Introduction - Luisa Herrmann
🫛 This edition of AI, Software, and Wetware features an audio interview on ethical AI with Luisa Herrmann, a 🇺🇸 USA-based 🇧🇷 Brazilian fractional CPO & tech strategist. Luisa is the author of Practical AI by Luisa Herrmann on Substack. We discuss:
Regrets on majoring in chemical engineering instead of computer science
How she got started in writing software and why vibe coding “isn’t for her”
Seeing LLMs as a non-deterministic software abstraction layer and why non-determinism matters
Using chatbots as a translation layer for writing and reading with ADHD and mild autism
Building technological literacy with her son on his science fair project for data analytics on wait times at Universal Studios
Designing Five Lanes to add a knowledge layer over information from different sources
How she fights back against data scraping to protect her date of birth, email address, house photo, pictures of her child, and more.
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 - Luisa Herrmann
Karen: I’m delighted to welcome Luisa Herrmann from the USA as my guest today on “AI, Software, and Wetware”. Luisa, thank you so much for joining me on this interview! Please tell us about yourself, who you are, and what you do.
Luisa: Sure. Thank you so much for having me. A little bit about me. I was born and raised in Brazil. I moved to the US for college. I graduated with a degree in chemical engineering, completely different from what I do now. But that was essentially when I fell in love with process design and process optimization. So my career essentially started with me trying to do process optimization at a manufacturing plant, and then went through different kinds of processes and different kinds of design and optimization.
Then I went to school in Philadelphia, eventually moved to Cleveland, Ohio to work in manufacturing, and then moved to Boston, where I live now. And it’s kind of hard to live in Boston and not work in software. But I’ve spent the majority of my career working in software through the changes that we’ve seen with the explosion of software and now AI, LLMs, artificial intelligence, all of that.
Because of my passion for process and engineering process design, process optimization, I’ve always been heavy on the data side. I love playing with data, optimizing based on data, collecting data, looking at data. And so I was at the right place at the right time for when all of this stuff started happening. Now I create products to simplify our lives, so take a whole bunch of data and then help people make sense of that data for their individual lives or for work. And I also work as a consultant for people in especially regulated industries who are trying to figure out how does this stuff work, how to use it safely and responsibly, and in a way that benefits people more than, you know, any potential drawbacks.
Karen: That sounds great. So tell us a little bit about your level of experience with AI and machine learning and analytics. Have you studied the technologies, or have you picked them up on your own? How have you learned what you’ve learned?
Luisa: One of my regrets was actually that I did major in chemical engineering. I hope my parents don’t listen to this podcast. Who am I kidding? I’m going to send it to them. But they know this already. I actually felt like I should have gone into computer science, because essentially all of the optimization that I was doing would’ve been helped by computers and algorithms, as it turns out. And I would’ve learned them a lot earlier if I had studied computer science. But I was actually scared because I felt like the people who were majoring in computer science had been doing coding before college, whereas people majoring in chemical engineering probably weren’t doing reactor design before college. Maybe they were. I’m not judging. But the level of experience that I had at that point was, like, coding my HTML website on GeoCities. So I didn’t think I could really compete with people who had built apps. Of course, that’s not true. A lot of people were coming in and just starting. What I realized once I actually did start coding for myself is that you pick this stuff up pretty quick. And there’s always more to learn, but the timeline to start creating things is way shorter than people think.
But because of this existence of data, my role was always in these startups. I ended up working software startups. Because I could both understand the market and what the clients, the customers, were asking for, but also sort of understand the technology, I ended up being that person sitting between, telling the engineers, “This is what you need to develop,” and telling the people who are not technical, “This is what’s possible. This is what the technology can deliver realistically.” From there it was sort of a small, and then large, slide into: I’m trying to figure out if these engineers are telling the truth about the timelines and about what they can and can’t build.
And then I had a problem that I wanted to solve, which is too many sources of data all at the same time all screaming at me. I can’t tell what’s important and what’s not important. And I was spending so much of my time going in between different tools to try to find the information that I needed, that I taught myself how to code, aided by GitHub Copilot. I knew very, very basic Python, and I extrapolated that into, like, “How can I write a function that does this, this, and this?” So I can tell you from personal experience that it’s not impossible. It’s not daunting. Once you get into it, it’s a lot more doable and achievable. But from there, I started just creating my own products, and that was the fast slide into developing tech.
Karen: So you’ve been using AI coding assistance basically to help you learn to code and learn to solve the problems that you want to solve with code?
