Introduction - Amy Swaner
This week’s edition of AI, Software, and Wetware features an audio interview with Amy Swaner. Amy is a 🇺🇸 USA-based lawyer, the author of AI For Lawyers, and the founder of legal tech startup Lexara. We discuss:
Why she sees AI as the perfect ‘field leveler’ for small firms and solo lawyers
How some in the legal community are enthusiastic about AI while others focus on safety and risks
Creating an AI-based tool to handle legal intake interviews and conflicts checking
How ChatGPT helped her install wainscotting in her house by herself
Using Claude Cowork to create an interface mockup for her startup
Why she emphasizes ‘humans in charge and AI in the loop’, not ‘humans in the loop’
How she handles client confidentiality and privacy with AI tools in her startup
and more. Check it out, and let us know what you think!
This post is part of our AI6P interview series on “AI, Software, and Wetware”. Our guests share their experiences with using AI, and how they feel about AI using their data and works.
This interview is available as an audio recording (embedded here in the post, and later in our AI6P external podcasts). This post includes the full, human-edited transcript. (If it doesn’t fit in your email client, click HERE to read the whole post online.)
Note: In this article series, “AI” means artificial intelligence and spans classical statistical methods, data analytics, machine learning, generative AI, and other non-generative AI. See this Glossary and “AI Fundamentals #01: What is Artificial Intelligence?” for reference.
Interview - Amy Swaner
Karen: I’m delighted to welcome Amy Swaner from Iowa USA as my guest today on “AI, Software, and Wetware”. Amy, thank you so much for joining me on this interview! Please tell us about yourself, who you are, and what you do.
Amy: Thank you. I am a fan and so I’m really excited to be here. I am a lawyer. And I also just recently started a legal tech company. I’m building AI tools for lawyers. I’ve been consulting with lawyers about AI and teaching them, and now I’m excited to be offering some really safe, compliant products for them.
I was bar-licensed in 2000, and I’ve been practicing or teaching law or running my company since that time. So a long time. I started working with AI when it really became popular back in, like, October of 2023, and fell in love with it. I started very naturally, like answering people’s questions about it, and started writing about it. And I mostly started writing about it so that I could learn about it for myself and help lawyers understand.
There are a lot of misconceptions now, but I feel like there were even more back then. So when I was talking to lawyers and law firms about it, I did not have a good tool that I could recommend for new clients coming in. So when I was practicing law, it takes a lot of time to interview potential clients and run complex checks. Those are very important tasks, but they take a lot of time and it’s not always billable time. And as lawyers, you only have your time to sell. And so if you spend an hour working on a matter that you’re not even able to take, because you have a conflict between the matter that you’re looking at and a client that you used to represent, you have just lost an hour of your income. And the more I learned about AI, I realized there needs to be a tool for this, because this is such a safe and compliant easy way to use AI. Since I couldn’t find a tool, I wanted to come up with one, and it’s called Lexara Engage. It conducts a smart intake interview.
Almost all AI tools are built on top of a frontier model. I built mine on top of ChatGPT and Claude. And it conducts an interview in a way that will get the questions answered that lawyers need in order to represent clients. And at the same time, while it’s talking to them, it also runs a conflicts check. So the idea is not to take away time from lawyers, but to give lawyers their time back, streamlining things that they need to do in their practice.
Karen: That’s a really great application that you’re addressing. These intake interviews, I guess the lawyers themselves need to do it – it’s not something that an assistant or paralegal or anyone else can do for them?
Amy: Actually, a lot of people do have an assistant do it for them. Or if they’ve trained their staff well, they’ll have staff do it if they can, because it takes such a large amount of time. I know that I had to do it when I was at a bigger firm when I first started practicing. I did a lot of intake interviews and so I was evaluating the claims. I didn’t do the conflicts checks. I feel like it definitely takes a level of skill. And so a lot of small firms and solo lawyers want to do it themselves, which is completely understandable.
But small firms and solo lawyers have so many constraints on their time. They are everything. They are the administrator. They are the boss. They’re taking care of so many things that if we can streamline anything for them, especially if it’s something that AI can do easily and we’re not taking away from their billable time, we’re giving it back, all the better.
