The legal profession is constantly changing, often prompted by new technology. Typically lawyers will not adopt new technology until market forces make doing so necessary to effectively represent clients. Under pressure to manage costs and work efficiently, lawyers must continue to adopt smarter and more efficient technological solutions that assist in delivering legal services.
Analytical tools that rely on artificial intelligence-powered software can assist lawyers in a wide range of document-intensive tasks that are critical to negotiating a transaction, conducting an internal investigation, or determining the evidence relevant to the prosecution or defense of a claim. AI uses algorithms to (1) identify and process patterns in data, increasing the accuracy and quality of the identification as more queries are processed (machine-based learning); (2) comprehend and respond to human language patterns (natural language processing); and/or (3) make predictions based on patterns found in sample data (predictive analytics).
In everyday life we are comfortable enjoying the benefits of AI tools: we ask Siri, Alexa, or Google Now to help us find the best nearby diner; we seek service assistance from AI-driven “chatbot” customer service representatives; we use smart home devices to adjust thermostats and lighting; and we trust Spotify, Pandora, and Google Home to select a playlist of songs based on a single request. Yet lawyers are reluctant to integrate AI analytical tools into legal practice, concerned that tools that rely on AI will lead to “artificial” and unreliable results, or may improperly replace a lawyer’s professional judgment for computer-manipulated results.
Intersection of Artificial Intelligence and the Practice of Law
There are AI tools to assist many areas of practice.
1. Contract Review and Analysis
Kira and eBrevia have software that extracts data from contracts to enhance due diligence. Kira can flag particular contract provisions (e.g., cooperation clauses) in a large collection of agreements and generate a “heat map” to show deviations from form. The eBrevia “Contract Analyzer” software converts collections of contracts into searchable text, extracts information based on user specifications, reveals deviations from “playbook,” and classifies documents by language and type.
2. Compliance and Threat Prediction
Several AI applications enable companies to monitor compliance and flag potential threats. For example, NexLP, Inc. offers “Story Engine,” which can review email and other communications in investigations to determine the timeline of key events, identify the participants and the relationships among these key players, and construct the context for communications. RAVN, from iManage, can read large volumes of documents and identify, classify, and extract information quickly and efficiently.
3. Research
ROSS Intelligence has software that permits lawyers to ask a legal research question in natural language, reviews more than a million pages of case law per second, and constructs an answer. ROSS also offers EVA, which analyzes briefs, checks cites, and identifies similar cases.
4. Brief and memo drafting
IBM Watson Debater scans a database (e.g., Wikipedia) for relevant content, identifies the strongest arguments, and constructs arguments on both sides of the issue in natural language.
5. Outcome predictors
Lex Machina applies natural language processing and mines court dockets to predict decision outcomes, favorable jurisdictions in which to bring a lawsuit, and successful motions and arguments before particular judges.
6. E-discovery
Technology assisted review (TAR) provides a means of sorting documents into categories to achieve a more efficient document review. Brainspace reviews documents at rates up to one million per hour and identifies key phrases and clusters in related documents.
The Legal Industry’s Vetting of TAR
TAR’s employment of AI tools is the example most familiar to lawyers, even though lawyers are slow to adopt TAR for even the larger document reviews where it can be most helpful. As with a manual document review, lawyers using TAR first identify the pertinent document custodians and download their documents to a review database to create the “master” collection that will be the source for all future searches.
Rather than relying on a team of temporary attorneys and/or paralegals conducting a manual review of documents that were responsive to search terms, TAR relies on one attorney to guide the review process through interactive testing. The lawyer may use search terms or establish criteria for judgmental sampling to create a seed set. The entire collection of electronically stored information is compared to the seed set. The computer determines a responsiveness score for each document in the ESI collection based on the review algorithm, which tests the similarity of seed set documents to the ESI collection.
The responsiveness metric identifies the success or failure of the TAR. For example, if TAR identifies 100 documents and 90 are responsive, the precision metric is 90 percent; if only 20 are responsive, the precision metric is 20 percent. A low precision metric means the produced documents are highly overinclusive, suggesting the lack of a pattern or a pattern that is not recognizable to the TAR algorithm. Documents that score above the chosen responsiveness threshold are marked as responsive. The lawyer checks the adequacy of the scoring by reviewing a sample of the documents tagged as responsive or unresponsive, and will re-code documents as needed.
This expanded set of attorney-reviewed documents provides the basis for the computer to automatically learn filtering rules, which will be applied to generate a new set of responsiveness scores. This process is repeated until the adequacy is acceptable to the lawyer. The resulting set of documents can be produced, subject to a privilege review.
