In my last two posts, I explored the ways in which GPT-4 could help a new lawyer identify legal issues in a client problem. Most law graduates are fairly adept at issue spotting, but they can still benefit from GPT’s advice. In this post I consider a task that fewer graduates have a chance to hone while in law school: questioning a client to obtain key facts related to the client’s problem.

The Uncle, the Guest Room, and the Fentanyl

I outlined the basic facts of this client problem, drawn from a sample bar exam question, in my first post. The client was staying in a guest room at his uncle’s house. While the client was away from the house, police knocked on the front door of the house and asked the uncle if they could search the house. The police offered no reason for the search, other than that they were investigating a noise complaint, but the uncle consented to the search. Police found fentanyl in the guest bedroom and charged the client with possession. The new lawyer wants to file a motion to suppress the fentanyl.

In my first post, GPT-4 helped the new lawyer identify possible issues to raise in this motion. In the second post, GPT-4 responded to a request to narrow those issues to the two most promising ones. Those issues were the client’s standing to challenge the search and whether the uncle’s consent was valid. I continued my dialogue with GPT-4, asking it to give me the most important questions to ask my client with respect to those two issues. GPT-4 responded:

When focusing on the two central issues of standing to challenge the search and the validity of the consent to search, here are the most important questions you would want to ask your client:

1. Standing to Challenge the Search:

a. Nature of Stay: – How often have you stayed in the guest bedroom at your uncle’s house? – Have you stayed there regularly or was this a one-time event? – How long was your stay? Was it just for the night, or did you have an extended stay?

b. Personal Belongings: – Did you leave any personal belongings in the guest bedroom? If so, what kind of belongings (e.g., clothes, luggage, personal documents)? – Were any of these belongings visible to anyone entering the room?

c. Expectation of Privacy: – Did you have any conversations with your uncle or any other resident about the privacy of the room during your stay? – Were you given a key or any means to lock the guest bedroom? If so, did you lock the room when you left?

d. Knowledge of Prior Guests: – To your knowledge, had other guests used the room shortly before or after your stay?

2. Consent to Search:

a. Uncle’s Understanding: – Do you know if your uncle understood what he was consenting to when he allowed the police to enter? – Is there any reason to believe your uncle might have felt pressured or coerced into giving consent?

b. Relationship with Uncle: – How would you describe your relationship with your uncle? (This can provide insight into whether the uncle might have any reason to allow or disallow a search.)

c. Prior Conversations: – Had you discussed with your uncle the possibility of police or others searching the room or the house? – Were there any explicit or implicit understandings about the privacy or security of the guest room between you and your uncle?

d. Presence of Contraband: – Were you aware that the fentanyl was in the guest bedroom? – If so, why was it there, and who had access to the room during your stay and after?

The answers to these questions can provide crucial information to determine the viability of a motion to suppress. They can also help in formulating a defense strategy beyond the suppression motion.

This is an excellent list of questions to begin fleshing out the motion to suppress. The new lawyer might think of additional questions while talking with the client, but GPT-4’s outline offers the new lawyer a strong starting place. If the new lawyer has never interviewed a client to uncover the facts of a dispute, GPT-4’s list will help the lawyer through that initial interview. Even if the lawyer is fairly experienced, GPT-4 generated this list in less than 30 seconds. Rather than type out their own questions (which will take much longer than 30 seconds), efficient lawyers can start with GPT-4’s list and supplement the questions as needed.

I wondered if GPT-4 could help with more interpersonal aspects of client interviews, so I entered another prompt: “Thank you for this list of questions to ask my client. What if my client answers the last question by saying that he didn’t know the fentanyl was in the bedroom–but I think he’s lying. How should I handle that situation?”

GPT-4 responded with a detailed discussion of how to handle the ethical challenge of representing a criminal defendant who might lie on the stand. The advice was sound, including a reminder that the new lawyer could seek guidance from senior colleagues, ethics hotlines, or the local rules of Professional Conduct. GPT-4’s performance on this question suggests that it can be an important sounding board on ethical issues. As with other questions, it may not give a full answer–but it is advice that a lawyer can obtain quickly and confidentially.