Luisa: Yeah, so when I started was way before Claude Code, even before Cursor, which is now my preferred coding assistance, let’s call it. It was very much a learning tool, more than “I’m just going to let the AI write it.” I was coming up with, I want the code to do these things in this order, and having the AI tell me, “Okay, this is how you write the function,” right? But I still had to keep track of the inputs and outputs for each file and then the routes and the processing, the services, the utilities, all of that. I had to keep track of what was coming in and out, and the formatting, which is honestly still the hardest. My current project is coming up with a better way to keep track of that, because I think there has to be a better way. I was actually writing these out by hand at one point. It’s prehistoric.
But you know, with GitHub Copilot, it was one file at a time. Cursor was all of your code base at a time. And so I was using those tools to really augment what I could do. And then I tried vibe coding, and it’s not for me, okay? But I think it has made a huge difference on making code more accessible to people, and I think that’s awesome.
Karen: Okay, can you say a little bit about why you feel it wasn’t for you?
Luisa: Vibe coding is incredible because it allows people who don’t speak code, essentially – because code is, you write, for you to talk to the machine, right? You just have to write it in a way that the machine is designed to understand. And what vibe coding things like Lovable or V0 allow people to do is essentially create something like that without needing to speak the language, which is fine. But that means that it sort of makes a lot of assumptions and skips a lot of steps that you would have to go through if you were the one actually writing it. And if you want to continue that analogy with language, you know, if you wanted to add any kind of nuance or personality, that’s not going to happen with these tools, which is fine. And that’s the thing that I say over and over again for these technologies. Like, it’s fine. You just have to understand what they’re good at and what they’re not good at. And then you use them for what they’re good at, being aware of what they’re not good at. And so you remedy for that, and you focus on this.
The joke that I tell people is that if you try vibe coding, this is the experiment. You do one thing. And then you tell the code that you just want to change something – not add, change. It always adds lines. It always adds more lines than it removes. And so whenever it makes changes, it never goes back and cleans up what doesn’t work anymore. It just leaves it there. So if you just keep going with this, after a while, you have spaghetti code. You don’t know if it leads anywhere or why it’s there, but then you remove it, and then something breaks because you didn’t understand it. I’ve seen this with my own code because at one point I was like, “I’m going to see how far I can take this.” Regrets. I have a lot. Because I actually liked the app that it ended up creating that I had to then tear down and rebuild from scratch.
But again, if you understand what it’s good at, it’s an amazing tool. I knew very basic Python, and without Copilot and Cursor, I probably would’ve just given up. I had a full-time job, I have a child, I have hobbies, and I wasn’t going to have time to dive into that very deeply, but that allowed me to do it anyways, so that was awesome.
Karen: That’s an interesting story. I started out with assembly code, you know, microprocessors. That was my first exposure to computers when I was a freshman in college. And then when we moved to higher level languages, people talked about how much got lost, and you lose a sense of control. And we’ve moved to higher and higher-level languages. And in some ways, some of these coding assistant tools act like a higher-level language, and you do lose that control. You gain ease in return, but there’s some loss of understanding, some loss of control that comes with that. And that’s just one of the trade-offs that people make, and maybe don’t always realize they’re making, but that is a trade-off that they’re making.
Luisa: And the interesting thing about that is that you’re absolutely right in that we’ve been adding abstraction layers to code by building higher and higher-level languages. But what I think this introduced was an abstraction layer that is non-deterministic. And that’s the part that scares me, is that up until now, at least, when you wrote the code and you used the abstraction layers, if you were running the code and then it stopped working, you could go and figure out why that was. Now, if you tell your Claude Code to do something, and then you go back and try to say, “Well, why did you do it that way?” It’s going to change its answer. It’s going to tell, “Oh, you’re right, that was the wrong thing to do.” Okay, so do I have to every time go like, “Why did you do that? Why did you do it that way?” Like, “Why? Why did we go that route?” And so I feel this enormous anxiety when I let it ... That’s why I don’t let it make decisions because it can’t explain. There’s no audit trail for the decision made, right? And so it’s like an abstraction layer. A non-deterministic abstraction layer is just insane to me.
Karen: Yeah, especially coming from a process optimization and control background, right?
Luisa: You can’t improve what you can’t measure.
Karen: Yes, exactly. So it sounds like you’ve used AI tools quite a lot professionally. How about in your personal life? Do you have any use for them, or do you avoid them, or what’s your experience been?
Luisa: So I’ve been using them for a while in my personal life. I don’t use the chatbots like the usual way that people use them. This is my office, and I have two servers just over there, because I like doing my processing in controlled environments, as opposed to just sending data out. And so there’s smaller models that do deterministic adjacent processing, still technically NLP, so natural language processing.
