Features

The AI revolution: A look back at 2023 and the future of AI in the law

 
 

by Alexander S. Rusek   |   Michigan Bar Journal

 

2023 may go down in history as the “Year of Artificial Intelligence” and if it does not, it should at least be recognized as the “Year of Accessible Artificial Intelligence.” In late 2022, ChatGPT, a large language model artificial intelligence program, was publicly released,1 finally bringing an accessible AI tool to the masses. Since then, ChatGPT has proven itself as a very capable AI program, even passing the Uniform Bar Examination.2 During this short period of time, there has been an explosion of AI tools being released that can do things such as generate text, sounds, voices, pictures, and videos. AI has also crept into the legal industry in multiple ways. This article provides a brief background on the flood of AI tools and what to expect from AI in the near future.

WHAT IS AI?

While there is no agreed-upon definition of artificial intelligence, Stanford computer science professor John McCarthy, one of the early pioneers in the field, defined it as “the science and engineering of making intelligent machines, especially intelligent computer programs. It is related to the similar task of using computers to understand human intelligence, but AI does not have to confine itself to methods that are biologically observable.”3

ChatGPT, when asked by this author, “What is artificial intelligence?” replied, in part, that:

Artificial Intelligence (AI) is a field of computer science and technology that focuses on creating systems and machines capable of performing tasks that typically require human intelligence. These tasks include things like problem-solving, learning, reasoning, perception, understanding natural language, and making decisions. AI systems aim to replicate or simulate human-like cognitive functions and abilities.

ChatGPT,4 which is short for Chat Generative Pre-trained Transformer, was created and is maintained by OpenAI and is the most well-known generative AI internet chatbot.5 The underlying technology “is capable of generating human-like text and has a wide range of applications, including language translation, language modelling, and generating text for applications such as chatbots. It is one of the largest and most powerful language processing AI models to date, with 175 billion parameters.”6

In practice, using generative AI tools feels and looks very similar to using a traditional messaging application such as Slack, Facebook Messenger, iMessage, WhatsApp, or one of the countless others available. Put simply, a user types in a command or question into the website’s prompt and the program then provides a written answer (or image, sound, or video, depending on the system.) Users can then enter follow-up questions or commands and the AI system will use the additional inputs to refine its output. OpenAI explains that “[t]he dialogue format makes it possible for ChatGPT to answer follow-up questions, admit its mistakes, challenge incorrect premises, and reject inappropriate requests.”7

WHAT CAN AI DO?

Accessible AI burst onto the scene in late 2022 and its use expanded greatly throughout 2023. New AI technologies like ChatGPT are being created seemingly every day. In a little over a year since the public release of ChatGPT, generative AI has been used to pass the Uniform Bar Exam;8 cheat on grade-school tests;9 write and debug programming code;10 create workout routines;11 create or reference recipes;12 write music and film and television scripts;13 write a résumé and cover letter;14 write jokes;15 solve math problems;16 get a B grade on a Wharton business school exam;17 attempt to diagnose health problems;18 explain scientific and mathematical concepts at different levels of sophistication;19 and much more. AI systems can also assist with internet searches (e.g., Microsoft Bing), create electronic art (e.g., OpenAI Dall-E 2, Dream Studio, Midjourney, DeepAI, Remini, and Stable Diffusion), and generate sounds, voices, and videos.

WHAT ARE AI’S LIMITATIONS?

As one asks ChatGPT to perform increasingly complicated tasks, its limitations quickly become apparent. OpenAI itself acknowledges that ChatGPT may occasionally generate incorrect information, produce harmful instructions or biased content, and has limited knowledge of the world and events after a specified date — remember, ChatGPT is not a search engine and may not have the most up-to-date information in its knowledge bank.20 At least one attorney has been caught using AI to write a brief wherein the AI system simply made up case law.21 AI programs are contoured by the underlying data used to build and train it, which can result in AI seemingly becoming racist, sexist, or otherwise biased because of the data it used to “learn.”22

Other problems have arisen over the last year stemming from the proliferation of accessible AI tools. Issues related to using AI to cheat on tests in grade schools and higher education is a major concern, as is the issue of deepfakes. Deepfakes are defined as “a fake, digitally manipulated video or audio file produced by using deep learning, an advanced type of machine learning, and typically featuring a person’s likeness and voice in a situation that did not actually occur.”23

