Embedded - 306: What Is in the Magic Box?

Episode Date: October 17, 2019

Dr. Loretta Cheeks (@loretta_cheeks) spoke with us about implicit bias in text, machine learning, getting a PhD, and STEAM outreach via Strong Ties (strongtiesaz.org). Also see: Loretta’s research on identifying implicit bias (The thumbnail image is from her work.) Lotetta’s TEDx talk on AI and remembering Yoshua Bengio wiki   Thank you to our Embedded Patreon supporters for Loretta’s mic, particularly to our corporate sponsor, InterWorking Labs (iwl.com).

Welcome to Embedded. I am Elysia White. I'm here with Christopher White, and our guest this week is Dr. Loretta Cheeks. We're going to talk about AI and getting people interested in STEM and STEAM. Hi, Loretta. Thanks for joining us. Thank you. Thank you for having me. Could you tell us a bit about yourself? Sure. I am a thought leader in artificial intelligence and also in a consultant in that role.

I research as well in that role as well. I am a STEAM advocate and also the founder and CEO of Strong Ties, a nonprofit that work with youth in STEAM education for engaging and introducing them to the world I love. Excellent. We're going to do lightning round where we ask you short questions and we want short answers. And if we were behaving ourselves, we won't ask how and why. Are you ready? Okay. Yes, I am.

Do you think we should bring back the dinosaurs? Yes, I think we should learn from the dinosaurs. There's a lot to be learned. What's something that a lot of people are missing out on because they just don't know about it? Artificial intelligence. Okay. What's your favorite artificial intelligence or ML algorithm? I guess it would be Q learning,

which is enforcement learning. Reinforcement learning. Yes. When someone finds out what you do, what question do they always ask you? They think I'm a rocket scientist. Is that bad? I mean, that's not bad, right? No, they just don't know what to do with me if they are not my world. What is your favorite fictional robot? Oh my gosh, a Beena.

A Beena? Beena. It's Beena. Okay, what's that from? Okay, so there is a robot called Bina, and Stephanie Dinkins is working on it right now to train her to be Black and to learn Black culture. Stephanie is an artist who intersects AI, and she's training her about the contextualized Black experience.

Bina. Okay. Hey, look, I can find a link and everything. Excellent. And one more question. Do you have a tip that everyone should know? Yes. I would say take every opportunity to learn as you go on your life journey. Learn. Enjoy the learning. Enjoy growing. All right. That's a great tip. But let's get to your career. You work on AI. What does that mean to you?

So AI is big, right? AI encompass supervised learning, which we know as classification, and encompass semi-supervised learning as well as reinforcement learning. So my work in AI is primarily focused on unstructured text using online news. So I am interested in modeling what we know as social influence and understanding migration paths about topics. My studies were in water insecurity, and I wanted to understand how attitudes, beliefs, and values change about the topic of water over time in space. Okay, water insecurity. This is like Flint, Michigan, or like the Sahara Desert? Well, I studied the Southwest region, but it can be any topic. The idea is to use machine learning to understand what's embedded within text. So water was the subject I chose. However, if you're talking about annual reports,

you may want to, before you have a Bernie Madoff, you may want to have insight into something like Bernie Madoff because it was embedded in the text. It's just that people, there aren't enough humans to consume all the information that's embedded in all these bodies of text. So you need machine learning to help. And that's what I do. So what sort of primary sources do you use as input? Yeah. So I actually created a data set using a news that was about the topic of water. I yeah, I wrote programs to basically crawl different seed links and I wrote my own program to actually read that using a bot. And so these, this was searching the internet or searching

government databases. You mentioned reports. So I'm just kind of, are we talking Twitter here or are we talking official information? No, no, no. Official news articles is the use case that I used as my data set. So I use national and also local news that's a regional news that's affiliated with the Southwest region. And when you talk about embedded, do you mean the biases and connotations that things are written with? A reporter's news are both internal biases or slants or frames human interest, management, which is authority, scientific, education. There is another one that's escaping me, but, oh, economics. And those are the five that are primarily used. So when combining that as well as internal biases, you can learn a lot by the cues that are embedded within the text.