But the biggest boon for me is that I am neurodivergent. I have ADHD and mild autism, which made it really difficult for me to write emails and read emails. So the way that I communicate and have communicated my entire life is very direct. I think in bullet points on, like, the data that I’m trying to communicate. And in person, I’m obviously not like that. I’m not talking in bullet points right now. But for some reason, communicating in writing was very difficult for me. And so what I started to do was just that I would write my emails in bullet points and have a model basically write it out. I would say, like, “Make this human and write it out as an email.” And then I would have it do the other way when someone wrote me an email, “Just give me bullet points”. And so it sort of became a translation layer for my brain. And that’s sort of where I still use it. If I want to communicate this, what is a way that is not dry bullet points that I can communicate it? Sort of, know your audience and know how they want to be communicated to.
Karen: That sounds good. And you mentioned that you have kids. Are your children getting exposed to AI in schools or using it? Or what’s that experience been?
Luisa: So my son is eight, so he’s not getting a lot of exposure to AI. They do have iPads, which surprised me. And, because you know that autism is hereditary, he came home one day and he said, “Mom, they told me we’re going to use iPads in school now, and I told them I’m not old enough to use an iPad.”
And I was like, “Excellent, I know you’re my child for sure.” Like, we had rules, and the rules are being violated, and I must say something. So he is not getting exposed to AI. He’s getting exposed to technology, but I am making sure to explain to him how these things work essentially.
He has a science fair tomorrow, and the science fair project is a data analytics project, because of course it is! But it’s because for spring break we went to Universal Studios in Orlando, and he started asking me questions about how they estimate wait times for rides. And so we started collecting data about it, about estimated versus real wait times for rides. And then we’re trying to figure out... You know, we went and we researched together, because he was the one asking the questions. How do they measure it? How do they figure it out? And then the whole science project is around that, which I’m really hoping, and I’m going to keep pushing for this, just leads to him understanding how these things work.
Because I think the technological literacy that we need everywhere – not just with children or teenagers or young adults; with everyone – is to understand, again, what these technologies are good at, how they work, what their potential drawbacks are. Because it’s going to be very difficult. Every time someone tells me, “Well, ChatGPT told me,” I’m like, “Nope. Mm-mm. No.” Even my father said ... I asked him, “Explain to me how ChatGPT works.” And he says, “Well, I ask a question and it looks it up in, like, an encyclopedia or something.” I’m like, “No, that’s not how that works.” It’s a statistical model, which means it looks at the things you sent it. It thinks of it as numbers. It tries to connect it with other things that look like it to sort of understand what you had said, and then it sort of finds in its training where that has, it has seen that before, and then it gives you the answer based on that. It’s not looking it up in an encyclopedia. It’s not a person sitting there going like, “Hmm, I wonder.”
The other example I gave about this was, there are these videos going viral on TikTok right now. It’s this guy who’s talking to ChatGPT and asking it, you know, “Time me” or “Tell me how high I jumped.” Which is funny in a vacuum. But somebody apparently asked Sam Altman about that, and his response was, “It’s going to be a few years until ChatGPT can serve as a timer.” And I’m thinking, “Stop. No. No.” ChatGPT is not a timer. You know what IS a timer? A timer. We have it. We don’t need it to be everything. Like, it can’t be a timer because it’s a statistical language model which doesn’t understand time. This was a great opportunity for Sam Altman to say, “Let’s talk about how these models work and why it’s not a timer,” instead of saying, “Oh, it’s going to be a few years.” It’s not a timer, you know? This is exactly the kind of thing that we need to get more familiar with, essentially.
Karen: Yeah. But that’s the kind of hype that drives up the IPO prices, right?
Luisa: I know, and I know that that’s why he’s a billionaire and I’m not. It’s because I could never do the hype thing. I could never say, you know, “This is going to revolutionize the way you think.” Like, no, just read a book. That’s going to revolutionize the way you think.
Karen: Yes. Books are awesome for that. I’m so impressed that your eight-year-old son started asking questions. I’m having flashbacks to queuing theory courses that I took in college. Eight years old. It’s an impressive leap that he’s looking into that. That’s pretty cool.
Luisa: Yeah, he’s a cool kid. Every now and then he catches me completely off guard with his questions about how things work. My approach is to always answer and always be thorough. I hear a lot of people say, “I don’t know, bud.” Like, “We’ll look it up later. I’m not sure.” And I’m just like, “Okay, so historically...” It’s going to be great while he indulges me. At some point he’s going to become a preteen and be like, “Mom, please stop, please.”