Karen: That makes total sense. A lot of the people that I talk to, and especially a lot of the people that I meet on Substack, are solopreneurs or entrepreneurs, or people that have just started to go off on their own and have left corporate life. We’re finding a lot of people who are open to this kind of assistance – ways to, as you say, get their time back.
Amy: Yeah. I feel like AI is the perfect field leveler for small firms and solopreneurs or solo lawyers, because it gives them so many advantages, they can actually now compete with bigger firms. But it feels like, in my time talking to lawyers about AI, I really see two different groups.
We have one group where they are so excited about AI, they’re diving in. They want to get all the benefits of it, because we are very competitive by nature. They even train you for that in law school. And the people who gravitate to law tend to be competitive. So I see some that are just diving in a little too blindly.
And then the other group I see are people who are so scared to touch AI and feel like it’s unsafe and it’s ruining the practice of law or what have you. And there are problems with both those approaches.
Karen: Yeah. Dr Sam Illingworth likes to talk about how the real progress to be made is in what he calls the messy middle – not at either extreme. And that’s where the more interesting discussions take place and where the progress could be made. I’ve talked with a couple of lawyers who are very concerned, looking more at the privacy aspects and also looking at what some people call hallucinations or confabulation where it makes up citations, which is obviously a very big no-no in the world of law. I’ve talked to other people that have worked on trying to solve that problem, of making sure that a tool doesn’t give you invalid references. But you’re approaching the back office of law operations. And that’s, I think, an interesting area, where obviously, from your experience as a lawyer, you’ve seen all of the burdens that lawyers have to deal with in actually just running their business.
Amy: Exactly. So after practicing at a firm, one of my daughters became seriously ill. It’s been an ongoing thing for her. So my husband and I took a serious look at our family and we decided that I would stop working to help her and take care of her. And I ended up, after only a few months, having a number of clients ask me to start my own firm. So I practiced as a solopreneur, a solo practice lawyer, for years. And it’s tough.
I understand the privacy concerns, but I think you can use AI in confidential ways with confidential information. And I get the worry with hallucinations. I feel like they are a natural part of this process. They’re baked into LLMs for various reasons. And that’s probably outside the scope of what we’re talking about right now. But in my mind, this is an area where hallucinations are so unlikely, so rare. I can’t ever guarantee that there’s never going to be a hallucination. But there are certain uses of AI that are a lot more hallucination prone and some that are a lot less hallucination prone, as you know. And I feel like this is one of the safer ones.
Karen: So applying it for things like the intake interviews and looking for the conflicts, you’ve had good success with that being fairly accurate, then?
Amy: Yes. It took a long time to get it right. The front end of it is a chat with the potential client and that looks like just any chat bot. It’s different because we spent so much time training it and giving it so many constraints and goals, and working with it to get a tone that’s friendly, that gets all the information that you need without getting too much information, that it won’t give legal advice. Again, there’s always a possibility for a hallucination. I am not a person who will say never. But it is something that we’ve spent a lot of time working on and we’ve had a lot of success with it.
Karen: That sounds great. So you mentioned that you started using it when ChatGPT first broke into common visibility. Have you studied any of the technologies underneath it, or is it primarily focused on how you apply it?
Amy: I studied the technologies beneath it. Because I’m one of those people where I have to know why and I have to know how. And I’m guessing from reading things that you’ve written, you feel the same way. I can’t say that anyone completely understands generative AI, but I wanted to know as much about the functioning of it as possible. And so I feel like that really was worth the time and the blood, sweat, and tears that sometimes went into learning more about it. Because I will sit through AI presentations and I will think, “Oh, that’s not accurate, because they’re not understanding the technology behind what they’re saying.”
Karen: Yeah, some people call it data literacy or computer literacy, but it’s really more than that. It’s more knowing enough to ask questions and to challenge some of what you’re hearing and to be able to really make an informed judgment about whether something can be trusted, whether it’s coming from a human or coming directly from an LLM.
Amy: Exactly. Yeah. I mean, you would verify most things or anything important that you hear from a human, you would verify. And so of course you need to do that with any output that you get from an AI tool.
Karen: I’m curious: you indicated that you have used it both professionally and personally. Can you say a bit about some of the ways that you have used generative AI tools personally?