It's kind of like if I were to write a paper about California, it probably would be very obvious that I like California, even if the paper was about water shortage. Maybe so. It depends on who you are. Yeah. But that may be a close correlation. Yes. And so the biases you were looking at, what kind of biases did you find? Economic frames, well, at 80% time led to the most articles that were written that were being fleshed out to the people. Very few human interests, oftentimes scientific because it's flooded with so much data. People are most not interested in that. It's a small sector of folk that are interested in that. It's a small sector of folk that are interested in that, but for the common

public, usually it combines both economic and managerial frames. So with that being said, when management or elitists, authorities speak, people listen no matter what it said, they believe this. And that's just a whole line of research that show, that suggests that when elites are people who are in authority and then you combine the economics, that is a strong slant. And it's a strong way to have people to actually change their attitudes about subjects. And what about the biases? The frames make sense. So, okay, so when you think of a news article, there are two coefficients that I calculated, right? So the internal is the producer or the news reporter's internal biases. So I created a mathematical formula that actually calculated the bias intensity. There is a level of intensity that can be showed over time. I combined the Cusco catastrophe theory to actually show over in this possibility space

how much biases was applied dependent on, for instance, I looked at variables, peace, meaning communities feel as though they are at peace, and also another variable that was used, frame type, of course, was used. Those are external variables. And does that cover the implicit variables of the writers? Yes. Okay. Yes, yes, yeah.

So it's looking at internal and also external. Another way of looking at it, yeah, that's, you know, with those two coefficients, I'm able to show that you can use multiple one or multiple different factors to show the impact of external forces on with combined with internal biases, the impact of the messaging. And that's very attractive for a lot of people at this time. So because they're looking at, for different reasons, they're looking at convergence for advertising. That's one of the things Adobe was interested in, who helped to pay for my PhD, as well as other things. And so how did this work? I mean, I'm a little familiar with like the recurrent neural networks and the LTSE, I don't know, random letters. Okay, so you get a bunch of articles from the newspaper from the Southwest region, and then a magic box, and then output. What's in the magic box?

Oh, you're so funny. So the first step was, because it's an issue, framing and computation had never been defined. So the first thing I had to do is define what that was. So I gave treatment to the first competition for framing, as we know in the news, that was the first thing. So what is a frame, right? And so what that looks like is you receive or you use input news articles, you clean that data. And that allows you to have a nice, you know, once the data is clean and goes through this integration process, you clean the data. And from there, you're able to do similarity matrices to figure out how similar different terms are to the next. You use TFIDF, which is a term frequency

to show how important certain words are as they compare to other words that are in this body of work. And then from there, the next step was to actually show, you know, the comparison so that you can actually compare, you know, over one article over the other, if what's the different pattern that's happening in one body, one document, as we called it, to the other document space. From there, now that you have that in conjunction with that work, we did content analysis as our ground truth. I was able to do that with some researchers in the Center for Policy Informatics at Arizona State University so that I can actually have ground truth to determine when I do actually start my classification, I'm able to have something to compare it with.

So the documents were the first thing to just get me to the point that I now knew what frame type was associated with every document. So at the end, so input, you have the news article for every article, there is a particular frame type that has been used and we call that a signature or a pattern. And from that pattern, then now that we know the frame type, the interesting part was to understand now with this body of documents, how are people either being influenced or affected by that document and that message? And how, what's the bias intensity? And then what's the possibility space? What actually can happen? What are the other factors that you can consider for moving

around in the possibility space. And for that, it looks more like if you look at the research that I did, if you're familiar with electrical engineering and so forth, you're in embedded software, right? Yes. Okay. So you're familiar with a state machine. Yes. Right.

And so it looks more like a state machine when you slow all things down. The frames or the AI pipeline? Both. I mean, so machine, yeah. So the framing plus the external forces of factors that were calculated, they were the actual variables that were used, the control variables and the bifurcation variables that were used to determine where you are in or what state you're in, in the possibility space. So, you know, just think of, you know, once you have frame, you have other variables that you're considering. For me, it was peace.

You can now determine the bias intensity within this possibility space, but the way that you move around is through similar, like a state machine. Okay, I think I get it. Yeah, and what it looks like over time is literally either shrinkages of growth over time from year to year based on messages that

were the message of interest to you. Now that's with online news. And again, if online news is your data set that you're interested in, that's fine. But any body of text that use more narrative, as you know, the reason that news is so flavorful is because people add lots and lots of words to make them colorful. With that, lots and lots of words, that adds a lot of subjectivity. And so I introduced this concept of double subjectivity. First level of subjectivity is if you witness something with your direct eyes. The second level of subjectivity is when you hear it, online news with a plethora of online news, there it is. It introduced a whole world of texts that you can have many, many, much, too much for the human to digest in indirect news or indirect subjectivity of any body of work, any text. And so this was supervised learning. You had to go through all of this data, didn't you? It was semi-supervised.