Alternatively, a firm may wish to do additional quality control prior to production, possibly sampling the responsive and non-responsive documents and running manual searches over those populations to exclude individual documents or document categories that may have been missed in the review process. Computer-assisted review can substantially reduce the number of attorneys and paralegals involved in the review process.
Litigants have challenged the accuracy of TAR algorithms or argued that the seed set was not representative of the entire ESI collection, but several factors typically weigh heavily in favor of the conclusion that TAR satisfies the “reasonable inquiry” standard under Fed. R. Civ. P. Rule 26(g). See Timothy T. Lau and Emery G. Lee III, Federal Judicial Center, Technology-Assisted Review for Discovery Requests: A Pocket Guide for Judges 6 (2017).
- With a sufficiently large volume of documents, “reasonable inquiry” is satisfied if the responding party is enabled to produce a higher volume of responsive material more quickly than otherwise would be possible, especially if the cost of the production is substantially lower because of the need for fewer attorney reviewers.
- The larger the review, the more likely that humans faced with the monotonous task of reviewing thousands of documents electronically will make errors. By comparison, the computer never tires of reviewing thousands of documents, and is consistent in its approach.
- The scientific approach to TAR, which involves iterative quality-checking of search results to establish the responsive document set, compares favorably to a document-by-document manual review, where quick and possibly inconsistent judgments are made by multiple reviewers.
Indeed, courts have consistently lauded the advantages of TAR (or predictive coding) over manual review. See, e.g., Winfield v. City of New York, 15-CV-05236 (LTS)(KHP), <-cite-parallel ref="BL\2017\423037">2017 BL 423037, at *5 (S.D.N.Y. Nov. 27, 2017) (court ordered TAR, which “hasten[ed] the identification, review, and production” of responsive documents and allows parties to “prioritize and/or categorize documents for purposes of document review and has been shown to produce more accurate results than manual review.”); Rabin v. PriceWaterHouseCoopers LLP, Case No. 16-cv-02276-JST, <-cite-parallel ref="BL\22017\277017">22017 BL 277017, at *2 (N.D. Cal. Aug. 8, 2017) (court accepted defendants’ argument that “TAR process is capable of achieving an exceptionally high level of accuracy” and that its use would expedite discovery); In re Actos (Pioglitazone) Products Liability Litig., <-cite-parallel ref="FSupp3d\274\485">274 F. Supp. 3d 485, 499 (W.D. La. July 17, 2017) (court noted that “[d]espite the initial ‘front loaded” investment of time, although not perfect or fully realized, [predictive coding] provided an innovative efficiency to the discovery process when compared to the existing, prevailing methods of review.”); Hyles v. New York City, 10 Civ. 3119 (AT)(AJP), <-cite-parallel ref="BL\2016\248010">2016 BL 248010, at *2 (S.D.N.Y Aug. 1, 2016) (concluded that generally TAR is “cheaper, more efficient and superior to keyword searching”); see also Maura R. Grossman & Gordon Cormack, Technology – Assisted Review in E-Discovery can be more Effective and More Efficient Thank Exhaustive Manual Review, XVII Richmond J. L. & Tech. 1, 37 (2011) (observed that manual reviews identified 25 to 80 percent of responsive documents, and TAR identified 67 to 86 percent of responsive documents).
At the same time, the benefits of TAR should not be exaggerated. As the first court that validated the use of predictive coding cautioned, predictive coding is neither required nor appropriate in all cases. Da Silva Moore v. Publicis Groupe & MSL Grp., <-cite-parallel ref="FRD\287\182">287 F.R.D. 182, 193 (S.D.N.Y. 2012) (magistrate judge’s ruling), adopted by <-cite-parallel ref="BL\2012\101971">2012 BL 101971 (S.D.N.Y. Apr. 26, 2012).
Moreover, neither manual review nor TAR should be expected to achieve perfection. Judge Denise Cote, after permitting a defendant to use predictive coding over the plaintiff’s objection, made the following observation:
The production of documents in litigation … is a herculean undertaking, requiring an army of personnel and the production of an extraordinary volume of documents. Clients pay counsel vast sums of money in the course of this undertaking, both to produce documents and to review documents received from others. Despite the commitment of these resources, no one would or should expect perfection from this process.” Federal Housing Finance Agency v. HSBC North America Holdings, <-cite-parallel ref="BL\2014\40542">2014 BL 40542 (S.D.N.Y. Feb. 14, 2014).
