Exploring the Evolution and Ethics of Generative AI with Sherry Marcus – Tech.Strong.Women. EP 33

Hi, everybody. Thanks for joining us for another episode of Techron Women, where we feature amazing women doing amazing things in tech. I'm Jody Ashley, executive producer here at Techron, and I'm here with my co-host, Tracy Reagan, creator, and CEO of Deploy Hub, and a weary traveler today.

She's been up in, uh, Seattle at OSS in the Linux Foundation, so we're happy to have her here. Before I introduce today's guest, I wanna give you a quick update about what's happening here at Techstrong. Um, the Artificially Intelligent Enterprise Virtual event is happening on May 21st.

This is a 24 hour global event, so please be sure to check it out and sign up to attend. Also, Textron will be live streaming from broadcast Elliott RSAC, Monday May 6th through the ninth. So be sure to stop by and say hello.

And we also have, um, our event that day DevOps Connect DevSecOps and generative AI security in the AI universe. And that will be on Monday, May 6th. Um, please go to the RSAC site and check that out and register to, to come, uh, registration is free, and we will give you guys passes to get in.

tv for great shows like this and tons of other interviews and programming. Hey, Tracy. So what's on your mind today?

Well, as you, uh, indicated, I just got back from Open Source Summit. Um, and of course, the big thing that was talked about there was the XY oodles, or what they call the XY back door, uh, vulnerability. Now, let's not be surprised, okay, there's a vulnerability, it's gonna happen.

Shocking. Um, this one was particularly a, a freak out moment because it's one of those back what they like to call a backdoor to, um, something called SSH. So, if you're a lytics person, and if you wanna get to root, you often can use SSH to get there.

Well, this particular update to the, um, this, this xz oodles, uh, library, uh, brought along with it some stuff it shouldn't have, uh oh, which allowed you to do, uh, to get really back to, to really get way down into the root. So BA basically said, Hey, here you go, and I'm gonna, I'm gonna give you a backdoor to, to do what you want at the Linux kernel level. Uh, so somebody from Microsoft discovered it, uh, this was in, uh, early April, and people are now sorting out what the impact was, if it, uh, if, if anybody was really seriously impacted by it, and now we have to get it fixed.

Obviously, it's gonna be fixed at the lower level, but that means every single company, every single, everybody, everyone who's using the Linux Kernel or Debian, they have to do an update. Oh, wow. This is our new, this is our new world.

Um, we oftentimes think that there are more vulnerabilities, uh, today than there may have been in the past, but I think we're just catching them. So we have, we're, we're, we're doing our job better, uh, and I'm glad that they caught this. I'm glad somebody at Microsoft were on their toes and, and, and saw something was wrong and dug a little deeper to find out what if happened.

But it was a complicated, this was a, this was an attack. It was a very complicated, um, process. The way they got it in, it took them months because they had to actually get a pull request approved to get it pulled, pushed through.

So it was a very serious attack, and it really highlights, uh, the need for us to always be on our toes. And I think it highlights the need for us to be looking at ways for AI to help us solve this problem, because humans, we have so much to do it. You know, we, if we have to rely on somebody who just happens to catch something, we, we shouldn't, we should acknowledge that we, we, we need help.

And, you know, I don't know what the next evolution will look like, but I really do believe that, uh, AI and generative AI will help us solve this. And I'm glad today that Sherry is going to help us with navigating that technology so we can better understand, um, how it can be used, how we can apply AI in these kinds of situations. Hmm.

That is why we invited Dr. Sherry Marcus to join us today. Um, we all know AI is such a big topic, and, um, she's a pioneering leader and responsible ai, and I'm really excited to have her here.

Um, Sherry, tell us a little bit about yourself. Yes. Uh, thank you Jody and Tracy for having me here today.

Uh, so my name is Sherry Marcus, and I lead, uh, bedrock Science for Amazon Web Services. Uh, bedrock, as you may be aware, is, uh, AWS's generative AI service that allows customers to choose from many different models, uh, world-class foundational generative AI models, and utilize them to create generative AI applications or their specific, uh, business areas. And so my organization in science, uh, creates the generative ai, uh, algorithms and analyses that powers, uh, many of the bedrock workflows.