So after, yeah, because if you look at the final work, it looks like clusters. Okay. So, yeah. So the first step was the classification or the detection of frames. And that would be the comparison was classification. And then later on, I had to move to clustering because I had to create a visual language whereby you can describe visually what's happening in space and time. I've always wished that reporters or any of my news sources or any of my content sources, I guess, could register what their explicit biases were.

So I'd know what to expect. You know, if I'm going- That would be really nice. Fox and this show has this sort of bias and that show- Good luck. I know, right? Well, I mean, I'm willing to say I have biases. I'm going to say I like California or I wouldn't live here.

I believe in women in tech is an important issue we should talk about, although I don't really want to talk about being a woman in tech. I have biases. Right. But there are ones I'm willing to say, okay, yeah, you listen to the show. You may be being influenced by the fact that I really want to have women guests as well as men guests. But how do we get that? How do we get a database of? Well, I mean, you can get it. I mean, if you do like a model like mine, you can actually see it, right? But unless you're doing it, the bottom line is unless

you're intentionally looking for those things, most of times the way we learn, it's not obvious. And with the amount of data that's coming in across our path every day, people are signaling or queuing. There's always some kind of clues or cues that is coming your way. No one's no language that anyone use. Usually I'm talking about news in the news and even for that matter on Twitter these days is not just for anything. If someone says something, you really need to ask, and why? Oh, so much.

Yes. Because I want people to like me. Oh. Yeah. And why? Like, for real? Yeah. So oftentimes there is a flat.

And when you, you know, you could kind of, what I found, you know, looking at managerial and economic frames, where there was a governor, there was always money. Or where there was a politician, you can guarantee there is always money. It's always tied to money and the interests of money. So what I show, for instance, through language, for a good example, you can think of, there are articles that now say the new welfare queen, right? And a new welfare queen is that baby in the arms of an immigrant mother who's crossing the border. All right? On the other hand, if you look at some of the articles

that are produced for the farmers, it's subsidies, the language is subsidies. That's the same thing. Both of them are basically using government or taxpayer dollars to get something, right? But some are painted as takers and some are painted as investments. Then, you know, so you have to look at

what's the agenda of the news producer. And what's most interesting is how do people take those mass medias, you know, these media outlooks, the Washington Post, you know, the New York Times, how do they take that to their my media, the Twitter spaces and start propagating those messages and getting people to listen and so forth. And there is research to say if people start to, for instance, if people, a lot of people start to listen to Alicia and Christopher, then there is an embedded something that starts to engrave in their brain.

It's really a physical action that starts to happen through this repetition. That's why propaganda is so good. Right. And it is. And I really like the idea of your research looking at these biases. And I want to go further. I want it to be that these biases are in a database somewhere where I can query, oh, I read this article by so-and-so, is this a climate change scientist talking good things, or is this some random person trying to convince me that the dinosaurs are here already? The dinosaurs are here. It would be nice. And there are some researchers who are definitely working. I do know some people are working on detecting what's faking, what's not, right? That's happening. But the level of intensity,

you know, to actually have some kind of measure of intensity, there isn't a lot of research to determine that. And the reason is it takes a long time. For instance, a lot of people can do a lot of things in AI, but how do you prove that it's any good, right? I mean, everybody can claim that something is good, but you know, as engineers, the proof comes in the pudding. Can you test it? Right. What is it tested to? And so, uh, which means like the work that I did, and it was a small amount of content analysis, but it took four people over a year to actually read through articles, to actually come up with these labels. Annotation is very hard. Um, it takes time. And for that, so a lot of people are making a lot of claims, but we want to be very good stewards when we're saying what is,

you know, a certain degree of biases. And I would say that my model does that. I am trying my best to make that available in a platform right now. So hopefully you'll get your bias intensity where you can dial it up to see what is this BS or is this somebody trying in their subtle way, trying to get your mind to twist this way of thinking. With so much looking at the articles and having people mark them, and you mentioned data cleanup, how do you figure out the biases you're putting into your bias detector? Oh, well, I do feature extraction, so I don't put anything in. I only extract out. Oh, because you mentioned clustering. That people were using ethnic names to do classification. In that case, you're adding a whole bunch of bias, right? So that means that when you see a certain name that you automatically classify and then lead to advertisement that ain't so good for people of color on the margin. So in that case, then there is a great degree, a great level of biases that's also being added that we do not want to introduce.