Deepfakes using the likeness of celebrities to sell goods and services without their authorization have begun to appear online.24 In one instance, a deepfake of actor Tom Hanks was used to advertise a dental plan.25 More disturbing are numerous reports of people using AI tools to generate pornographic images of others without their knowledge or consent.26 Several states have implemented criminal penalties for creation of nonconsensual pornographic deepfakes.27

Recently, authors such as John Grisham, Jodi Picoult, and George R.R. Martin have sued OpenAI alleging “systemic theft on a mass scale” for generating output based on their copyrighted material used to train the system.28 Intellectual property issues arising from the use of generative AI are only expected to grow. It should be noted that this article has only touched briefly upon the numerous legal and ethical issues surrounding the use of AI, and this author expects these issues to be hotly debated in the coming years.

AI IN THE LEGAL FIELD

In the legal field, AI can assist with research and cite checking; drafting and reviewing pleadings and other papers such as complaints and motions to compel discovery; drafting, analyzing, and reviewing transactional documents such as buy-sell agreements or operating agreements for a limited liability company; and analyzing and reviewing discovery, amongst other uses.

Both Thomson Reuters and LexisNexis have rolled out AI tools and intend to release more in the future. Relativity, perhaps the most well-known e-discovery platform, also offers AI-driven discovery review tools. Scott Wrobel, managing member of Michigan-based N1 Discovery, told this author that in the future, “it will be interesting to see how generative AI tools will assist attorneys with determining appropriate search terms based on the type of case or concepts within a case.”

TESTING AI IN THE REAL WORLD

ChatGPT and other AI systems are relatively easy to use but can be difficult to effectively utilize without recognizing and accounting for its limitations. Always keep in mind that AI can — and routinely does — provide inaccurate or incomplete information and is not a substitute for the legal advice of a licensed, qualified attorney. Attorneys should also use traditional research tools to confirm the accuracy of AI-generated documents and research. In general, AI out put is most useful and accurate when the user provides the program with detailed information and parameters for the desired output.

To try out ChatGPT for the first time, go to www.chat.openai.com. After creating a free account, you will be taken to the main ChatGPT web page where you are greeted by a simple search bar somewhat reminiscent of Google’s home page. Input your question or command; ChatGPT almost instantaneously begins generating its response on your screen. After ChatGPT delivers its initial response, users can ask additional questions and provide additional commands. ChatGPT will then craft a response that, hopefully, has taken the additional input and builds upon its previous answer. Users are also able to rate the accuracy of the response, which can be used to further train the system to provide more accurate and human-like responses in the future.

Let’s look at an example of how generative AI can help lawyers. In this example, ChatGPT was prompted, “Draft a cease-and-desist letter from the owners of the real property known as Blackacre to the owners of the real property known as Greenacre demanding that the owner immediately stop trespassing on Blackacre. Use Michigan statutes and case law to support the cease-and-desist letter.”

ChatGPT produced the template below: and provides a quick and easy starting point for attorneys to draft the needed letter, even if only as a means to overcome writer’s block.

CAUTION AND CONCLUSION

While there is significant room for improvement, AI can be effectively used by attorneys who understand its power but are also aware of and respect its limitations. As with most tools, how AI is implemented, and not necessarily its inherent characteristics, should be the measure of its usefulness. Attorneys aware of AI’s limitations and the knowledge to implement it within those parameters will be best positioned to leverage it for their clients’ benefit in the future. However, with that said, AI is not likely to replace attorneys anytime soon.

Finally, attorneys must always keep in mind their ethical obligations under the Michigan Rules of Professional Conduct (MRPC) when incorporating new technology into their practices. At a minimum, attorneys should keep in mind their obligations to become and remain competent under MRPC 1.1 and keep client information confidential under MRPC 1.6 — ChatGPT does not guarantee that your inputs will be kept confidential. While a full discussion of the ethical concerns surrounding AI is beyond the scope of this article, numerous resources have been published to help guide attorneys through the ethical minefield that AI presents.29


 

ENDNOTES

1. See OpenAI, ChatGPT https://chat.openai.com/ (all websites accessed January 19, 2024).

2. Debra Cassen Weiss, ABA Journal, Latest version of ChatGPT aces bar exam with score nearing 90th percentile, [https://perma.cc/ U5LZ-ZFR6] (posted March 16, 2023); see also Daniel Martin Katz, SSRN, GPT-4 Passes the Bar Exam, [https://perma.cc/DA4R-4EJG] (posted March 15, 2023).