Karen: Yeah.
Luisa: Or he’s going to start asking ChatGPT things, and then I lost, so.
Karen: Right. Well, you’ve already kind of covered a specific story on how you were using the coding assistants. Can you say a little bit about the application or what problem you were trying to solve by writing code for that?
Luisa: Yeah. So the first one was really that I had six or seven email accounts. Each one has an associated calendar and, you know, there’s documents and people refer to documents, and sometimes they’re attached to the email, sometimes it’s a link to a Google Drive, sometimes it’s OneDrive, sometimes it’s Dropbox. I don’t know. And the tipping point was that somebody said, “I sent this to you on Slack. You should have seen it.” And I said, “I have 26 unread channels on Slack. About the same amount of direct messages. Let’s not get started on my email. I cannot possibly have read everything that people have sent me.” But I also realized that just saying that was not helpful, and that people prefer to receive information in different ways and to communicate in different ways. Some people prefer Slack, some people prefer email. Some people do phone calls or text messages. Some people use, like, Asana or Notion or Jira, right? People communicate in different ways.
So I wanted to create a tool where people were still using their preferred ways of communication, but they were able to access information in the way that they wanted to access it. Essentially, it’s the concept of a knowledge layer, so pulling in information from different sources, and then having that information so that you can access it however it is that you want, but you don’t have to go chase it, essentially.
So that’s what I built for myself. I called it AI Fred because it looks like Alfred, and my son was really into Lego Batman at the time. And I was just like, “This is like my butler.” I just ask it, like, “Where is it?” And it tells me. And I can go and find that information, and no one ever gets to say, you know, “Well, I sent this to you, but you didn’t see it,” because it’s surfacing these things if I say that they’re important.
But I’m type A, so I didn’t want something like a superhuman which makes choices for you on, you know, which emails to highlight and which to not show you. I need to program the algorithm on what’s important to me, as opposed to having it try to guess. So that was my first product, Alfred.
Karen: Very cool. Do you think that’s something that other people might be able to use? Would you consider sharing it or open sourcing it or turning it into a product?
Luisa: I went on a journey with that one. A lot of people... All of it was deterministic. There were small deterministic models. Like, there wasn’t a lot of decision-making happening. And then a few things happened. When I would show it to people, they would start thinking about integrating it with their LLMs. Cause, you know, then I can ask it to summarize things for me or to write emails for me or to give me, you know ... Fair. That’s a use for it. And also things like OpenClaw and or the Claude team or whatever. Essentially, there are other things doing similar things. The latest iteration of it is something that I have been working on with the goal of open sourcing by, like, the end of this month, so.
Karen: Oh, very cool.
Luisa: Yes. That one will be called Five Lanes. Because I name things poorly, so.
Karen: Five Lanes, okay, yeah. And when you were talking about having all this information and people contacting you in different ways, I think we’ve probably all run into that, and especially if we have more than a single immersive job that we deal with and that is our entire life.
I mean, even with, say, on Substack and the She Writes AI community, and we’re writing these books together, and some people want to be in the Google chat, and some people want email, and some people only really keep track of their Substack direct messages. The idea of having a single portal that would let me see, “Okay, here’s all the messages I’ve gotten, all the links I got from all these people.” Even without some intelligence trying to decide what I want to see, I think that would be really useful. I could definitely see value in having it pull everything. Because every one of these tools wants to be the place where you live, but that’s not reality. And so we’re all juggling ... I have multiple email accounts. You said multiple calendars. There are tools that try to solve that, but you have to put all your eggs in that basket.
Luisa: Yeah, and that’s by design, right? They’re trying to capture a market. They’re trying to have a captive audience that is sticky. And I’ve worked in this for so long that I can tell you the conversations usually go something like, “You know, we could just connect to this other thing.” But if you are just the connector, if you’re the middleware, then you lose mindshare in the customer’s mind. So when it’s renewal time, they’re like, “Oh, yeah, no, I haven’t thought about that tool in so long.” So everyone wants to be top of mind. They want to be the tool that you go to, so that you’re stickier, so that you’re providing value. But also, they don’t want to connect with other tools because they don’t want to become just a pass-through entity. So that is not necessarily a decision that’s made based on what’s best for the people who are trying to get their information, but it’s purely a strategic decision.
Karen: Well, I’m going to have to keep track of Five Lanes when it comes out. So actually it may be out by the time this interview publishes, so we’ll have to include a link to it in the interview.
Luisa: Yeah, that’d be great. I created the GitHub organization yesterday, so...