So, educate us, you know, there, this is a new field and for, you know, there's only a few of us like yourself that really are experts in, um, these areas. Could you go through some of the different common, um, you know, generative AI machine, uh, machine learning, na, you know, natural language processors. Can you give us just a kind of a quick education around these types of tools and how they're applied?

Sure. So, you know, I think if you go back maybe 20 years-ish, um, uh, the field begins with, uh, natural language processing, where it really started, where you have, uh, text, you know, when you wanna be able to, you know, search the text as you see in, you know, things like, uh, you know, any kind of search engine, or you wanna be able to take a text document and put it in a structured, uh, format of such. So you have like a text document that's maybe a transcript of, uh, an earnings report or, uh, you know, a or, or some meeting.

And you take the, the text and then you say, okay, here's the date of, you know, of the document. And you parse that out into a record, and here's the people in the, you know, people's names in the document, and you parse that out. And then maybe there's a subject you know about it, and you can create a, uh, topic, you know, algorithm, which can find topics associated with the, uh, document.

So NLP really, uh, tried, you know, in summary, a lot of what it, you know, did was try to take the amorphous world of textual documents and put structure around it so that we can search it, you know, in, um, lots of different ways. And, uh, then, uh, you know, artificial intelligence, uh, you know, came along, uh, with, uh, neural networks, um, uh, and neural networks, uh, really looked to determine how to do various types of, uh, completions of specific, you know, text documents or, or specific types of, of words. So you might say, you know, Mary had, and based on looking at, you know, a, you could res, you could a neural network might predict a little lamb, or, you know, Mary had maybe a bad day.

And, you know, there would be various, um, probability distributions of, of how you could complete that sentence. And essentially, uh, with the rise of, uh, three different events, um, came generative ai, you know, out of ai and this kind of completion idea. The first was, uh, we saw we have lots and lots of data on the internet.

The second is the rise of, uh, GPUs and much more fast AI accelerators. And, uh, the third thing is, uh, this notion of a transformer, which is a new type of neural network architecture, which allowed one, not only to do, uh, you know, completions of, uh, with one or two words, but to be able to complete or create entire documents. So that's the rise of gender of ai, those three trends.

And that happened about, I think, I wanna say 20 19, 20 20 when it really hit the marketplace. So what gender of AI can do, right, is it can do much more than the sentence completion. It can summarize a document, it can, uh, you can do question and answering with it like, you know, some kind of chat bot.

Uh, you can ask it, you know, to write poetry, you know, or different kinds of documents. And not only can you do this in text, but you could do this multim modally, you could do this and, you know, say, I wanna see a picture of Sherry with blonde hair or black hair, you know, jumping, things like that. So, I know that was a bit of a long answer, but I hope, uh, it, it, it gave you a little bit of a historical view.

Uh, I love it when people can break things down into really a simple way of understanding, uh, because I think that as everybody talks about ai, it becomes, it's mysterious, right? Yes. When you break it down into the simple structures and the, it's, it's history.

It doesn't become so mysterious. And then we can start thinking about, oh, that's not that hard. It's the, it is just an evolution of software, right?

And GPUs, the first time I ever saw A-G-P-U-I swear, or a system that was county GPUs, I was at a developer, uh, kind of a meetup, and I wanted to get up and run around. I was so excited. I felt like it changed my world at that moment.

Yes, yes. So, in the, in the space, you know, you talk about data and how important data is. Yes.

And if we talk, if we think about software development, and I think about this all the time. We have a, we have a, a huge amount of data in, in things like GitHub Mm-hmm. GitHub has an enormous amount of data.

But beyond GitHub, we don't have a lot of data to, in terms of the lifecycle of a, a software. If we wanna try to solve this problem of security in software and try to stop things like XE backdoor, I feel like we have to have more data to do it. And we don't have, we have data on the left side of software development, which is stuff and Git Mm-Hmm.

But we don't have the data on the right side that shows the flow of, of software going into, you know, other open source utilities or even into enterprise inner source. What amount of data do we need to be able to do that? What kind of, what, how large does, do these data sets need to be?