And it's done a lot because as well, biases can be entered if you're only looking at a certain group of individuals, right? And you're just not capturing the diversity of a group of people. Biases can be entered if you choose to. I've seen things like people taking pictures of a certain zip code and seeing a pickup truck and automatically classifying it in some way to say that this group of people is certain, you know, this is the class for this group of people, which usually aren't favorable. Why is it we only put in the unfavorable biases? Because it sells yeah it sells um unfortunate unless you're a reader meaning you know you have a elevated unless you have elevated your thinking and also have a consciousness to be connected sometimes it's almost like gravity. We kind of gravitate toward things that are negative and feed the negativity. That's the unfortunate thing. People want juicy juice,

you know, they want juice. And if you start drinking juice, it's really hard to get out of the well. You have to kind of fight against it, and it's not always easy. Yeah, you really have to. And the importance of that leads to what we know in networking is creating new bridges of information, talking to people who are different from you, reading things that may go against your grind. Just even see if there's any credibility to the text, right? It's very important. It's very important getting to travel to meet other people from different places so that you can understand how others may think about the world. I am more of a reader than a traveler, but yes, I totally agree. I remember speaking to Bailey one year, many years ago, and her goal was to read one book from every section of the Dewey Decimal System.

And I just thought that was kind of cool. You know, it was like, all right, that's a little odd, but yeah, that would give you a cross section. And I think that was her goal every year or something. I didn't do it, but I was kind of, I looked at it. I'm like, yeah, okay. All right. Yeah. Yeah.

So I try to stretch myself in that way. And I'm going to go back to the technology a little bit. I work on some convolutional neural networks where they look at an image and respond to it. They have no memory. Every image is a brand new world to them. But you've looked at remembering in AI. And the The acronym I was trying to come up with earlier was LSTM, long, short. Oh, the long, long term. Yes, yes, yes. Long, short term memory. Yes, yes.

So, yeah. So what else? Yeah. So that's what you work in? No, that's one of the things. I wondered, you look at remembering in artificial intelligence. Is that the sort of thing you do or are there other ways of remembering? Well, okay, so you're referring to my TED Talk, right?

Yes. That's the information you obtained from the TED Talk. So, you're so funny. This came from my internet stalking of you, yes. Yeah, yeah. So in the early 90s, I studied the brain and cognition for understanding damage. And if you can recall, the AI went into what we know as its second winter, right? And from there, things kind of died down.

You didn't hear much about AI. Lisp went away, which is what I programmed in. But through that work, I've never used, I'm not even, I can't probably even say the name, the CNNs. I've never used that type of neural network, but one of my students last year wrote her dissertation in that area. And it was more aligned with autonomous vehicles. utilize short term, but what they had to do is store it away. Short term, like short burst, like short amount of memory that could be used to feed the neural network, but it wasn't long term because long term would of course eat up. It's very costly. It just takes so much RAM at every step. Yeah, exactly. Right. And so in my, in my discussion, I, uh, through my Ted talk, I was admonishing individuals that memory, uh, the computer uses memory and has to use

memory. It's critical to any type of AI, every, anything Carlo, Bayesian networks, Bayesian algorithms. It uses a level of memory. Even I mentioned the Q-learning up front. You asked me what was my favorite algorithm. And I really like Q-learning because it doesn't, it automatically update. It has this update phase, right? But the update phase considers as well, some level of memory. And so in my TED talk, I was saying, you know, if AI understands the value of remembering and remembering, remembering so that it can move ahead it's very important for human intelligence to not continue to blaze through life without looking back a little bit you have to look back on your both uh this embedded information as well as past experiences that may impact you in the long run.

That's what I was showing, the correlation, because even in AI, it doesn't differentiate between bad or good. All memories are used. You learn so that you can learn about your environment over that course. So there's not a differentiation for like bad memory, as we know, like we get traumatized by the things that happen to us. Well, when we train, there's no sense of trauma. It's just learning. Only I could learn that easily. We all, right? It takes a long time. Exactly. That's yes. Yes. I think we are all in that cycle of continuing to learn that, you know what, it's not bad to look back. It may be a little painful, but I need it so that we can move forward. And AI is definitely considering that. It has to remember in order for anyone or even for AI to learn, there must be some level of memory, even if it is temporary memory. Okay.