Karen: I am definitely going to have to follow that. And maybe I can make some good use of it for myself, because it’s a problem I think a lot of people are experiencing with all the different channels of communication.
Luisa: Yeah, and for me it’s Five Lanes because I at one point realized that there were five lanes of communication that I was trying to keep straight. Whether it was, you know, home, work that I was doing, consulting, personal brand. I just wrote it out, I was like, “Okay, I’m going to keep track of these five lanes.” And that’s where the idea came from.
Karen: Very cool. Yeah. So you’ve made a lot of good use of different AI-based and non-AI-based tools. So are there any things that you avoid using AI for, or that you would say maybe that you would never use AI for? Thinking about writing your own personal AI policy, which is something I wrote about in my book on Everyday Ethical AI. Daria Cupareanu shared this nice idea of three buckets. One bucket is things that are sacred to me, where humanness is the point, where I will never, ever use AI for that. And then there are things that, this is just routine, it’s a burden, I don’t care. The more I can automate that, the better. It doesn’t have to be AI necessarily, but certainly automation. And then there are things in the middle where, yeah, I could use some help here, but I still want to be in control, or the augmented zone. And so I’m wondering, is there anything that you would say is in your sacred zone, things that you would not or do not ever use AI tools for?
Luisa: I have quite a few, for different reasons actually. So I’m not an artistic person, but I don’t use AI for art ever. The logos that I’ve created for my companies – if I don’t have the money to have someone design it, I just do it on Procreate on iPad, best I can do. It’s simplistic, lots of lines. The one that I did for my company, AI Nova, I actually sketched over coffee with a friend when I wasn’t even paying attention. It was like half of a brain and half of a node, like a neural node network.
I never use it for advice on anything personal because I have to remind myself that this is a statistical model and so it’s programmed to agree with you and to tell you what you want to hear and find justification for it post-hoc. I’m not going to fall for that. The hallucinations, the AI delusion is I think a concern there. So I’m never going to use it for that.
Or anything that requires sending personal data of any kind, because that’s data that you’re giving away. Anything that you use these large models for, you’re helping train them further with your own data. So anything that I wouldn’t want to gift to a large corporation for them to use to their benefit, I don’t use these models for. So there’s not a lot.
The larger models, every now and then I’ll think of an idea for an article or something and I’ll write a first draft that is my bullet point way of thinking and then have it just flesh it out, but then I rewrite it after that. There’s a tweet that I saw a long time ago that said, and I immediately resonated with it, which was “I didn’t know if I believed in the human soul until AI came out.” And now I do, because you can tell how soulless... like, I never understood when people said that, but now you start looking at AI art, and human art started having much deeper meaning for me, and that’s how I feel about writing as well.
I want the things that I put out into the world under my name to have my soul attached to them. So I draw the line at anything that I put out there that’s supposed to be coming from a special brain, mine. It has to come from my brain. And actually, I was having a conversation this morning about the fact that I think something is lost when people don’t have to write their own output and have multiple versions of it and, and go through, and sleep on it and come back and rewrite it and think through it.
Because honestly, I have this folder and it has loose-leaf in it. When I need to think about something, I just start writing. I just write pen on paper, and that’s how it helps me think. So the idea of something else doing the writing for me, and therefore the thinking, is sort of very dystopian.
Karen: Yeah. It sounds like the main reason that you would choose not to use it is the sense of the output not having the human soul that you feel should be going into it.
As you said, when you use one of these tools, you give it data and you’re helping to train it further. There are a lot of questions about where these tools even got the data that they use for training in the first place by scraping YouTube videos and other things that they claimed were, you know, publicly available, which is not the same as public domain because they are copyrighted. Books have been stolen, videos have been stolen, images stolen. Is that a factor in your decision-making? Like, if you knew of a tool that did get its data ethically, would that change your view? Or would you still say, “You know what? The output still doesn’t have the human soul in it, and so I’m not interested”?
Luisa: It would make a difference, but not for that. I would still not use it because the process of being ingested and regurgitated by a machine makes it not human, regardless of what the machine is doing. It’s really interesting that you bring this up because the way that the original data was collected to train these models is unethical at best, illegal at worst.
And there’s, you know, like I said, the post-hoc rationalization that now that the models are out and everyone’s using it, they’re like, “Well, if we hadn’t done the crimes, then people wouldn’t have these models that they like.” Which is, you know, I guess a response that you could have that they have chosen. Again, that’s why I’m not a billionaire. But I think having an understanding of how these things work and how they were assembled into, you know, sucked from the outside into these models to then create this value juggernaut that they have now makes me definitely uneasy using these tools, because I know that they are profiting handsomely from not their work.