Are we talking about, let's say, 50,000 pipeline transactions or of a, you know, of a software, a piece of software, you know, Jenkins workflow? What would we need to be able to build a, a true environment where we could start applying these techniques to the right side? How things are getting deployed and how things are getting consumed?

You know, that's a great question. Um, so the first, uh, uh, you know, what I was thinking as you were saying that is, you know, access to the data, you know, because if I'm deploying a piece of software in my environment, right? Or, you know, in a customer's environment, um, I have to be able, uh, I, you know, and, and for me, and in order to learn from those kind of environments so that you could train, uh, you know, with it on a gen ai, uh, system, you have to, you know, be able to access it.

So it would be interesting to me just to be able to get access to that kind of data, uh, because it is sensitive, right? You know, people's environments and what they're training on. But I do think that there should be, uh, some kind of, uh, you know, repository, you know, where, uh, you could train, you could use a gen AI to train, uh, uh, systems, right?

That go on different environments. And then, you know, the process of using a gen AI would be to, uh, ask it, you know, it's chain of thought using a chain of thought reasoning function within a gen AI that says, how would you attack, you know, how is my, um, uh, library, you know, as you mentioned, vulnerable, you know, in this situation, in, in this environment. And then, um, the gen AI would be able to go back and look at all of the vulnerabilities associated with the environment, right?

And then the features associated with the, uh, library that you're looking to deploy and do that analysis. So I'm answering these, I, I guess I'm answering this question thinking out loud now, but if you had lists, okay, of the functionalities of the libraries and how it runs the software, and you had the, uh, environment variables that you're deploying it to, as well as, you know, all the information about its vulnerabilities. You could train a gen AI to be able to look at, uh, known vulnerabil, you know, known vulnerabilities.

So it's not so much how much data would you need, Tracy? It's really just getting it all. Um, Yeah.

So when I was at the open source summit, you know, I was questioning, I was talking to people about this, and we have something that has a, a lot of information, an SBO m right? We have this software bill of material report that has a ton of dependencies in it. For every single object that's out there, every single open source library that's produced has an sbo MM-Hmm.

Um, and, you know, I was talking to folks about if we could pull all those SBOs into a central repository mm-hmm, we could at least begin to see the flow of the packages and how they're being consumed by all of these different, um, open source libraries. It feels like that could be a starting point for that kind of information. Now, the side we don't have would be the corporate side, the environment variables and the deployment process and, and how they're being consumed.

So let's say we had a public access of the SBO m information, which would have quite a bit of data about the, the, the, the packages, the dependencies, and then we had on the right side private repositories that could consume that data and associate to their deployment, uh, endpoints. Yes. Is it possible for a, in a private company to keep their information private and build that kind of data so that they could do that kind of learning?

A hundred percent. Um, that would be the, uh, type of functionality, uh, that, uh, private companies would build. So that is a possible solution, right?

Mm-Hmm. Because now if we get there, we begin, we, we can probably do better job of threat modeling. Yes.

Yes. I mean, the additional data that you would need, right? Is you would need a library of all known types of attacks, you know, against different, you know, components, uh, as observed between different components in software modules and the environments, you know, and that would be the basis, right, of being able to learn and infer, um, you know, the, the necessary, uh, inferences made, you know, within the SBOs, as you said, that could, that could ultimately open up an environment to vulnerabilities.

I think that there is so much, um, that we're leaving on the table when it comes to taking advantage of the data in the SBOs Mm-Hmm. Um, and being able to, um, dissect it and apply it. Yes.

Uh, and I feel, I, I feel like it's a conversation that, and, and from the, the DevOps community that we really need to be, we really need to be looking at. Um, yes. But a couple of the people I talked to about that, they look like, well, they looked at me like I was green, and they said, there's no, that's gonna be this massive, huge database, you know, would be thousands and thousands and thousands of transactions.

And, you know, I used, I came from the financial industry and thousands and thousands of transactions doesn't seem like much. Right? I did too.

Yeah. A hundred percent. Yeah, a hundred percent.