Indeed, TAR cannot displace lawyers in the discovery process because a lawyer’s professional judgment is a necessary part of the TAR review. Attorneys must classify the sample of documents and provide the parameters for the computer’s search. Lawyers who understand the case, the implicated documents, and technology-assisted review must select the most appropriate protocols for the review and may need to defend those choices in court. If a pertinent document custodian is vague or uses coded language in emails discussing a topic of interest, a lawyer will need to correct the review by training the computer on a sample of responsive documents. Finally, the typical algorithm assigns each document a probability of responsiveness. Lawyers may have to hand-classify documents for which the responsiveness probability is too low.
The Ethics of Using AI Software in Legal Practice
The ethics rules support reliance on vendors who supply AI tools to assist in the practice of law provided that the tools are compatible with a lawyer’s professional obligations and assist lawyers to competently represent their clients. In 2012, Comment 3 to
Moreover, Comment 8 to Rule 1.1 provides that “[t]o maintain the requisite knowledge and skill, a lawyer should keep abreast of changes in the law and its practice, including the benefits and risks associated with relevant technology” (emphasis added). The practical significance of this duty is two-fold: lawyers have an affirmative duty (1) to be proficient in the technology they use in the representation of a client (either directly or through those who assist in the representation); and (2) to consider technological advances that may improve the professional services they provide to their clients. The first requires that lawyers who use technology during a client representation (e.g., produce documents using a discovery vendor’s search and storage software) understand the technology well enough to ensure compliance with the lawyer’s ethical duties. Lawyers must also supervise nonlawyers who are involved in e-discovery, and they remain liable for e-discovery vendor errors. See In Re Seroquel Products Liability Litig., <-cite-parallel ref="FRD\244\648">244 F.R.D. 648 (M.D. Fla. Aug. 21, 2007) (a party is responsible for the errors of its vendors, citing Sedona Principle 6: “Ultimate responsibility for ensuring the preservation, collection, processing, and production of electronically stored information rests with the party and its counsel, not with the nonparty consultant or vendor.”). Thus, if a lawyer uses AI in her legal practice, the lawyer must either directly possess the competence to use the technology or hire competent vendors, in which event the lawyer must conduct sufficient due diligence before hiring the vendor and during the vendor relationship.
The second leg of the duty suggests that a lawyer has an ethical responsibility to consider whether she may provide better service using technological tools – including AI software. Currently there may be no instance in which AI software represents the standard of care in an area of legal practice such that its use is necessary to a representation. See, e.g., New York Ethics Op. 1053 (2015) (“if a lawyer needs a sign language interpreter to communicate effectively with a client, then, unless the lawyer utilizes such an interpreter, the lawyer would be unable to provide ‘competent representation’ to the client, as required by Rule 1.1.”). A lawyer has an affirmative duty, however, not only to determine if certain technology is necessary to a client representation, but also to consider whether technological tools will improve service to the client.
Assessing When AI Tools Should be Used in Legal Practice
AI technology and other new technology has already allowed lawyers to provide more efficient services to their clients in several ways. Legal research that previously took days may be accomplished in hours, and large computer networks and files may be reviewed in a fraction of the time. Innovations have reduced, and will continue to reduce, the overall number of lawyers and staff necessary to review and process documents critical to each engagement.
Lawyers can assess whether to use AI tools or other technology by considering the following questions:
- Is the technology recognized in the industry as a means of facilitating the pertinent aspect of practice?
- Is the technology priced competitively as compared to manual or old software approaches?
- Is the technology easily supported by current technology at the firm or will the firm need to invest in outside technology support, and if so, is the outside technological support cost-effective?
- Is there a foreseeable benefit to the client based on the cost, scope, accuracy, efficiency, or some other important factor relevant to the specific project?
- Does the new technology have the potential to create greater efficiencies in the management of the workstreams?
Conclusion
Use of AI tools and other new technology in legal practice is ethically appropriate and eventually may be required in order for a lawyer to discharge her duty of competence and remain abreast of relevant technology. Lawyers remain responsible for determining when AI tools should be used in their practice, and ensure that such usage is appropriately supervised. Clients may begin to require the use of these tools to reduce costs. Use of AI technology will free lawyers from more routine tasks and focus them on issues requiring professional judgment – those areas of practice that remain squarely in the lawyer’s domain.
Learn more about Bloomberg Law or Log In to keep reading:
Learn About Bloomberg Law
AI-powered legal analytics, workflow tools and premium legal & business news.
Already a subscriber?
Log in to keep reading or access research tools.