And they’ll argue it’s their work because their work is the algorithms that scraped, processed, and now spit it back out. But one fun thing that I think is happening is that, I don’t know if you remember, but right when ChatGPT came out, they said that they had a cutoff date, that after that cutoff date there was no new information being put in. And at the time, we talked about how that was, because as soon as people found out that they were doing this, they would try to seed information into it. Just like people tried to hack Google AdWords and stuff, SEO, right? They’re essentially going to do SEO on steroids if the information was still being ingested. And so there was a cutoff date to essentially say like, “You can’t game it now. It’s too late. There’s a cutoff date.” Obviously, that’s not the case anymore. So people are trying to game it. And in that process, it’s an ouroboros where they’re feeding AI slop to create more AI slop. And at one point they will have to come up with better models that are able to at least differentiate on some level, otherwise it’s just going to become unusable. And some of them already are slowly becoming unusable for that reason.
So I think the future isn’t necessarily an ethical model, but it is models that understand what they’re supposed to do and are more purpose-built. And I hope that they’re also ethically built, you know, with consensually-sourced materials. You know, because right now, essentially it was like, “Well, there was no law against it.” I mean, debatable. I remember Napster. But People need to be asking more questions of how we’re gathering that data, who owns the data, who profits from the data. And I think that’s part of the future is having companies that have models where there’s ethical data sourcing and accountability.
Karen: Yeah, there’s a very small number of companies that have been certified as what they call Fairly Trained, meaning they got their data ethically. AI music was one of the first things I started writing about two years ago, and I found a handful of ethical companies and a very large number of unethical companies. But there’s one chat bot, for instance, built in Switzerland, ethically trained, run on renewable energy – everything you could think of that they could probably do to make it ethical. There’s, of course, people who talk about it being theft or “all AI is theft”, well, here’s a tool that isn’t based on theft. Is that still something that you see value in or not? And it raises some interesting questions.
You also mentioned about big models, and they’re saying, “Well, you know, we had to do this in order for you to have something that’s useful.” There’s a project called The Common Pile that you may have heard of, where they used just ethically sourced materials, and they got a model that gave just as good results as one of the LLMs that preceded it. Okay, so, that was a lie; you didn’t have to steal this data to provide something that would be useful. And that doesn’t get, I think, as much attention as it deserves.
But I want to go back to your comments about the companies that are profiting. They are making tons of money from having these tools, but they are not compensating anyone. Also, I don’t know if you’ve heard, you mentioned about it being legal, but there are well over 100 lawsuits nowadays on copyright infringement. A lot of them are in the music space, but some are also with the books, and there was the Anthropic settlement. So it’s not that there weren’t laws to say that this was not okay. The laws have been there. But it’s going to take probably decades for all these lawsuits to settle out.
Luisa: Yeah. And the thing with that is that there was no precedent for what they did, and so there had to be an argument. And that’s what the lawsuits are, to argue, “Does this apply in this case, yes or no?” And that’s what’s getting decided, but the technology moves so much faster than the law and the application of the law. That’s sort of what we’re seeing now is the, “Is it legal? Was it legal?” We’ll find out.
Karen: Yeah. I’m wondering what your thoughts are about companies that do take this data. Do you think that they should be required to get people’s consent, to give them credit, and to compensate them when their data is used? It’s what CIPRI calls the 3Cs of creative rights: consent, credit, and compensation. Do you think that AI tool providers should be required to give all of us the 3Cs, to have the right to consent and be credited and be compensated for use of our data?
Luisa: Yeah, I think part of that comes back to the education piece. If people were educated enough to give their consent to understand how those companies were getting value from their work so that they could be adequately compensated, that would go a long way. So far there hasn’t been an incentive for those companies to do that, both from a perspective of it would cost money, it would cost profits, but it would also slow things down significantly. And that goes against everything that Silicon Valley stands for with “move fast and break things.” It would be great.
I think the best thing that we as consumers can do is understand how our data’s being absorbed and used by these companies. Because then when we talk about requiring consent and being compensated, then that will be a process that says like, “This is the kind of consent that you need from me, and this is the kind of compensation that I want from you,” as opposed to just telling them that they have to figure out how to provide that, right? Which is sort of the issue that we also have with legislation, because the tech companies are telling legislators like, “Hey, you don’t have to understand this tech. Like, just write the legislation this way and it’ll be fine.” But if you don’t understand the ramifications of that, then it’s not fine, right? So we can’t have them writing the laws just because the legislators don’t understand it, essentially.