You know, the, you could also do the other side, right? As you know, uh, with GitHub, right? It's being used as a repository to do, you know, to teach, uh, software engineers, you know, how to cope, you know, with, you know, better coding ideas.

So it can also be used, uh, you know, these libraries that you mentioned, you know, can also be used as a way to do training and to help system designers build, uh, less vulnerable systems, okay. By, uh, using the AI to help you figure out how someone could attack your system, you know, as you're building it. And, uh, you know, to me, uh, you know, that's, you know, that's where I would start, uh, you know, just before I would even bring any system into, uh, you know, my environment.

I would work backwards from that. Very, very interesting. I'm, I'm so glad that I, that you are our guest today, because it was something that was so fresh on my mind, uh, leaving that conference and seeing feeling, um, that, you know, maybe I was crazy.

Maybe I am green, uh, but I, I really believe that some kind of shared, you know, SBO m evidence store that could be managed by an open source foundation who wants to help solve this problem. Mm-Hmm. Would be a super useful, uh, be starting point for applying AI and starting to look at threat models and even the threat, the, the, the, the, the xe, uh, backdoor that we just discovered.

There is a, that's a whole model we, that that whole model was, should be something that we should be able to capture in some way to see if we're seeing that pattern somewhere else. Yes. Yes.

I completely, So a lot of times we talk about, and I've read a lot of this in your bio, and, and we, this is a conversation, Tracy and I have a lot, 'cause AI can, on some levels really terrifies me. Um, but the, the idea of responsible AI and ethical ai, how does that play? When you guys talk about all this data and let's make it available to everyone, um, how do we do that in a way that is responsible?

And what, how do you define that? So, you know, to start, uh, when you think about, you know, responsible ai, it's a team sport, alright? So it, you know, not only my organization and science, but you know, senior executives, policy makers, um, and all relevant stakeholders, you know, really, uh, uh, should be, uh, are part of the community, right?

To think about, uh, responsible ai. But, you know, the path to responsible ai, you know, as we were, you know, talking earlier, Tracy is really, it begins with data, right? And so, uh, data, uh, is used as, you know, to train, um, you know, generative systems and data is, uh, inherently biased.

You know, this is something that none of us, you know, can control, you know? Uh, and, uh, part of what, you know, I do, you know, in my day-to-Day, uh, activity is work very hard to, uh, uh, remove this bias, you know, within the, uh, uh, within the bedrock platform. And there are a few ways, you know, to do this.

Um, so you've trained a model, okay. On all this data to begin with, you know, and, you know, without doing anything, it's gonna be biased. But then what you can do is you can correct it, right?

And, you know, teach it, uh, more neutral ways. So rather than saying she is pretty okay, which you might see in lots of different kinds of examples of completing the sentence, she is, you could say she is smart, right? Or you might see examples of, uh, when you say a doctor, okay?

You might think of, you know, a certain demographic and, you know, these responsible ais might, you know, might give you alternatives. So that's kind of the first step, you know, that we do. And then, um, after that, uh, you know, as a second, uh, there are other steps, but another big step is something we call guardrails.

And this is really in business areas, right? Where, uh, you just don't wanna have any mention of any kind of biased or toxic topics, okay? At all, right?

So you are an airline, you know, doing business, and you don't really need to know about crime you, when you're committing, you know, talk about crime in chat bots, right? So we eliminate things like that. So, um, as well as doing a lot of other, uh, automated abuse detection functions, uh, within, within Bedrock.

So, um, you know, at a high level, that's, that's the data side. Uh, the second is in the area of education, you know, and we're committed, uh, by 2025 to provide free AI skills, uh, to 2 million people, uh, on ai, uh, data security and cloud courses through our online learning programs. Uh, we also have invested a hundred million dollars in the AWS Generative AI Innovation Center, which is a program that rolls out and helps bring a lot of our bedrock applications to customers responsibly.

And we brought, you know, this to, you know, many, many companies like, you know, NatWest, Bridgewater, United Airlines, uh, Delta. And lastly, uh, we've encouraged and supported, you know, white House voluntary AI commitments sponsor United Nations session on generative ai, as well as, uh, the AI Safety Summit in the uk. Wow.

