chevron-down Created with Sketch Beta.

Probate & Property

May/Jun 2023

Technology—Property: The Robots Are Coming: The Threat and the Potential of Artificial Intelligence

Seth Rowland


  • ChatGPT and similar applications of artificial intelligence (AI) in law.
  • With AI, the computer is “fed” a few dozen, hundreds, or even thousands of documents.
  • The practice of law is ever-evolving and the need to embrace the bots while understanding their limitations.
Technology—Property: The Robots Are Coming: The Threat and the Potential of Artificial Intelligence
Richard Newstead via Getty Images

Jump to:

Technology—Property provides information on current technology and microcomputer software of interest in the real property area. The editors of Probate & Property welcome information and suggestions from readers.

Since ChatGPT was released to public beta at the end of 2022, lawyers have been asking, “What will happen next?” You can tell ChatGPT to compose a legal brief, but therein lies the rub. How good is that legal brief, and is ChatGPT threatening my livelihood as an attorney? With ChatGPT, you can ask a detailed question and get an answer in written form. With a Google search, you get a search result with an “extracted text” response from the top search result, followed by a list of web pages ranked by relevance. By contrast, with ChatGPT, you get a comprehensive written answer but no citations and no links to web pages.

There have been disturbing reports that ChatGPT makes up stuff; that is bad news for lawyers whose stock in trade is credibility. Hallucination is the term used when ChatGPT makes an entirely unsupported statement of fact. According to an article published in the online magazine, Datanami, “Roughly speaking, the hallucination rate for ChatGPT is 15% to 20%.” The article continues: “So 80% of the time, [ChatGPT] does well, and 20% of the time, it makes up stuff.” See Alex Woodie, “Hallucinations, Plagiarism, and ChatGPT,”, January 17, 2023.

ChatGPT doesn’t reveal its sources. A first-year law student knows that a legal brief without accurate and verified citations isn’t worth the paper on which it is printed. Moreover, Chat does not “know” anything after 2021, when it stopped learning. ChatGPT is only a glorified parlor trick that can provide helpful but unreliable information. So, is there a place now for ChatGPT and similar applications of artificial intelligence (AI) in law? Incidentally, there are reports that when ChatGPT is incorporated into the Bing search engine later this year, it will include footnotes linking to web pages that support the statements made in the “answer.” And indeed, the Bing implementation will include a crawler that keeps learning and is aware of current development.

The Promise of Artificial Intelligence

As a legal technology consultant specializing in automated document assembly and workflow, I have long dabbled in automated intelligence. I review dozens of transactional documents to identify transaction changes in a typical consulting engagement. I translate those changes into simple English questions and decision points that mirror how a lawyer thinks about a transaction. Then, I organize those questions into an interview that can be presented to a junior attorney, a paralegal, or a client. Using business logic, I mark up document templates and load them into an online platform, be it XpressDox, ContractExpress, HotDocs, or PatternBuilder, which presents the interview to the end user. Then, as if by magic, complex, highly accurate, and verified documents are produced in minutes.

AI is different. With AI, the computer is “fed” a few dozen, hundreds, or even thousands of documents. It reviews them, searching for linguistic patterns. AI needs to be trained, but unlike a dog, most AI engines already understand basic English. AI engines don’t yet understand English’s legal meaning or the law’s nuances. That is where the trainer builds a linguistic model so the AI engine can understand what it is reading and how the trainer wants it to respond.

AI engines are currently very good at identifying patterns and extracting data that is semi-structured. For example, an AI engine can read a lease and identify the landlord and the tenant. It does this by looking for the word “Landlord” in quotes and then stepping backward through the document to find the closest “entity,” which it determines to be the landlord’s name. An entity to AI is the name of a person or a business entity. The AI engine can then move forward through the document, noting other properties of that entity, such as the type of organization, where it is incorporated, the business address, and the authorized signatory. One can also train the AI engine to spot dates and numbers and identify what they are for from the context of those dates and numbers. A human might take five minutes to abstract a single lease; an AI engine can abstract 100 leases in five minutes. Initially, the human will more accurately identify the key provisions and who the parties are. Over time, with corrections by humans, the accuracy level of the AI will improve to the point that the human overseer is needed less frequently.

When we, in turn, ask the AI engine to construct a new lease based on those 50 models that meet our particular needs, the best current AI engine fails to deliver. It is constrained by what it reads and cannot extrapolate what you need from the source material without the risk of hallucinations (see above). Further, statistical biases from the source documents limit its creativity to inventing new linguistic constructs. A true understanding of why a provision was chosen differs from statistical probabilities that a particular legal provision should be used. AI engines are an aid for good legal drafting and not a replacement. The AI Lawyer HAL 9000 is a parlor trick, not a real threat.

AI on Display at ABA TECHSHOW 2023

In March 2023, I attended the ABA’s annual Tech Show conference in Chicago. At the meeting were several startup companies using AI technology to power their service offerings and some existing vendors who have added AI to enhance their software. In this article, I will look at some current approaches to using AI to change how lawyers deliver their services.

