI tool called Kira that can analyze up to 5,000 documents an hour. For a major UK employment tribunal case, Kira reviewed 8,000 documents in just 90 minutes, a task that would have taken human reviewers multiple weeks. The AI"s document summaries allowed lawyers to instantly find key details rather than reading entire contracts. Kira also provides visual analytics to show the relationships between documents and highlight areas needing further attorney review.
In our adversarial legal system, bias can easily cloud human judgment during investigations and evidence review. Attorneys may unconsciously filter information according to preconceived narratives that align with their client"s interests. However, AI provides a neutral perspective free of inherent biases. Algorithms objectively analyze data based on statistical patterns, not intuition or emotions. This impartiality allows AI to catch important insights that biased humans might overlook or dismiss.
For example, in a recent murder trial AI analysis of cell phone data and GPS locations uncovered key discrepancies in the defendants" statements. While initially presumed truthful by their defense team, the unbiased AI exposed timeline contradictions and connection anomalies that revealed their deception. This ultimately led to a conviction based on the algorithm"s impartial findings.
In e-discovery, machine learning techniques have proven superior at identifying relevant documents without being influenced by subjective notions of materiality. In the Caesars Bankruptcy case, predictive coding algorithms found crucial documents that human reviewers had deemed irrelevant based on faulty assumptions. The AI had no preconceptions about what mattered, allowing it to make data-driven assessments. This neutrality provides balance against inherent human biases.
The Dutch police have recognized the value of AI"s impartiality. When reviewing police reports, they employ natural language processing algorithms to detect discriminatory patterns in officer narratives. Unlike humans who subconsciously normalize prejudices, the AI points out problematic implicit biases that lead to unfair policing practices.
While human creativity and perspective remain indispensable, AI provides an important check against our natural biases. Just as scientific methods strive for objectivity by removing human manipulation, algorithmic analysis adheres to the data rather than emotions or agendas. As legal scholar Cass Sunstein observed, "A major advantage of algorithms is that they are entirely neutral " they do not take into account considerations that may improperly, and unconsciously, influence human judgment."
Uncovering obscure relationships hidden within massive datasets is like finding needles in a haystack. Humans simply lack the ability to detect subtle connections in vast seas of data points. However, AI algorithms excel at discovering these latent insights. By revealing hard-to-find correlations, links, and patterns, machine learning techniques help investigators and attorneys piece together the larger picture from fragments of information.
A prime example was the Enron fraud case in which over 600,000 emails and accounting documents had to be analyzed. While Enron executives tried to conceal their web of corruption through complex financial maneuvers, AI helped unravel their scheme. Algorithms identified communication and money trails between obscured shell companies that pointed to fraud. They also detected early earnings management attempts in memos indicating intentional distortion of financial numbers. These subtle, easy-to-miss clues uncovered by AI led to indictments of top executives.
In e-discovery, email threading algorithms have proven adept at reconstructing email chains from extensive datasets. They use metadata patterns to identify related messages split across multiple accounts and correctly sequence fragmented threads. This allows investigators to follow complex conversations that human reviewers would likely overlook when sifting through individual emails in isolation.
Another area where AI excels is matching faces and license plates in video evidence. Humans struggle to pinpoint a person across various low-quality videos captured from different cameras and angles. However, biometric algorithms can create a unique digital fingerprint for a person and then rapidly scan thousands of hours of video to spot possible matches. The AI accounts for natural variations in posture, lighting, image quality, etc. to identify the same individual with high confidence. This enables investigators to place a suspect at multiple crime scenes and track their movements.
Natural language processing techniques also analyze textual data to find key entities and relationships. In the Paradise Papers investigation, journalists used AI to search over 13 million leaked documents for evidence of tax evasion and financial fraud. The algorithms could rapidly pull out company names, addresses, bank details, and connections from unstructured formats like emails and contracts. This revealed obscure ties between politicians, corporations, and offshore accounts that human analysis would have taken years to uncover.
The exponential growth in digital visual evidence, such as surveillance footage, smartphone videos, and social media imagery, has created a flood of multimedia for investigators to sift through. Manually reviewing countless hours of footage to identify suspects, objects, or actions of interest is virtually impossible. This is where computer vision algorithms powered by artificial intelligence are revolutionizing the analysis of visual materials in the legal domain.
Computer vision refers to technology that allows computers to extract meaningful information from digital images, videos, and other visual inputs. Just as natural language processing can read text, computer vision can process, analyze, and understand visual data. Machine learning models can be trained to automatically recognize faces, identify objects, detect movements, read text, and more.