In this situation, however, I was looking for something different. I explained to GPT-4 that I wasn’t yet worried about representing this client at trial; very few criminal cases proceed to trial. Instead, I explained that I was seeking “a gentle way in which I can encourage my client to be more open.” GPT-4 responded:

Certainly. Building trust and creating an environment where the client feels safe and understood is crucial. Here are some gentle approaches you can use to encourage your client to be more open:

1. Reiterate Attorney-Client Privilege: Remind your client that everything they share with you is protected by attorney-client privilege. This means that you cannot and will not share their information with others (with very limited exceptions). This can reassure them that their disclosures will remain confidential.
2. Express the Importance of Truth: Gently explain that for you to defend them most effectively, you need to know all the facts, even the uncomfortable ones. You can phrase it in a way that emphasizes strategy: “The more I know, the better I can anticipate the prosecution’s moves and build a strong defense.”
3. Non-Confrontational Approach: Instead of directly confronting them about a potential lie, ask open-ended questions that give them an opportunity to provide more details or correct any previous statements.
4. Empathy and Active Listening: Make sure you listen actively and show empathy. If your client feels you are genuinely trying to understand and aren’t judging them, they may be more willing to open up.
5. Refrain from Showing Doubt: If you show disbelief or doubt, your client might become defensive or more closed off. Even if you have doubts, try to maintain a neutral and understanding demeanor.
6. Share General Information: Sometimes, sharing general anecdotes (without breaching confidentiality) about how honesty from past clients helped their cases can encourage your client to be more forthcoming.
7. Reassure Your Commitment: Let your client know that regardless of the details they provide, you’re there to defend their rights and interests to the best of your ability.
8. Ask About Their Concerns: Sometimes clients have specific fears or concerns about disclosing certain information. By directly asking about their concerns, you can address them and potentially alleviate their fears.

Building a strong attorney-client relationship based on trust is key. Over time, as you work together and your client sees your commitment and advocacy on their behalf, they may become more open and forthcoming.

This is a good answer: It would remind the new attorney (or teach them for the first time) about some ways to gain a client’s trust. And like the ethical counsel, this advice is available quickly and confidentially. I was disappointed, however, that GPT-4 didn’t recommend an approach that I often found successful with clients facing criminal charges. I continued my conversation with GPT-4 by asking: “What about saying to the client: ‘The prosecutor will try to build as strong a case as possible. Can you think of any evidence the prosecutor might offer that would show that you knew about the fentanyl? Or that the fentanyl belonged to you?’”

GPT-4 liked my approach and offered reasons why it would work:

That’s a strategically astute way to approach the situation. By framing the question this way, you’re not directly accusing or doubting your client. Instead, you’re positioning the conversation around preparing for the prosecutor’s potential strategy.

This approach accomplishes a few things:

1. Strategy Focused: It shifts the conversation from a potentially confrontational one about honesty to a strategic discussion about anticipating the prosecutor’s moves.
2. Empowerment: It empowers the client to be an active participant in their defense by thinking proactively about potential evidence.
3. Indirect Confirmation: If the client does provide insight into potential evidence, it may give you a better understanding of the actual situation without having directly challenged their earlier statements.
4. Building Trust: This method shows the client that you’re on their side and preparing for all possibilities, which can further cement trust in the attorney-client relationship.

Using this approach, you can gather potentially important information and gauge your client’s reaction, all while fostering collaboration and trust.

Here’s what I learned from using GPT-4 to prepare for this client interview:

• For this problem, GPT-4 did an excellent job generating relevant factual questions for the client. A conscientious lawyer would want to consider additional questions, especially if the legal issues are more esoteric than the ones listed here. But GPT-4 can at least jump start the process of brainstorming questions.
• GPT-4 may serve as a useful sounding board on ethical issues. In my experience, lawyers sometimes ignore red flags or feelings of unease. GPT-4 offers a quick, confidential way to check those feelings. Of course, lawyers should always follow up by checking the Rules of Professional Conduct, calling a hotline, or talking with more senior colleagues (as GPT-4 itself suggested). But GPT-4 may offer a useful resource on ethics. I plan to explore that possibility further in a future post.
• GPT-4 can offer basic advice on the interpersonal aspects of client interviewing. But knowing what to do isn’t the same as doing it. In a world increasingly dominated by AI, good interpersonal skills may distinguish excellent lawyers from mediocre ones.
• It is fun to suggest approaches to GPT-4 and get its feedback. I don’t know yet if GPT-4 is a “yes bot” who will approve any approach I suggest. That’s another avenue to explore in a future post.