What I saw fell into five categories: (1) the AI-powered contract abstracter, (2) the AI-empowered “virtual receptionist” and intake clerk, (3) the AI-enhanced litigation support paralegal team, (4) the AI-turbocharged research engine, and (5) the AI-adjacent contract reviewer and drafting assistant. AI technology, while disruptive, has not reached the level of HAL 9000, the “Heuristically programmed ALgorithmic computer” that powers a space station in Stanley Kubrick’s movie 2001: A Space Odyssey. AI engines are quite a long way from becoming Skynet, the self-aware computer system in the Terminator movies series, a network of computers built by Cyberdyne Systems that deemed humans a threat to its existence and sought to terminate the human race.

The AI-Powered Contract Abstracter

The obvious first use for AI is reading and extracting information from legal and business documents. AI technology is now part of your Office365 subscription. For a modest extra fee, you can write an application that reads vendor invoices attached to incoming emails and automatically inputs your bills into a spreadsheet or your accounting system. With PowerAutomate tools, you can train the AI engine by uploading a few dozen invoices and marking on them where critical data is located: the invoice number, the name and address of the vendor, the amount owed, etc. YouTube videos and free all-day seminars teach you how to do this. See

If you are a corporate counsel and need to review and manage hundreds of agreements, more powerful AI tools are optimized for reading and understanding legal documents. Someone else has taken the trouble to educate the AI engine to read and understand an agreement at the level of a paralegal or corporate contract manager. FoundationAI ( showed me a system that could pull several hundred similar agreements from a DMS like NetDocuments or iManage and spit out a report of parties, terms, and key provisions in short order. As Will Parkhurst, VP of Sales at FoundationAI, told me, contract managers whose sole job is reviewing contracts and producing spreadsheet reports don’t know they are already out of a job. With a proper model, the task of managing contractual relations just got a whole lot easier.

The data extraction can also be applied to something as unstructured as a user’s email inbox. With FoundationAI’s Extract Filer, users can train the system to tag, categorize, and file the emails and attachments that overwhelm a user’s inbox. It can even split large files into separately categorized PDFs and link the documents automatically to the correct matter. Along the way, it can capture essential information about each document.

Filevine ( took another tack with AI for data extraction. They have an immigration law module that fills out US Immigration and Naturalization forms. Much of the information required for these forms is written in a foreign language. Filevine added a module that takes images and documents provided by clients—such as birth certificates, marriage licenses, passports, and tax returns—first translates them, if they are not in English, and then extracts and labels the data. Once extracted and labeled, the client data (name, date of birth, etc.) is then stored in Filevine and can be used to auto-populate immigration forms. ( uses tools for entity extraction to identify confidential and personal protected information in documents. This information is restricted and must be redacted before disclosure to other parties, or the law firm could be liable. The redaction process is tedious and error-prone. will safeguard this sensitive data by auto-redacting entity information in documents. It looks for entity information and wipes it out for document production. Integrating with a document management system (DMS) like NetDocuments or iManage creates a new protected document version while preserving the original.

Another exhibitor, Wisedocs (, has focused on understanding medical records. These documents often appear to be written in a different language. Deciphering medical bills, costs, and adjustments can be challenging. Wisedocs trained its AI to understand medical records about legal claims. It can process hundreds of pages an hour. It can shorten turnaround in document review and de-duplicate the documents.

For personal injury attorneys, another venture, Truve (, consolidates case information into a dashboard. Truve has a range of tools to manage insights into how your law firm is managed. An AI engine powers its case value estimator that pulls data from your CMS, CRM, and accounting systems. It will let you put a value on cases that consider multiple factors and rank the profitability of cases in the law firm’s portfolio.

The AI-Empowered Receptionist and Intake Clerk

Yes, she has a name—Amelia. has built a proprietary AI engine that replaces your receptionist and intake clerk. Amelia is your personal concierge. She is smart, learns quickly, and knows everyone’s calendar. She knows which clients you want to talk to and which clients you would rather send on to someone else. She has a pleasant voice, speaks in complete English sentences, and can simultaneously talk to thousands of people. She can answer your inbound calls, greet your website visitors, qualify leads, and make outbound client calls.

Not only can Amelia answer the phone and take messages, but she can ask questions and record the answers. If you are running an advertising campaign that drives potential new clients to your office, she can run through a series of questions to gather critical data from your callers. You design the script, and Amelia executes it flawlessly. Amelia can engage in a conversation, unlike the typical phone tree, which requires the user to punch in a series of numbers. If the interaction stresses the caller, Amelia can sense the caller’s frustration and route the caller directly to a live receptionist, someone in your office, or schedule a call back from a human. Amelia can work 24/7 and doesn’t charge for overtime. was another automated reception service exhibited at ABA TechShow. SmithAI offers a bundled service that includes live agents working with AI to provide receptionist services 24/7. Like OneLawAI, the service can pull data from your calendar or practice management system for a better user experience.