So it’s the same thing. It comes back to: we need to be educated on how our data is being used, extracted from every conversation that we have with chatbots, everything that we post online, every photo, every, every LinkedIn post, everything, right? Right now the reason why Instagram looks the way that it does as opposed to what it looked like when we started is because it’s just become a way to suck up data constantly about everything that you do all the time, and sell it. So I miss the internet when it wasn’t that.
Karen: Yeah. Those were the days. I’m with you there.
So with the different tools that you use, different tools and different systems, do you feel like the companies have been transparent at all about where they got the data? Or, on the other side, what they will do with your data if you use that tool?
Luisa: I don’t think it’s transparent at all, especially on where they got the data. A lot of them claim that that is, like, trade secrets, and they’ve been somewhat successful with that argument. And how they’re going to use that data, it goes back to the education that if you read privacy policies, if you read terms of service, they do sort of say some things, but they leave it as vague as possible as to how they’re going to use your data. And the thing is, those are not legally enforceable, right? They can change privacy policies. They can change terms of service.
And, you know, raise your hand if you’ve never gotten a notice of a breach, right? Because there was a privacy policy, and then your data leaked anyways. This is not very comforting for me. And so usually... I actually had this really funny interaction at a doctor’s office where they said, you know, “Do you consent to having this transcribed with AI?” And I said, “Well, what tool are you using? Let me take a look.” And I said, “I’m going to be really honest with you.” And they’re like, “No, no, no. They’re very private. They do things blah, blah, blah.” And I was like, “Okay, it doesn’t matter,” because the odds of me getting an email saying, “There was a breach and your data may have been exposed,” is probably 90%. Because the people who are building these systems are not fully aware of how they are supposed to work.
So it’s just a reality that There’s too much happening. Your data is getting taken, and we are slowly ... And that’s the other thing. I don’t want a sense of helplessness that I keep hearing from people. It’s like, “Well, they’re going to get my data anyways.” We shouldn’t give up. You know? Yeah. We should be fighting back. I mean, I never put my actual date of birth on any website. I very rarely use a real email address. I use masks, which send the email to my inbox but aren’t my email. I always reject cookies, always. I clear my cache all the time. I use Firefox Focus, which is a privacy browser. So things like that, I will never stop doing. My house is blurred on Google Maps. It’s little things that I want to educate more people on so that they will start doing it as well, because I think as individuals, once we start realizing what’s happening with our data and the power that we have, we can make a change, but that takes longer.
So my new goal is to try to help legislators close that gap, where, you know, they’re trying to make laws for things that they don’t understand, and people are trying to sell them things all the time.
Karen: Yeah, you mentioned the medical scribes, and that’s a really interesting area, I think, because in some cases we have a choice, and if you say no, the doctor will say, “Okay,” and they’ll just take their notes by hand like they used to.
In other cases, I’ve heard of practices that won’t accept you as a patient for treatment, and you don’t get to keep the appointment, which means you don’t get the care that you need. And so there’s this idea of consent, but there’s also informed consent, and then do you actually really have a choice? And I think those are questions that we really need to give more thought to.
Luisa: And there’s an interesting nuance in that specific use case. And most of these use cases have very interesting nuances. But in that one is the fact that we have engineered our medical care system in a way where taking notes and then transcribing notes is a huge effort that needs to be done by someone. And, you know, doctors hate it. There used to be medical transcribing services which were then shipped overseas, which caused problems, different kinds of problems.
So the idea that you could have a system help with that is, in theory, I think, really good. The problem is not the theory, it’s the implementation. Why are we implementing this as third parties with cloud access who are not careful with data? Because if you just said, like, “This is a medical practice,” and what happens is they record the meetings, and then they take all the information, they throw away the actual recording so that your voice recording is not saved, and it just adds this information to your chart. All of this is auditable. It shows every single piece of it, how they did it, and where it went, and the consequences of it downstream. That is completely different from a black box where it just gets sent to some model that then fills out something, and it’s non-deterministic, and we don’t know what happened.
And I’m sure there have been cases of this already where it’s caused problems for prescriptions, right? For anything that didn’t get added to the chart or did get added to the chart. And so again, we cannot outsource decision-making, but we can use these technologies to help us significantly if we understand how they’re working, if we are still responsible for it, if we understand the drawbacks.
But that’s a great example of no, I don’t like the way that they’re doing it now, but boy, do I understand why they want to do something different from just taking notes and then having to go put that in the chart over there or transcribe those notes. I get it. And again, I go back to, there has to be a better way.
Karen: You mentioned earlier about, you know, Instagram and LinkedIn and other services that are basically just there to get our data. Do you know of any cases where your data has been scraped and used in a way that you felt was not appropriate?