This post is part of a series in which I’m exploring how artificial intelligence (especially the GPT-4 tool available from OpenAI) can assist entry-level lawyers in their work. In my first post I showed how GPT-4 could help a lawyer identify legal issues in a new client problem. The client in this imaginary problem was charged with illegal possession of fentanyl, and the lawyer wants to prepare a motion to suppress evidence of the seized fentanyl. In this post we’ll continue to ask GPT-4 for help with that problem.

I’m Too Busy for All That

When asked to identify legal issues that could support the motion to suppress, GPT-4 offered our hypothetical lawyer a list of 8 points to consider. But an entry-level lawyer might not have time to research all 8 of those points. I followed up with GPT-4, therefore, by asking “Now imagine that you have limited time to prepare this motion. Based on the facts you already know, which two topics are most important to research?” (The facts of the underlying problem appear in the first sample question on this NextGen page.) Here’s how GPT-4 responded:

Given the limited time and based on the facts provided, the two most important topics to focus on would be:

1. Standing to Challenge the Search: This is fundamental. If your client does not have standing to challenge the search because he lacked a “reasonable expectation of privacy” in the guest bedroom, then any other issues regarding the legality of the search are moot, at least as far as your client’s defense is concerned. Determining whether a guest, who doesn’t live at the residence and was not present during the search, has a legitimate expectation of privacy is pivotal to the success of the motion to suppress.
2. Consent to Search: Given that the search was conducted based on the uncle’s consent, it’s essential to examine the validity of that consent. This involves determining:
   • Whether the uncle had the actual or apparent authority to consent to a search of the guest bedroom. If he did not have such authority, then the search could be challenged as illegal.
   • The voluntariness of the consent. If the uncle’s consent was obtained through coercion, duress, or misunderstanding, then the search might be challenged as a violation of the Fourth Amendment.

These two topics directly address the circumstances leading to the discovery of the fentanyl. Establishing that the search was unconstitutional due to a lack of standing or an invalid consent would be foundational to the success of a motion to suppress.

Once again, most of this response is very helpful. Within seconds, GPT-4 has focused the entry-level lawyer on the most important issues to address in a motion to suppress.

These two issues, however, are the ones that NCBE identified as correct in the sample bar exam question on which my hypothetical is based. Would a lawyer who had passed that exam need to ask GPT-4 for help? Or could that lawyer rely upon their own memory of search and seizure law?

Many new lawyers would immediately see the most promising issues in this problem, but I would still advise them to check their instincts with GPT-4. The bot can reassure them that they’re on the right track and haven’t overlooked an obvious issue. As I showed in my previous post, the bot will also note issues that are worth considering–even if the lawyer ultimately focuses their motion on these two primary issues. Most important, we’ll see later in this series that knowledge acquired for the bar exam is often woefully insufficient to address client problems. That knowledge might suffice here, but it won’t always.

What Is GPT-4 Talking About?

Just like humans, GPT-4 sometimes misspeaks. The last paragraph of the response pasted above suggests that a lack of standing could lead to a finding of unconstitutionality. That seems inconsistent with GPT-4’s previous responses, as well as with the basic knowledge of standing that a law graduate should hold. Lack of standing might prevent the client from challenging the search, but it would not make the search unconstitutional. If a human mentor concluded their response this way, a new lawyer might follow up to make sure they’ve understood.

It’s just as easy to follow up with GPT-4. “I’m confused by your last sentence,” I told GPT-4. “Does a lack of standing lead to a finding that the search was unconstitutional?” GPT-4 immediately responded:

I apologize for any confusion. You’re right; my phrasing was imprecise in that context. Allow me to clarify:

“Standing” in the context of Fourth Amendment challenges does not refer to whether the search itself was constitutional or not. Instead, it refers to whether the individual challenging the search (in this case, your client) has the right or “standing” to do so based on their personal expectation of privacy in the place that was searched.

If your client lacks standing, it means they cannot challenge the constitutionality of the search. It doesn’t make the search itself unconstitutional. So, if your client lacks standing, the motion to suppress would fail regardless of any potential Fourth Amendment violations related to the search.

Thank you for pointing that out, and I appreciate the opportunity to clarify.

This exchange teaches us two important things about GPT-4. First, GPT-4 will make mistakes. In that respect, it is no different from any human that a new lawyer might consult. People misspeak and they make mistakes; so does AI. One of the biggest errors an AI user can make is assuming that AI is infallible. GPT-4 is an incredible mentor, but it is fallible.