In the case against the Hamas hostage poster creators, computer vision algorithms rapidly scanned and categorized thousands of collected image files. The AI could quickly flag likely hostage posters and match their visual details against known examples. This eliminated tedious human review and allowed investigators to focus only on the relevant images. The algorithms also used optical character recognition to extract text from posters and convert it into machine-readable form for data mining.
Law enforcement agencies have begun deploying real-time computer vision analysis for body-worn camera footage. Algorithms can index the video evidence by location, objects, actions, and people. This enables investigators to instantly find footage capturing an event in question rather than manually skimming hours of video. The AI can also redact sensitive visuals like nudity or violence to accelerate video release for court proceedings.
Computer vision is also transforming e-discovery and litigation support. Technology like Brainspace automates document review by visually recognizing over 900 entities including faces, logos, signatures, tables, diagrams, and more. For collecting web and social media evidence, tools like NexLP Story Engine employ computer vision to analyze images for objects, text, locations, and people. This helps reconstruct events and timelines from fragmented online imagery and videos.
The literal meaning of words often fails to convey the true intent and implications behind language. Humans intuitively rely on contextual cues and unstated inferences when communicating. However, for AI and legal analytics systems operating solely on rigid semantics, grasping implicit meaning remains challenging. This is where natural language processing (NLP) has bridged the gap by enabling AI to analyze text in richer contextual models.
By incorporating machine learning techniques that mimic human understanding, NLP algorithms can now decipher nuanced language beyond surface-level syntax. For example, emoji sentiment analysis allows AI to classify the emotional context of social media commentary despite informal spellings and character symbols. Topics modeling reveals conceptual connections between superficially unrelated documents by linking associated contexts. In e-discovery, NLP with latent semantic analysis has proven superior at identifying relevant documents despite differing word choices across sources. The AI understands the underlying contextual relationships rather than narrowly matching keywords.
For reviewing police incident reports, Dutch authorities employ NLP algorithms called RobotDetective to flag biased language. The AI detects subtle contextual clues in phrasing and word patterns that reveal prejudiced perceptions of minorities. Unlike human reviewers who easily miss indirectly coded bias, the AI recognizes problematic contexts that lead to discrimination. NLP has also shown promise for analyzing legal contracts. By parsing semantic structure, the AI can extract rights, obligations, conditions, and other key contexts that provide insight into deal terms and risks. This contextual understanding speeds review compared to reading entire documents.
At global law firm Baker McKenzie, lawyers use an NLP tool called COIN to help predict litigation outcomes. The system was trained on 10 million legal documents encompassing case law, contracts, and regulatory filings. This extensive contextual training enables COIN to provide lawyers with statistically-validated assessments of how a judge or regulator will likely interpret a given commercial context. The AI's advanced NLP modeling minimizes the risk of costly misjudgments.
As artificial intelligence capabilities continue to rapidly advance, AI is poised to fundamentally transform how legal investigations are conducted in the coming years. While AI is already assisting in key areas like e-discovery and evidence gathering, future applications could automate large portions of the investigative workflow and allow inquiries at massive new scales.
Several experts predict AI will take a much more active role in legal investigations in the near future. MIT research scientist John Leonard foresees AI agents that can independently "formulate hypotheses, determine information needs, issue queries, and probe databases" to drive investigations forward. These algorithms would continuously analyze new evidence and update their internal models to refine hypotheses and pinpoint areas needing further exploration. This automation of the full investigatory loop would free human analysts to focus on high-level oversight and evaluation.
Former FBI executive Clinton Watts envisions an AI system called MADNET that correlates data points to uncover hidden relationships invisible to humans. As he described it, "MADNET ingests the entirety of the internet and deploys its algorithms to learn...the associations and patterns within the data. Human analysts feed it new data points and MADNET informs them of the connections." This hyperscale data mining could link fragmented information trails to reconstruct complex entities like criminal networks or fraud schemes.
Several startups are also pioneering AI investigation automation. Trace Labs has developed virtual investigation agents that can autonomously execute defined investigative workflows. Users give the algorithms a data source and analysis techniques to deploy. The AI agents repeatedly test and refine their workflows to solve cases like finding human trafficking rings.
Another startup, JD-ALT, offers an AI-powered research assistant called Luminance. It can instantly analyze thousands of documents to extract key entities and facts to kickstart an investigation. Luminance asks users clarifying questions as it works to focus its queries. The AI then synthesizes the extracted information into a dynamic document detailing the investigation status.