So, effort is being done, effort, lots of effort is being done around this to solving this, this problem. Um, unfortunately, there's never gonna be a way to stop the bad actors. It's just gonna happen.

You say it's a group effort, it's still kind of frightening, isn't it? 'cause bad actors are part of that group, right? They are.

They are. You know, and, uh, again, what I can say is that every part of the Bedrock organization, you know, from ideation to production, you know, we build in responsible AI because we, uh, you know, the importance of it is, is really part of the DNA of the product and what we believe, uh, are the right solutions to bring to our customers. So let's shift the conversation just for a minute.

Um, I, this was a question I asked one of our other guests, and I took the recommendation, read the book, uh, the, one of our other guests suggested, uh, to read the, the, the logic of failure, which I found absolutely fascinating. Uh, it's gotta be, I I, I'm gonna have to go and read it again because it, it was so incredibly insightful, the logic of failure. Is there a book in particular in the, in AI or in any, you know, domain, but it would be great to get one in, in the AI domain that our listeners should really consider reading.

Do you have a favorite? Yes, I do. There is a book by Steven Wolfram, uh, and he recently, it's on Kindle, uh, and I think if you're on Kindle, like the free part of Kindle, you can buy it, you can get it for free.

But it basically explains generative AI from like, foundations. I forgot the name, like the exact title. It could be Generative ai, but it's by Steven Wolfram.

And the book just goes from literally first principles, uh, of generative AI in a very, um, uh, explainable way. But he also goes, you know, very deep. And I had been looking for a book, you know, that I can, that I could give to, uh, you know, customers who, you know, wanted to learn something but didn't kind of want all the hype associated with it.

And it was just a very well thought out, uh, uh, book. Did you? Yeah, did, did, how do you spell his last name?

Is it, Is it, can AI solve Science? Uh, I was looking on his website. That's the beauty.

You know, we can do this in real time. Yes. This was from March 5th, but I'm not sure if that's his book or if it's a paper.

Okay. Just gimme a second and I'll look. Uh, I learned so much from, just from a management perspective and a life perspective, reading that book, the Logic of Failure, um, what is, and I know I'm, I watch it on the happening in real time in politics and I'm like, wow, it happens everywhere.

The name of the book is What is Chat? GPT Oh, it's the chat GPT book. Okay.

Um, but it's not really about chat GPT. It's about how generative AI works from the beginning. Right.

And, Uh, it, it's, it's a very short book. Um, but it's just very, very well written. Great.

Okay. It's on my list. And I love it that it's on Kindle 'cause I'm on a plane a lot and there's no better place to read than when you're stuck on a plane and nobody's calling you and you're not having any meetings.

No, it's great. I love, I love that you asked that, Tracy. 'cause it's always good to have a takeaway for someone to go check something new and different out.

Mm-Hmm. So, I, I have to ask this 'cause I think that your historical, um, and your bio is really cool. How has being like part of our national security organization and the I a, I mean, obviously that was some data that you, uh, that you touched that was really high level.

How has that helped you in what you're doing now? Yes. So, uh, when I started, uh, at the CIA, um, you know, it was big data then, right?

And it's, uh, probably we would call it tiny data or, you know, uh, now, right? But, um, you know, one of the things I really learned a lot, you know, from, um, you know, working in the national intelligence space is the, uh, there really are so many different signals, you know, in the world that, you know, then I never thought about. Right?

There's, you know, what they call signal intelligence, which is telephone calls and things like this and transactions, and there's, um, mobility intelligence, okay. Where people are okay, and how fast or where things are right. And how fast are things going from A to B and, and all kinds of human intelligence about what people are saying and things like that.

So the goal of, uh, of a lot of what I did was like, how do you fuse it, you know, all together Okay. To get a single cohesive picture of, you know, what your target was, right? And that was called, uh, you know, fusion.

And, uh, but just in, in terms historically, uh, and, and I see that often that theme of, you know, when we do that here at, at, at Amazon, right? We have, uh, a lot of different kinds of customer data sets, whether they're transactional, whether they're, uh, like a database, whether they're text, you know, like free text documents, whether they're images, you know, contained within PDFs, you know, and you have to take all of that and turn it into content, right? That you can actually reason about.