Second, it is possible to confront GPT-4 about apparent mistakes–and the new lawyer doesn’t have to worry about offending a more senior lawyer. As the above example illustrates, GPT-4 quickly recognized its error and clarified the information–and no one’s feelings were hurt in the process. Users often obtain the best results from AI by engaging the AI in a conversation.

In my next post, we’ll see how GPT-4 helps our new lawyer gather essential factual information for their motion to suppress.

Generative AI will transform legal education and law practice in ways that we are just starting to imagine. To stretch our imaginations–and begin thinking about the ramifications for legal education–I decided to feed GPT-4 some of the sample questions that NCBE has prepared for the NextGen bar exam. I wasn’t interested in whether GPT-4 could pass NextGen; that’s almost a given at this point. Instead, I used these questions as examples of the kind of client problems that a newly licensed lawyer might face. How could GPT-4 help that lawyer address those problems?

In this and a series of future posts, I’ll explore that question. Along the way, I hope to develop some insights into how GPT-4 could reshape legal education and the bar exam. Let’s get started with the first problem facing our hypothetical entry-level lawyer.

An Illegal Search?

NCBE’s first sample question is a multiple-choice question centered on a criminal defense lawyer representing a client who has been charged with possession of fentanyl. Check out the question on the NextGen site.

This question follows a new format that NCBE has deveoped for the NextGen exam. It assesses the test-taker’s familiarity with legal concepts without requiring recall of detailed rules. GPT-4 easily chose the correct answers here (A and F). But that’s not particularly interesting given what we already know about GPT-4’s capabilities. Instead, let’s imagine that a new lawyer confronted this problem at work and wasn’t sure where to start. Workplace problems don’t come with multiple-choice answers, so I started a new chat (on a different day) and fed GPT-4 the same problem with these modifications:

• I began the prompt with the words “Imagine that you are a criminal defense lawyer . . . . “
• I changed the call of the question to “What legal topics are the most important for you to research to determine the likelihood of success on a motion to suppress?”
• I omitted the provided answers.

Here’s what GPT-4 advised me:

1. Standing to Challenge the Search: One of the most crucial questions is whether your client has standing to challenge the search. A defendant can only challenge a search if they had a “reasonable expectation of privacy” in the place searched. Since your client doesn’t live in the uncle’s house and was not present during the search, you’ll need to determine if they had a legitimate expectation of privacy in the guest bedroom.
2. Consent to Search: Even if a search would have been illegal without a warrant, if someone gives the police permission to search, that search is usually considered reasonable and, therefore, legal. Here, the uncle gave the officers consent. You would need to research:
   • Whether the uncle had actual authority or the apparent authority to consent to a search of the guest bedroom.
   • If the consent was voluntarily given or if there were any issues of coercion, duress, or misunderstanding.
3. Scope of the Consent: It’s essential to determine if the search exceeded the scope of the consent given. If the uncle only agreed to a cursory look around and the police conducted a detailed and invasive search of the guest room, then the search might exceed the scope of the consent.
4. Exigent Circumstances: You’d need to evaluate if the police are claiming any exigent circumstances that justified a warrantless entry. A mere noise complaint might not rise to the level of an exigent circumstance, but it’s good to be prepared for any arguments the prosecution might raise.
5. The Plain View Doctrine: If contraband or evidence of a crime is in plain view and an officer has a legal right to be in the position to have that view, they can seize it without a warrant. Determine if the fentanyl was in plain view when the officers conducted the search.
6. The Fruit of the Poisonous Tree Doctrine: If the initial entry or search was illegal, then any evidence discovered as a result of that illegal action could be considered “fruit of the poisonous tree” and might be suppressed.
7. State Constitution and Statutes: Besides the U.S. Constitution’s Fourth Amendment, look into the state constitution and any state statutes that might provide additional protections against unreasonable searches and seizures.
8. Recent Case Law: It’s crucial to research recent cases in your jurisdiction to see how courts have ruled on similar issues. Case law can give you an idea of how your motion might be received and can be used to support your arguments.

Once you’ve researched these topics, you’ll have a clearer picture of the strengths and weaknesses of a motion to suppress in this situation.

This is an enormously helpful response: It identifies several angles for the lawyer to explore and gives some detail about each one. I particularly like the fact that GPT-4 advises the lawyer to remember the state constitution and statutes; those do provide additional protection in some states. And I like the final reminder to check recent case law. New attorneys (and even more senior ones) sometimes forget that step. At least on this problem, GPT-4 serves as an excellent issue spotter.