The AI-Enhanced Litigation Support Paralegal Team

Litigators have different needs than transactional attorneys, but they, too, can benefit greatly from AI. For litigators, AI engines enhance document, email, and video searches. Keyword and name searches in traditional eDiscovery tools are limited to full-word or partial-word matches. To find all references to “Alfred E. Neuman,” you would need a series of alternative searches for “Al,” “Alvin,” “Afred,” “Neuman, Al,” etc. With an AI-enhanced search, your search engine would automatically identify all variants and return them in the result set, including his childhood nickname of “Alfie.”

AI engines can provide a matrix of relations between documents and users, organized in a timeline using keywords and entity extraction. In addition, the newer AI engines can also search for concepts. Relativity is an eDiscovery engine that has technology-assisted document review ( AI can weigh the value of documents on a scale from responsive to non-responsive to a given question. If you find a beneficial document to your case, you can ask the system to find “more documents like this one.” In addition, you can enhance the search with sentiment analysis that looks at tone and emotion in the document.

LawDroid—The AI-Turbocharged Research Engine

Most practicing attorneys have an associate, law clerk, or paralegal who does legal research for them. The attorney states a position or legal argument she wishes to make and sends her assistant to find legal support for that position. The assistant returns shortly with a list of legal citations and language supporting that position, case summaries, and some contrary cases. The LawDroid (, Jurisage (, and CaseText ( are AI-powered research assistants who do the footwork for you in seconds. Their droids are trained in case law. Using a chat interface, you engage in a dialog. Each system can find your supporting case law and provide case summaries.

LawDroid’s Copilot will answer research questions. It will first help you narrow down the issue and refine your search. Once done, it will help you draft emails, letters, and document summaries. These systems can write the first draft or fill in the legal argument to support a point. LawDroid’s AI engine was trained on cases stored in the Harvard Law School’s Caselaw Access Project, which contains 6,930,777 unique cases from 612 federal and state case reporters. LawDroid Builder will let you design your own ChatBot for your website to engage with your clients and answer their questions, much like Amelia described above. This engine can showcase your law firm’s expertise and attract new clients.

CaseText offers a research engine that covers case law but can also research business history. The same tools that can answer questions about the law can also answer questions about your organization’s or clients’ contracts. It can evaluate unstructured data in the agreements and tell you which contain trade secret provisions or escalation clauses.

The AI-Assisted Contract Reviewer and Drafting Assistant

As a profession, our greatest fear is that robots will take over the legal drafting process; when they do, what will be left for us humans? Though there have been great strides in this direction, the first drafting products out of the gate are more akin to spelling and grammar checks. Picture a side panel in Microsoft Word. As you are drafting a brief, the side panel makes suggestions. If you like the suggestion, click on it, and it will be added to your document, whether it is a legal memorandum, a letter, an email, or a contract.

At ABA Techshow, I spoke to two vendors that offered drafting assistance software. Docgility ( focuses on contracts. It converts them into a model and allows you to view similar provisions in multiple contracts side by side. You load the contracts into the system and identify the key concepts. Then, as you review the agreement, you can review similar provisions in other agreements.

Spellbook ( appears in the side panel in Microsoft Word. As you highlight text in the agreement, it lets you cast a spell. The spell runs a search in the Spellbook AI engine and returns text to put into your document. You can describe what you want the language for, and it will find it. Spellbook can also be used to analyze the contract. It can list key provisions and even explain the contract to a five-year-old. Depending on your goal, it can beef up a contractual provision or tone it down.

The Future of AI-Powered Document Assembly

Chris Pearson, Director and Business Head of XpressDox (, spoke at ABA Techshow on how AI will affect document automation. In his presentation, Pearson pointed out that AI tools have several hurdles to overcome before they become widely adopted. They need to prove themselves trustworthy, they need to demonstrate consistent results for similar inputs, and they need to guarantee confidentiality. And at present, they need to overcome lawyers’ resistance to change.

He points out that the bridge may be document assembly. The same tool that provides entity extraction can be reversed to markup documents for automation. AI can be used to identify all the data points in a model agreement and place the correct variable tag in the form. AI can review a set of similar agreements, present alternative versions of a contract clause, group them by concept, and allow the document assembly coder to assign business logic to decide which alternative clause gets used.

In this way, AI can be used to turbo-charge the development of document assembly templates. These templates would be trustworthy because the lawyers chose the model forms. Assembly of new agreements based on these templates would be consistent because the end user follows a dynamic questionnaire that produces consistent results from similar inputs. And the data entered into these agreements would be confidential because the data would never leave the protected workspace surrounding the document assembly system.

Pearson points out, “If we look back on history, automation has never cost jobs—although it has changed them.” The practice of law is ever-evolving. We need to embrace the bots while understanding their limitations. They will enable transaction attorneys to write better contracts, contract managers to better understand their portfolios, litigation attorneys to have more insight into their cases, and appellate attorneys to write better legal briefs. But they will not replace us.