Right? And over the last, uh, you know, 20 odd years, we've come a long way, you know, in really being able to understand how to do that. But historically, I just wanna say like, one thing I think that I feel I've been through, um, at least three revolutions, you know, um, that, that have affected me technology wise.

Um, and the first was obviously the internet. Okay. But the second was mobility applications within the internet, right?

Because with a cell phone, you know, you can track wherever people are or things are, and then you have all of these, you know, great applications like Uber, um, and, and figuring out how how many people are walking into your stores. Um, and, you know, that was a very new thing, you know, uh, still is a very new thing for lots of people, people. And then of course, generative ai, you know, was the third.

Um, I guess to answer your question though, about like, data and, and national security, and I've seen all of these things Mm-Hmm. You know, after seeing all, like, so much data, you know, like prior to Amazon, you know, I worked in banks, you know, like I was, you know, head of AI at, at BlackRock prior to this, I had seen lots and lots of data and it becomes clinical, you know, it just, you know, it's just data, you know. And I try not to pay attention to the content.

I really am just trying to solve problems, uh, for, uh, the business. Yeah, I bet Having, having access to that high, a level of stuff that could be flying around the world has been helpful with ai. 'cause it gives you like a whole leg up on what you should be looking for, right.

And what could bad actors could be doing. And you probably had more exposure to that than the average person diving into AI today. A hundred percent.

A hundred percent. You know, and I think even the notion of privacy, you know, generationally, um, and maybe I'm going a little bit off topic, you know, has changed quite a bit because, uh, you know, I have two children, um, you know, one is in college and the other is out of college, and they seem to have much less expectation of privacy, you know, than I do. And, uh, uh, I just find that very interesting.

Um, Yeah, I'm, I'm in the same place. My kids are outta college, and, and it amazes me what they're okay with. Mm-Hmm.

And I'm like, wow. I don't, I don't want, it creeps me out that people can chase me around if my locator device is on, or I walk in and out of a store and someone sends me a, you know, an ad for something the next time I'm online. Yes.

My kids are like, eh, whatever. They just, they don't think that's a big deal. So Yeah, I agree with you.

Until it impacts them in a bad way. Well, you know, you know, that's, they haven't, they, they don't have the history of what could potentially happen. We don't really either.

But I think that we have been, you know, we were raised to be told, don't get in a don't get in a car with a stranger. And now that's gone. Right, Uber.

Exactly. So we were raised with different, uh, parameters than they have that they're, they're experiencing, you know, I've spoken to people who have been to China and they're young, and they come home thinking differently that maybe it's not so cool to have so much of your location who you're interacting with, um, be exposed to, um, you know, big government. Uh, and, uh, and you know, and here we have a, in the United States, I think we have a problem with how much does a government know what we're doing, and how much does the capitalism know what we're doing, right?

Because there's money to be made off of, of, of what we're doing. Uh, Sherry like you pointed out, how many people are walking into your store? Should you keep that store open based on that?

So, you know, I think that there's going to be a, a, a learning experience for younger people, uh, as we've moved down this kind of road of exposing everything that we're doing and telling everybody, taking pictures of our food all the time, and, and kind of, you know, too much information, uh, uh, culture. I think it will, I think it will shrink. I think it's gonna, I think it's gonna go, go the other direction after we get used to it and decide that maybe it wasn't the greatest idea.

I think you're right, though, between privacy expectations that I have and the over the, the immense amount of oversharing that goes on, It's insane Social media person. I, I, I keep up with what my friends and and acquaintances are up to. I don't engage in a lot of it.

Um, but it, it just amazes me. I'm like, I don't need to see 250 pictures from your vacation. Yes.

Why do you push that on? So show Me five or six of your favorites, and I'm good Favorites and, and pictures like I've always been told. Make sure you take a lot of pictures and make sure people are in them like you're in them.

'cause otherwise you just have pictures of stuff and that you see 200 pictures of stuff and you're like, okay, great. Yeah. Like you said, just pick your five favorites and post them.