Amy: I love that question, Karen. I am excited to answer that. I’ve used it for things that I’m not amazing at. I’ve kind of taught a lot of my friends who aren’t lawyers, “Oh, just look it up on ChatGPT” or “Send a picture of that to ChatGPT”.
One of the things that I did, where everyone was out of my house except for me for a whole week, and to entertain myself at night I put wainscotting in my master bedroom. And I am lousy at math. So I just gave all of the information to ChatGPT at the time. This was probably a year ago, and it gave me every single calculation I needed for “Here’s where you put all those frames”. Because wainscotting, as you probably know, but in case anyone doesn’t know, it’s almost like square or rectangular frames on the wall that you put in measured spaces. So you want to have the right amount of space between each one so that it looks good on the wall and looks well-spaced. Each square individually is the same size. And it was amazing. It was phenomenal at that. My husband came home after a week and was like, “You did this?” And I loved it, yeah. There are all sorts of things to use it for. We’ve used it for everything from setting up a party to figuring out catering for my daughter’s wedding. All sorts of things.
Karen: Nice. Those are good examples. I’m kind of pleasantly surprised to hear that it did so well on the measurements for your wainscotting, because I’ve heard a lot of stories about how it’s not at all good at math.
Amy: I definitely feel like ChatGPT was not good at math for a long time. And I remember the days, I’m sure you do too, where it couldn’t count the number of Rs in strawberry, and it couldn’t add 3 + 3 correctly, or whatever. But it did, it was amazing. Of course, I double-checked before I used it, but it did a phenomenal job with those. And I feel like we are really seeing those improvements with the capabilities of AI tools.
Karen: Very good. So you’ve already given a good description of how you used Claude and ChatGPT for Lexara, for the business. I’d like to hear about some other AI tools that you’ve used. It doesn’t have to be just for processing words, like for your intake interviews, but it could be anything. For instance, there are AI tools and LLMs that don’t just create texts, they also create images. They create videos. Some of them create music and such. Have you ever experimented with any of those?
Amy: I have. I remember back in the old days when ChatGPT images based on DALL-E had weird fingers and too many hands or whatever. I look back at the pictures that I had ChatGPT create a few years ago, and they’re horrible compared to what it can do now. And that’s exciting to me, and also a little bit scary to me, because we are losing that uncanny valley where we can easily see what’s AI and what’s not AI.
But back to your question, one of the things that I’m the very most excited about, and this is for my business – I am so excited about Claude Cowork. I’ve used all of the general-purpose tools and a number of the legal specific tools throughout the past few years. And I used to really like ChatGPT. Then I felt like Claude kind of took over. And I tried Claude Code and I didn’t love it. I felt like, “Oh, it’s kind of fun to create this interface” or whatever, but I’m not a coder. And so I thought, “I’m just going to try this out.” I wanted to be able to do a dummy interface where I could show what my AI tool does, but without gathering any information, without storing it anywhere, but still have a fairly authentic experience of what it was like to use my AI tool.
I have three software engineers and I asked them “Can one of you build this for me?” I thought it would be very quick. And one of them said, “I can do it. It’ll be probably about two weeks.” So even if you think 20 hours a week, that’s 40 hours. Or if you think 30 hours a week, that’s 60 hours. I did it with Claude Cowork in 20 minutes and it was beautiful. It works incredibly well. It linked right to my website and I was astounded.
I don’t think we’re in jeopardy of losing humans in the workplace. I don’t think we’re in a post-work society. But we are definitely in a shift similar to the industrial revolution, where we’re going to need to learn how to work with these tools. Because they are phenomenal at doing some of the heavy lifting that we can let them do so we can do more fun stuff.
Karen: Yeah, if you have three software engineers, I’m guessing that you’re using them to handle things like making sure that whatever databases you are using, or systems mechanisms you have, are secure and scalable and all the important qualities which AI tools tend not to get right, but the humans are generally very good at. I think the tools haven’t replaced that part of it yet. But for just cranking out a prototype or a website or something like that where those aren’t considerations, they can really save a lot of time, from what I’ve been hearing from people. And it sounds like you had that experience.