Luisa: I mean, everything I’ve ever posted on Facebook. Photos, all of it was used to train facial recognition models, everything. Instagram also was used that way. I know that that’s what they were doing. I mean, Gmail data was used to train Gemini, right? I need to double check this because I can’t just be saying these things I haven’t checked. But I heard that there’s an option to disable AI in Google now, in Gmail, but that means that you don’t get, like, spam filtering because that’s AI.
So I know that all of that data has been scraped and used. I know that emails in Gmail accounts and photos that I’ve posted on any social media up to and including LinkedIn have been scraped. Reddit is one of the largest contributors to LLMs because of the question-answer format. So I know all of that has been scraped. Like, I am aware, which is why there are no pictures of my child on the internet. Because I would like to give him the option, which honestly we weren’t given, because when we were posting our photos 20 years ago, we didn’t know that this was going to happen, and nobody asked us afterwards, right, if it was okay to use it this way, and we didn’t know.
Karen: Last question. Public distrust of these companies has been growing, AI and tech companies, and that’s probably healthy because now we know more about what they’re doing with our data. But is there any company that you do trust, or what do you think it would take for you to actually trust a company with your data?
Luisa: They would have to be very clear. So what happens right now is because a lot of people don’t understand the fundamentals of the technology, a lot of the documentation and the external statements that these companies make are targeted towards people who aren’t in the industry, right? And so they’re vague, they’re high level. To be able to trust a company with my data, I would have to actually have a conversation about, Which databases do you use? How do you store data? For how long? How do you do data sovereignty? Do you separate? Do you store data in the US, in the EU? What are your policies for retention and deletion? How are you processing? And I usually go back to, you follow the money, right? How is this company making money? So for example, Gmail isn’t free. They were using your data to train models to either sell things through ads or create Gemini.
Apple, for example, doesn’t have a LLM. I mean, they’re trying to make Siri work, but it’s like making Fetch work. It’s not going to work. I trust them more because I understand that their business model is more related to selling very expensive phones, hardware, and some software, and not necessarily suck up data to sell, vs. AWS that has the hosting and all of that. So understanding the financial motives is usually where I go. Like, “How are you actually making your money? Why do you want my data, and what are you going to do with it?” And so it would take me actually having to understand their financial model to trust them with, with my data. Or it’d be data that I don’t really care about.
Karen: Okay. So is there any company that you do trust right now, or no one has really met that bar?
Luisa: It depends on for what. I post on LinkedIn. I have an Instagram. It’s private, for what it’s worth. So there are companies that I trust, but I understand what their point is, and I treat them accordingly. There’s no company that I just trust unreservedly. You know, when Apple Intelligence came out, I turned it off immediately. So there’s no one that I trust just blindly, which is exhausting, if I’m being honest.
Karen: Yeah, fair enough.
Luisa: I feel like there’s no legislation that’s protecting me.
Karen: Yeah. Well, especially not here in the US. In other countries, they’re doing a better job of being more proactive about it and maybe not getting ahead of the technology, but at least trying harder to keep up with it and to rein in some of the worst abuses.
Luisa: That’s true, yes. We can learn from them.
Karen: Yeah. Well, I really appreciate you making time for the interview with me, Luisa. Is there anything else that you would like to share with our audience?
Luisa: No, I mean, keep learning however you can. Search out sources to understand how this technology works. If I can be helpful – this is not a plug – if I can be helpful, just reach out to me. I love to talk about this stuff, to teach people how it works. I think it makes us all better for it. So stay curious. If you feel overwhelmed, start small. Build something for yourself, see how it works. That’s always fun. You can do it in, like, 15, 20 minutes. It’s not that hard, so ... I encourage people to get their hands dirty because it’s not going to stop.
Karen: Great. So what’s the best way for someone to reach out to you if they have an interest? LinkedIn, or do you have another way that you prefer?
Luisa: LinkedIn is fine. My website is luisaherrmann.com. L-U-I-S-A H-E-R-R-M-A-N-N. I was gifted all of the consonants. My email is also luisa@luisaherrmann.com. So those are the easiest ways, because my tool means that it surfaces to the top and I don’t have to go digging through all the inboxes.
Karen: Perfect. Great. Well, Luisa, thank you so much. I enjoyed our conversation, and I’m glad we had time to chat today.
Luisa: Me too. Appreciate it. Thank you so much, Karen. Thank you for having me.
Interview References and Links
Luisa’s Five Lanes project on GitHub
Luisa Herrmann on LinkedIn
Luisa Herrmann on Substack
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 😊)