That always strikes me crazy how much people think the rest of us wanna, wanna know. So this topic brings me to TikTok. Okay.

Oh, boy. Is, is TikTok a a really a problem? Um, now I know I've spoken to people at like IBM who are working more closely with the government, and they're like, yeah, it is a problem.

Sherri, what is your opinion on TikTok? Is it, are we, is that a place where AI could be used against us? So I'm not, uh, an expert, um, you know, in, in, you know, uh, the internals of how TikTok works or what they collect.

Um, you know, my personal opinion, right, is that, uh, I just think, uh, we spend much, much too much time on TikTok, you know, uh, rather than, um, uh, doing other, you know, things that might be, you know, more productive. Uh, I just see, you know, taking the, uh, train so many younger people on TikTok or listening to TikTok, and I don't even understand, you know, most of the little lips, you know, they kind of go completely over my head. I'm sorry.

That wouldn't I agree. Well, there is another good book to read on this particular topic, and it's called The Coddling of the, uh, American Mind. Um, I, uh, I think every parent should read it, and it's about, um, you know, young, we're talking children, uh, you know, 10 to 14, uh, what too much social interaction on a machine does to the American Mind.

Um, so it's called The Coddling of the American Mind was written by a, um, social media, um, specialist, uh, somebody with a PhD in social media. And I've heard him talk a few times. He was on, um, bill Maher, and I was fascinated by him.

So it's another book to read on this topic if you're, if you're a concerned, uh, parent. Yeah. But on that topic, you know, I do, uh, as I mentioned, I have two girls and, uh, you know, as a math PhD, uh, you know, I, it was important to me that they had a lot of exposure, you know, to stem and, um, a lot of different areas, you know, without, you know, doing it in a nice way without making them, um, you know, crazy.

Like, you know, go away, mom. And, uh, you know, one of the things I remember doing was, uh, you know, I think my daughter was in the sixth grade and she had to do like, you know, the area of a triangle, you know, or square. And then I like took the piece of paper and I crumbled it up, and I said, okay, how do you compute the area, you know, along the surfaces of this paper?

You know, because that's, that's a math topic. Okay. Uh, it's a whole branch of like, mathematics.

And, uh, so I had both my kids in the room, you know, and one of them was like, you know, trying to think about, okay, well, do I add up all of the different, you know, flack parts or like, what do I do? And the other one decided she was just gonna draw a picture of it because she thought she likes art. Um, but, uh, you know, I just think, you know, I know that's a simple example, but I think there's just so many things parents can do to just bring, um, uh, more productive content, you know?

And there's so much great content now in STEM for, uh, not only girls and boys, right. To like, introduce lots of interesting topics, uh, uh, you know, you know, AWS amongst many others. And so I just wanted to, uh, you know, back to TikTok, you know, I would just rather see, uh, youngsters thinking about that a bit more.

You know, Well aim into that. I was pretty lonely at Open Source Summit. There were not a lot of women there, and I was very disappointed.

I, there were not a lot of women presenters. There were, you know, there was a, any activities I went to afterwards, it was like, I'd look around and we'd look at each other and there was like four of us and, you know, 70, 80 men. So yes, ladies, we need to see you at these, these, these shows.

We need to see you in person and, and get your daughter's learning stem. Hmm. A hundred percent.

Absolutely. Well, this is probably a good place for us to wrap today. Thank you.

Um, thank you so much for being with us. This was, this was a really great conversation. Um, I just think AI is such an important topic, and it's, it's great to see, um, everything that you've done.

I mean, your, uh, your, your bio's really impressive. And we were very excited that, uh, Emma, Emma reached out to me and said, Hey, would you guys like to have Sherry on? And Tracy and I were both like, absolutely.

So thank you so much. We, we really need your voice and we need you here talking to all of us, and we appreciate you being here with us today. My pleasure.

Thank you so much. Okay, everyone, that is a wrap for our latest episode of Textron Women. Please stay tuned for all sorts of great programming on Textron tv and we'll see you next time on Techron Women.

Thanks. Thank you.