Amy: So one of my software engineers uses AI a lot as he’s coding, and then he just manages it, and he gets so much done so fast. One of my engineers has his training in cybersecurity and he does use some AI, but like AI in any other use case, there needs to be a human in charge. And I don’t even like to say ‘human in the loop’. I like to say ‘human in charge’, because that’s really what we are. We shouldn’t be getting acted upon. We should be taking the initiative and controlling it.
I used Claude Cowork to make another thing for me, a personal calendar, and it was like talking to a little kid. I had to tell it every single thing. And I had to say. “Wait a second. You said this was where I could capture items. You didn’t capture them; you caught them and released them. That’s not what capture means.” So that one took a lot more time.
But yeah, you’re right. I will not code without humans. I don’t feel like just coding on my own with Cowork is safe. I need a human there, knowing that it’s going to function the way it’s supposed to function, not break something else.
Karen: Yeah. I like your characterization of humans in charge and AI in the loop, as opposed to the humans being in the loop and the AI being in charge.
Amy: I keep meaning to do an article or a post about that, and I haven’t done it yet.
Karen: Ah, you definitely should. And please tag me when you do.
Amy: Definitely, yes.
Karen: There’s so much talk about how it’s taking over jobs. It doesn’t seem like what you’re doing with Lexara is going to take over an attorney’s job. As you said, it would really make it possible for smaller firms to compete with larger firms that have assistants who can handle some of those things.
Amy: Yeah, I do think that it will still potentially take up some things that a human was doing. But it’s all part of that adjusting. I read a great book about the weavers back in England when the industrial revolution was happening in the UK, and how they were burning down the factories where yarn was going to be treated and wool was going to be woven. And I’m not saying this very clearly, but they were trying to stop progress and they lost. And I wrote an article about it because I feel like that’s where we are. If we’re fighting against AI, we’re going to lose. It’s transformative. We need to use it and make the most of it.
And another article is on how we’re staring straight in the face of a situation where it will be malpractice in law not to use AI. That’s my opinion. No one’s told me that, or necessarily even agreed with me, but I feel confident we’re close.
Karen: Yeah, the example with the Industrial Revolution, that comes up a lot. And I think the Luddites are often misunderstood. It wasn’t so much that they didn’t want the technology, but it was the exploitative way that the companies were using it and affecting their jobs. And that’s what they were railing against, more so than the machines themselves. And I think that analogy also does still apply in the world of AI. I think a lot of people look at the eight-figure tech bro, massive companies that are dominating the industry and saying, “They’re using this exploitatively. And it’s not that we don’t want or like the technology. But the way that it’s being developed and deployed is where the concerns come in.” And that aspect of the analogy, I think, tends to get overlooked.
Amy: That is such an excellent point. And you are exactly right. I really appreciate how you said that because I do think that big tech bros have used a lot of data in very exploitative ways. And how data got cleaned up was exploitation of workers. Especially, I know you focus a lot on privacy, and that’s definitely a concern when we can’t easily opt out or when we can’t see where our information is going or what’s being done with it. I understand that there is so much about us floating around in cyberspace right now. Like everything from our buying habits to our height and our eye color and how many kids we have or how many, or who our partner is or whatever. But there’s a difference between letting us know that they’re using it and just taking it and profiting from it. You’re exactly right. It’s the exploitation that’s the problem.
Karen: You mentioned privacy and yeah, I do talk quite a lot about privacy! I’m curious about how you’re handling that with your product when, if you’re using like Claude and ChatGPT and doing these interviews, obviously there’s a confidentiality aspect to that. And there’s also, as you said, just the personal privacy aspects of when you’re using the tool for other purposes. How have you been navigating that?
Amy: Confidentiality in law is our highest duty. It is uncompromising. You have to be so careful with all of your client’s data and information. And in order to make this work, I knew I had to have agreements with both OpenAI and Anthropic that they would not use any of our prompts or any of our information. They wouldn’t be storing it. They wouldn’t be using it in any way. So we have those agreements.
And then with my company [Lexara] personally, we do not see a thing that anyone puts into a firm’s intake interview. Each client gets their own identifier, or each law firm gets their own identifier, and then that’s what we can see of them. We do not see any data that is put in. I don’t process it. I can’t even look and see if they’re using it or not. I don’t even want it to be usage-based, because I want them to get as much benefit as they can from it. And I don’t want the responsibility of babysitting any data that I don’t have to babysit because it’s too big of a risk.