3. John McCarthy, What is Artificial Intelligence? [https://perma.cc/9DMX-62XN] (posted November 12, 2007).

4. See OpenAI, ChatGPT https://chat.openai.com/.

5. See OpenAI, ChatGPT https://openai.com/.

6. Alex Hughes, ChatGPT: Everything you need to know about OpenAI’s GPT-3 tool, BBC Science Focus (February 2, 2023) [https://perma.cc/X8SS-F96M].

7. See OpenAI, Introducing ChatGPT (posted November 30, 2022).

8. Weiss; Katz, supra n 1.

9. Patrick Wood and Mary Louise Kelly, National Public Radio, Everybody is cheating: Why this teacher has adopted an open ChatGPT policy, [https://perma.cc/2QVN-RBAS] (posted January 26, 2023).

10. Ani Madurkar, Medium, ChatGPT is the End of the Beginning of the AI Revolution, [https://perma.cc/P4YP-FSR3] (posted January 2, 2023).

11. Id.

12. Id.

13. Id.

14. Maxwell Timothy, MUO, 11 Things You Can Do With ChatGPT, [https://perma.cc/YYR9-RT6M] (posted December 20, 2022).

15. Id.

16. Tech Deck, Indian Express, What happens when ChatGPT has to solve a basic math problem? Check out its response, [https://perma.cc/A5FZ-PYGE] (posted December 28, 2022).

17. Lakshmi Varanasi, Business Insider, A ChatGPT bot passed a Wharton business school exam, but a professor says he would’ve only graded the effort a B or B-minus [https://perma.cc/VVU5-VT3U] (posted January 23, 2023).

18. Kevin Roose, New York Times, The Brilliance and Weirdness of ChatGPT [https://perma. cc/7G5C-GJ6Z] (posted December 5, 2022).

19. Id.

20. See Open AI https://chat.openai.com/chat.

21. Benjamin Walker and Nate Schweber, New York Times, The ChatGPT Lawyer Explains Himself  [https://perma.cc/W8L8-HYA7] (posted June 8, 2023).

22. Cade Metz, New York Times, Who is Making Sure the A.I. Machines Aren’t Racist [https://perma.cc/58XU-WQUE] (posted March 15, 2021).

23. Dictionary.com, deepfake, https://www.dictionary.com/browse/deepfake.

24. Benjamin Edwards, Arstechnica, Deepfake celebrities begin shilling products on social media, causing alarm [https://perma.cc/JLX5-RLBX] (posted October 3, 2023).

25. Id.

26. Manuell Viejo, El Pais, In Spain, dozens of girls are reporting AI-generated nude photos of them being circulated at school: ‘My heart skipped a beat’ https://english.elpais.com/international/2023-09-18/in-spain-dozens-of-girls-are-reporting-ai-generated-nude-photos-of-them-being-circulated-at-school-my-heart-skipped-a-beat.html (posted September 18, 2023).

27. Isaiah Poritz, Bloomberg, States are Rushing to Regulate Deepfakes as AI Goes Mainstream https://www.bloomberg.com/news/articles/2023-06-20/deepfake-porn-political-ads-push-states-to-curb-rampant-ai-use (posted June 20, 2023).

28. Hillel Italie, AP News, Game of Thrones’ creator and other authors sue ChatGPT-maker OpenAI for copyright infringement [https://perma.cc/6SVN-U5S6] (posted September 21, 2023).

29. Bobby Allyn, National Public Radio, A robot was scheduled to argue in court, then came the jail threats [https://perma.cc/5HX9- KBYM] (posted January 26, 2023); Liz Dye, Above the Law, World’s First Robot Lawyer Shorts Out [https://perma.cc/S3TV-3HPZ] (posted January 26, 2023); Aimee Furness and Sam Mallick, Law360, Evaluating The Legal Ethics Of A ChatGPT-Authored Motion [https://perma.cc/2RCM-NJAL] (posted January 23, 2023); Foster Sayers, Legaltech News, ChatGPT and Ethics: Can Generative AI Break Privilege and Waive Confidentiality? [https://perma.cc/8X8Q-Y5E3] (posted January 26, 2023); Lance Eliot, Forbes, Generative AI ChatGPT Can Disturbingly Gobble Up Your Private And Confidential Data, Forewarns AI Ethics And AI Law [https://perma.cc/ DV86-DUYU] (posted January 27, 2023). The author of this article does not own any of these resources or guarantee that any of the information or viewpoints expressed in them is current or accurate.