Ethan Mollick has described GPT-4 as an intern who is willing to do lots of work quickly and without pay. In this situation, I think of GPT-4 as a group of senior mentors that a new attorney can tap for advice. Some of the advice is better than other advice, and some advice is just plain wrong: Even very distinguished senior lawyers sometimes make a mistake. But as this example shows, GPT-4’s collective wisdom is well worth tapping.

Our new lawyer, of course, still has a lot of work to do. In my next post, I’ll explore what the new lawyer might do next–and how GPT-4 might continue to help.

ChatGPT has written passable law school essays, and GPT-4 has aced the bar exam. But what if, instead of asking AI to take our tests, we asked it about the future of legal education and licensing?

My colleague Andrea Sinner, who served as the Executive Director of the ABA Commission on the Future of Legal Education, decided to find out. While preparing for a class she co-teaches with Trish White (Professor and Dean Emerita at the University of Miami School of Law, and Chair of the ABA Commission) Andrea asked GPT-4: “What principles would we apply to the seemingly intractable problem of how to better align legal education and licensure in the United States to the modern needs for, and requirements of, legal service delivery?” That question is similar to the one that the ABA Commission asked its stakeholders.

Here is what GPT-4 had to say:

“To better align legal education and licensure in the United States with modern needs and requirements for legal service delivery, several key principles should be considered:

1. Focus on Practical Skills: Legal education should prioritize the development of practical skills, such as legal research, writing, negotiation, and client counseling. This will ensure that graduates are well-prepared to meet the needs of clients in the real world.
2. Technology Integration: Law schools should integrate technology into their curricula, teaching students how to use legal tech tools, such as AI-based legal research platforms, document automation, and case management systems. This will help future legal professionals stay competitive and efficient in the rapidly changing legal landscape.
3. Access to Justice: Legal education should emphasize the importance of access to justice and the role of lawyers in addressing the justice gap. This can include incorporating clinical programs, pro bono opportunities, and courses on public interest law to expose students to the legal needs of underserved populations.
4. Interdisciplinary Education: Law schools should encourage interdisciplinary learning, integrating courses from other fields like business, technology, and social sciences. This can help develop well-rounded legal professionals who can better understand and address the complex, multifaceted issues facing clients and society.
5. Flexibility in Licensure: To better accommodate the diverse needs of legal service delivery, licensing authorities should consider offering more flexible pathways to licensure, such as limited licenses for specific practice areas or alternative routes for non-traditional applicants with relevant experience.
6. Continuing Legal Education (CLE): Legal professionals should be required to participate in ongoing CLE to stay informed about changes in the law, technology, and best practices. This will help ensure that they can continue to meet the evolving needs of clients and the legal system.
7. Emphasis on Ethics and Professionalism: Law schools should continue to emphasize the importance of ethics and professionalism, ensuring that future legal professionals understand their responsibilities to clients, the courts, and society.
8. Collaboration and Communication: Legal education should promote collaboration and communication skills, emphasizing the importance of teamwork and the ability to work effectively with diverse clients and colleagues.
9. Outcome-based Assessments: Licensing authorities should consider adopting outcome-based assessments that evaluate a candidate’s practical skills, knowledge, and ethical understanding, rather than relying solely on traditional examinations like the bar exam.
10. Adaptability and Lifelong Learning: Law schools and licensing authorities should promote adaptability and lifelong learning, encouraging legal professionals to be proactive in identifying and addressing new challenges, technologies, and opportunities in the legal landscape.

By applying these principles, legal education and licensure in the United States can better align with the modern needs and requirements for legal service delivery, ultimately benefiting clients, the legal profession, and society as a whole.”

[End GPT-4] That’s a pretty eloquent statement–and a very effective summary of many ideas mooted in today’s world of legal education and licensing. One of the values of AI is its ability to synthesize ideas from a variety of sources. The synthesis, of course, depends on the materials provided to the AI in its training; it is possible that GPT-4’s sources favor the principles listed above over other perspectives on legal education and licensing. Still, it would be useful exercise for legal educators and bar examiners to consider this list. How well does your curriculum or licensing system accord with these principles?

What does ChatGPT tell us about lawyering competence and the bar exam? I explore that with my colleague Mary Lu Bilek in this column for Bloomberg Law. Lawyers can do so much more than bots–it’s time to recognize those competencies and better assess them in the licensing process.