I will sign agreements with law firms if they ask about privacy. That information’s not going anywhere from the very beginning. Part of the reason that I wanted to train myself about AI was so that I understood privacy and privacy concerns. And from the very foundations of it, the servers that we’re using, the authentication that we’re using, all sorts of things, every single choice I made was based on privacy.
Karen: Yeah, that absolutely makes total sense. It’s great that you built it in from the start. You know, one thing that they say with startups and new companies is that the values of the founders tend to imprint the products and the companies from the beginning. And so the fact that you’re starting from a perspective where those are your values, that’s a good basis to build on.
Amy: Thank you. Too many years practicing law!
Karen: Yeah. So we’ve talked a lot about the different ways that you have used different AI tools, mostly generative. Are there any situations where you avoid using AI tools? And if so, can you share an example of when, and then why you chose not to use it?
Amy: There are not very many examples. I do feel like there are some things. One of my software engineers said, “You know, Amy, you should only use AI where traditional methods can’t accomplish the same goal.” We ended up reworking parts of our intake tool, and it was a long rework of our conflicts checking. It really delayed launching. Because I saw his point. Let’s not use AI where we don’t have to use AI. That’s a business application.
For example, I know people do fun things like take a picture of their fridge and then ask ChatGPT to tell them what meal to make. I don’t use it for things like that. I don’t feel like there’s enough value for that. I am pretty intentional about my AI use and even though it’s broad, it’s not all encompassing. If I can do it faster or easier or better myself, I’m not going to get AI involved.
Karen: That sounds like you got some pretty wise advice from your software engineer.
Amy: It was a little bit tough to take. We fought over it for a while, and I’ve talked to a number of technology experts, and I came to see the light.
Karen: I think one thing that we’ve all learned from years of working, and now it’s even more true, that we have to keep learning and keep an open mind, and being willing and able to change our minds when it makes sense and to adapt. Because things are not going to stop changing anytime soon.
Amy: You know what? That is perfect advice. I feel like that’s what society needs right now. We could bottle up what you just said and share it with everyone because that’s exactly what we need right now, is being open to learn, and smart about the choices we’re making. So, however you said that, good thing you recorded it, you should write a book focused on that. I guess you kind of did!
Karen: Yeah, I kind of did! But I appreciate the compliment and maybe we’ll pull that out, restack it.
Amy: You should!
Karen: But mostly I want to restack what you’ve said, not so much what I’ve said!
I want to talk a little bit about all of these different AI and machine learning systems. They all run on data. And you mentioned earlier that some of them have gotten their data in exploitative ways and use data that they didn’t have permission to use. I’m wondering how you feel about companies that use people’s data and works for training their systems and tools. There’s a concept called the 3Cs from CIPRI, which is the idea that people who create should have the right to consent to whether their works are used, to be credited when they’re used, and to be compensated. And I’m wondering what your thoughts are about that with respect to different AI tools. You mentioned using ChatGPT and Claude, for instance.
Amy: That is something that I’ve spent so much time thinking about and studying and kind of stewing over or contemplating. Because our AI tools are the very best when they are trained on the largest amount of data, especially dynamic data like conversations and work meetings. So I see so much value, because I love AI tools. I want them to be the best that they can be. But at the same time, I feel like all of the people whose information was copyrighted, yeah, they should have had a choice, or they should have been compensated before that was used. I am seeing now some collective licensing, which I think is a great step in the right direction where people are getting compensated, maybe not enough. I feel confident there will be some people who are probably never compensated enough for how they contributed to the AI tools that we have now. And I don’t know exactly how to solve that.
The bigger companies knew what they were doing when they used data, and it’s one thing to just scrape open data that’s sitting out there on the internet that someone just put out there, you know, for the world to see. But when you’re taking someone’s creative work, that is exactly what all of our IP laws are against. We’re trying to protect people’s right to benefit from their creativity.
Karen: It sounds like you feel like people are entitled to the 3C’s, but reality isn’t matching up to that.
Amy: That’s well-stated. That’s a perfect summary of how I feel. It’s not matching up. People are entitled. I appreciate all of the works that have gone into training our AI models and I wish there was a way to compensate people better.
Karen: As someone who has used, for instance ChatGPT and Claude and some of the others, do you feel like they have been transparent with people about sharing where they got the data that they used and whether the original creators did consent to its use?
Amy: Oh, definitely not. You talked about personalities of the creators coming out in their AI tools. I wrote an article about the personalities of the creators of the large foundational models or frontier models about how their personalities come out in their AI tools. And I was really encouraged by Anthropic at first. I felt like they were probably the most privacy-interested and concerned, even though they weren’t perfect. I’m amazed that at this point we are still running into AI tools that are really hiding how they’re using our data and what they’re doing with it.
I’ve talked to people who have said, “Yeah, you can’t trust a single thing with AI.” And I don’t feel like that. I’m not that much of a pessimist, but I do feel like we, as a country in the US, we have failed. We have been way too lax on what has to be disclosed. I have tried to get out of training on certain tools and it’s been like a scavenger hunt. It’s been like, “Okay, now go here. Now you have to do that. And we’re still taking it this way because you didn’t say no to this.” And that’s fairly deceptive. They’re making it as hard as possible to opt out. And I look at other countries who have made it a lot easier to opt out, and it feels discouraging that we haven’t been able to come up with some good privacy laws for AI and the way that LLMs use our personal information.
Karen: You probably heard about the fuss with LinkedIn and the way that they automatically opted all of us into use of everything that we had put into the site up to that point. That was not cool. And then of course, Meta – I don’t know if you use any of their social media tools, but they were pretty flagrant about saying, “Yeah, if you’re not under GDPR, we don’t care. We’re going to use your stuff anyway.”
Amy: You’re not going to be able to opt out. Those are two of the examples that I was thinking of when we were talking just a minute ago. LinkedIn when they switched it and didn’t announce it, and then we had to see from LinkedIn posts or other people, whatever, “Hey, go set your privacy because they’re using all of your data and information.” And I don’t use a lot of Meta AI tools, but I am very disappointed in Meta’s direction in just saying, “Yeah, you don’t have GDPR. We don’t care.”
Karen: Yeah. I know a lot of parents have had challenges with that. They want to use social media to share photos of their kids with family and with friends. And they feel constrained by it because there’s so much misuse, especially with kids’ photos and such nowadays, that some people have just gone off the social media sites. And some people put little smiley faces or something over the kids’ faces, but that just makes the pictures look kind of odd. I know it’s a challenge a lot of parents have dealt with.
Amy: I’ve been careful. I haven’t used social media very much for a lot of years. The one that I do use is LinkedIn. And I guess Substack now is more of a social media form than it used to be, but I am very careful about that too, especially where kids are concerned. My grandson, I won’t put his real face online because that’s not my choice to do. But you’re right, a little sticker on their face, that’s not the same, that’s not sharing a photo. It’s totally different. I wish that we didn’t have that situation. And it’s not even so much that I worry that a kid that looks exactly like one of my children or like my grandson, I’m not so much worried that their exact image will come back. I just feel opposed to how people are using it. Kids should have the highest protections. They don’t have a choice about it. And just out of respect for children, I don’t want to do anything that would help promote misuse of child photos.
Karen: Yep, absolutely. I’m with you there. Yeah. And likewise, I also have grandkids. They don’t have a choice. And they have to live with choices people make about them and for them now. And once it’s out there, it’s never going to be undone.
Amy: The funny thing is, I have one good friend who’s a lawyer who thinks he’s off the grid. He’s never had social media. He doesn’t post or chat or whatever. But you can Google him and you can find all sorts of things out about him. So there is a lot of data floating around cyberspace, about all of us. It’s not like we can avoid it. But I wish that we had some better laws in the United States.
Karen: Yeah, a lot of these data brokers, especially in the US, they have scraped up information from us, bought it, and that information has leaked out in a lot of places. Like your friend who thinks he’s been off the grid, there’s a lot of information about us, personal data.
Amy: Oh, yeah.
Karen: Do you know of any cases where personal information has been used or by an AI based tool or system?
Amy: I know that my information specifically has appeared in data leaks. I know for sure that my information is wrapped up somewhere in some LLMs. But I can’t think of a specific example. I guess the closest one that I could think of was one of my employees was using Otter AI and I had to stop him from using that in our meetings. I’m not comfortable with the way that they use data and don’t disclose it. And not just the language, which I probably would’ve been okay with in a lot of contexts, but images and voice prints. And we’re getting into security issues. That’s really unfortunate that they feel comfortable using that without even asking or telling.
Karen: There’s always a trade-off with these tools. Yeah, this is convenient. It will summarize my meetings for me. But on the flip side, then they have all the information about you and about your meetings and what was said in the meetings, and as you mentioned, images as well.
Do you know of any company that you gave your content to, or that you used, that made you aware that they might use your information for training at all?
Amy: Nope. No. I think you’ve made the point a number of times before about how that information is buried somewhere in the terms of service or their privacy policy. Everything from the grocery store where you use the card to get a discount where they’re tracking your purchases. I think that I really fully felt the consequences of that when I had a miscarriage, and I was really excited for the baby. This was a number of years ago. After I had had the miscarriage, I kept getting these coupons for diapers and for everything else. And my next youngest kid was way too old for diapers or anything, and I thought, “Of course they’ve realized that by now I should have had this baby. And they are using that data.” So they’ve definitely been using it for a lot of years. That was 10 years ago. It’s astounding how long, how much data has gone into these LLMs that we’re not aware of.
Karen: Miscarriage is so common. I want to say it’s one in four or something like that, which seems awfully high, but I think some of them may be very early. But that must have been painful to have that reminder.
Amy: It was not my favorite.
Karen: Oh, that really sucks. But yeah, a lot of the exploitation of data that we see really predates generative AI coming onto the scene. But it meant that there was this large base of data out there that it was very easy for them to buy, or to set up a data sharing agreement with, and that is part of where these tools have gotten their data. And in many cases, it’s out of date, or just plain wrong, and we don’t really have any way to get it corrected. Or if we get it corrected at the source, we have no way to know where else it’s gone. And so it’ll never get corrected in all the places that that data flowed out to.
Amy: Exactly like you said. It’s so old. It’s been mushed around several times. It’s gone around several times through the cycle and yeah, in ways that we have no clue. Another thing, when people think they’ve privatized something because they’ve anonymized the data, but they don’t realize how easy it is to reconstruct that data. And so they thought they were being private. The thing that’s coming to my mind the most is healthcare data where they were sending it into a central location to share and to study, and they thought that it was anonymized. But so many times that’s so easy to reconstitute. If they’ve just taken out names, that’s easy to put back together. So it’s disheartening.
Karen: There was a study recently about de-anonymizing data and that it is surprisingly easy to do, and so a lot of people aren’t protected. You mentioned medical data and that’s obviously a very big concern, especially with some of the genomic data. Companies like 23 and Me, and then what do they do with that? Or if you take a medical test where you’re providing blood samples, stool samples, anything like that. And what do they do with it besides just perform the test that you’re paying them to do?
Amy: Exactly like you said, we’re paying to give them our information. And, thankfully we have HIPAA, but HIPAA hasn’t protected us like it should have when, like you said, we’re paying for this test. And the company thinks it’s anonymized our data. But I wrote an article about that a while ago too, about how it’s easy to reconstitute that data. I’ll have to look. When was that study from?
Karen: Oh, I remember reporting on it within the past few months. I’m sure I can find it again. It was just showing how easy it was to de-anonymize data. Even just looking at, for instance, someone’s driving patterns, data from our cars, GPS or any of the information. You don’t have to know somebody’s name if you see their path of where they drive every day – where do they work, where do they drop off their kids, or anything like that – you can pretty easily figure out who that person is. There’s just so many ways, and so many different sources and types of data, that it’s not as hard as people might assume. They might think they’re safe, but really, they’re not.
Amy: No, it’s not hard to put it back together. I’ll look for that study.
Karen: Yeah, I’ll look for the link for you.
[Reference: “Large-scale online deanonymization with LLMs”, Simon Lermen, Daniel Paleka, Joshua Swanson, Michael Aerni, Nicholas Carlini, Florian Tramèr, 2026-02-18, last updated 2026-02-25, https://arxiv.org/abs/2602.16800]
Last question: With everything that’s going on, we see that public distrust of these companies has been growing. And that’s probably healthy, that we now know what they’re doing with our data. But if there was one thing that these companies could do to earn and then to keep your trust, what would that one thing be? Or is it even possible? Maybe that’s another question. I’ve had a few people say they didn’t think it was possible.
Amy: I think that if they would be a lot more transparent about how they’re using data and ask for permission and allow us to opt out. I know that’s actually three things, but that would go such a long way toward making me feel comfortable and confident with what they’re doing with AI and what they’re doing with data, if that makes sense.
Karen: Transparency is a big thing. There’s so many other aspects that we could look at, but if they aren’t transparent about what they’re doing, it’s hard to know if they’re doing the other things right. To me, that’s really a key enabler for any of the other aspects that would need to go into trust. So without transparency, I don’t think trust is possible.
Amy: I completely agree. That’s exactly what I was going to say. It’s inherently a little sly or a little deceptive. If you’re not telling us how you’re using it, you may be using it in the best way possible, but if you are not telling us how you’re using it, that’s inherently untrustworthy. Inherently deceptive. Absolutely.
Karen: Yeah. Well, thank you so much for sharing your thoughts on AI in this interview! I really appreciate you making the time. Is there anything else that you would like to share with our audience?
Amy: So I’m excited about my AI tool, Lexara Engage, and I am incredibly excited about my next AI tool that’s coming out soon. That is going to be an absolute game changer for bigger law firms and doing conflicts checks. So I’ll be excited to share more about that when it’s complete.
I just want to say that it’s been a pleasure to talk to you. I love talking about AI, and you know so much about AI and privacy, and so it’s been wonderful. I really appreciate you taking the time to talk to me and being always so generous with your time, and your forums you share. Oh, Karen, I’m not saying this right, but I just, I have so much respect for people who help other people, and I feel like you are that person. You’re very generous with helping people around you, and you certainly don’t always find that in law, but you don’t find that everywhere. So thank you so much, again, for your time.
Karen: You’re welcome. Thank you, Amy.
Amy: Thanks, Karen. Talk to you later.
Interview References and Links
Lexara website and intake tool for lawyers
Amy Swaner, JD | AIGP on LinkedIn
Amy Swaner on Substack (AI For Lawyers)
About this interview series and newsletter
This post is part of our AI6P interview series on “AI, Software, and Wetware”. It showcases how real people around the world are using their wetware (brains and human intelligence) with AI-based software tools, or are being affected by AI.
And we’re all being affected by AI nowadays in our daily lives, perhaps more than we realize. For some examples, see post “But I Don’t Use AI”:
We want to hear from a diverse pool of people worldwide in a variety of roles. (No technical experience with AI is required.) If you’re interested in being a featured interview guest, anonymous or with credit, please check out our guest FAQ and get in touch!
6 'P's in AI Pods (AI6P) is a 100% human-authored, 100% reader-supported publication. (No ads, no affiliate links, no paywalls on new posts). All new posts are FREE to read and listen to. To automatically receive new AI6P posts and support our work, consider becoming a subscriber:
Series Credits and References
Disclaimer: This content is for informational purposes only and does not and should not be considered professional advice. Information is believed to be current at the time of publication but may become outdated. Please verify details before relying on it. All works, downloads, and services provided through 6 'P's in AI Pods (AI6P) publication are subject to the Publisher Terms available here. By using this content you agree to the Publisher Terms.
Audio Sound Effect from Pixabay
Microphone photo by Michal Czyz on Unsplash (contact Michal Czyz on LinkedIn)
Credit to CIPRI (Cultural Intellectual Property Rights Initiative®) for their “3Cs' Rule: Consent. Credit. Compensation©.”
Credit to Beth Spencer for the “Created With Human Intelligence” badge we use to reflect our commitment that content in these interviews will be human-created:
If you enjoyed this interview, my guest and I would love to have your support via a heart, share, restack, or Note! (One-time tips or voluntary donations via paid subscription are always welcome and appreciated, too 😊)