And now I want to go back to the past since we're talking about memory. That was my smoothest transition ever. You were a senior software engineer at Honeywell and a principal engineer at General Dynamics and other systems and software engineering jobs. And then you left industry to get a PhD in computer science. How long had you been in industry? Sort of. Oh, sort of. Okay. So tell me the story. No, because the steps were, yes, I did leave industry, but I didn't leave it to get a PhD.

I started the PhD while I was in industry. I completed the PhD out of industry. But I had been in industry for 20 years, 20 years exactly, when I stepped away to commit my time to focus on the PhD. And it was a hard reboot. It wasn't like I just volunteered to step away. I call it a hard reboot. And it is in computer science, isn't it? Computer science, undergrad, graduate, and a PhD in computer science. Computer science all the way. All the way. What does your PhD get you that you didn't have before? I mean, you did software. You were a senior engineer.

I was principal. Principal is a big deal. As a principal engineer, I was over $650 million projects. I didn't need a PhD for that. PhD allowed me to think about problem spaces that I never would have even thought about. It gives me options. It gives me the ability to be a thought leader, which allows me to think freely. And thinking is a big deal because our thoughts are bombarded with technology and the things we love, which can be very disruptive. And it gives me an ability to see, to see over different domains, experiences, and also ways of thinking about things that others may not have the privy to actually contextualize and be able to say, this is how you can create value either for your customer,

this is how you can create value for the world, this is how you can create value for your community based on what's in your hand. So the PhD really gives me options and places me at a, gives me the credentials to say that I am a thought leader in this space because I've worked really, really hard for a long time. And that coupled with my experience, it's just like icing on the cake. I'm really able, like for instance, when I go to conferences, I was just at Grace Hopper this week and the week prior in New York. So when I go to these conferences, I'm able to look at products

of today, but I know what I've been around since. I'm not going to tell you how old I am. I know you're going to ask that, but I've been around for some time. And with that, I'm able to look at different problems and basically understand what's in the background or what's behind their interface. And also to understand what it takes to actually really give people the insight that they're looking for. Use cases. People are looking for use cases in AI right now because a lot of people know the word AI, but they aren't really understanding how they actually can aggregate through their data and deliver insights. So that's what this does. That's what the PhD does for me.

Cool. Do you wish you'd gotten a PhD right after college or was the time in industry critical? No. Matter of fact, I wrote down when I was 20 something years old that I would get a PhD. And I wrote also that I would work in industry first. And the reason for some people, you know, I don't knock people's path, right? For some, I wish you well. But for me, I want, I'm talking about my personal conviction was if I said that I can meet some requirement at 99, 99%, I didn't want to read that this book said that it could meet the spec, but I wanted to prove it. In order for me to do that, I needed to have done some work, which I have done. And most of the time when they tell you something like that, it's the extreme case. You'll never,

ever get that. So I thought it was very important to get a lot of industry experience. Plus, the industry was really good and giving you skills that were beyond what you studied, which I did everything from business development. Right. I mean, you're doing, you know, depending on where you are. When I was at Weston House, oh, my gosh, we entered a nuclear project and it was five years behind and we had just got the contract because somebody else had messed up their contract and we got it by default and so we went into a mandatory overtime um and and we stayed there for at least three years just working on a project that mean that you were the database person. You were the clear case. Clear case was just coming in. You were the clear case trainer of your whole team. Sorry. You were the SQL. You were wearing like four hats just to get things done. And you were happy to be at the party. So industry affords you that kind of like flex. Right. And you're able to do some things, you know, depending on where you are. And my only thing, you know, my thing I struggle with in industry was management. People managing people is hard. It's really hard. It's really, really hard. And from my mind, it was really taxing. But it gives you the opportunity

to gain a lot of skills, which is pretty cool. People have said that I am such an engineer, I can barely talk about theoretical things without wanting to apply them. And so the few times I've thought about getting a PhD, I look into it, and then I look at the classes, and then I start thinking about how to apply them, and then I think about getting a book, and maybe I get a book, and maybe I read some, and then it's all about the application, and I lose interest in the PhD because I now have a new toy to play with. Was that a problem for you, or did you switch over?

No, it wasn't a problem because, I mean, now mind you, I wasn't trying to stay in school a long time. And most of my research was not directly with a professor. That was one thing. I created a dream team, like literally. I needed a mathematician. I had two. I needed a social environmental, like literally. I needed a mathematician. I had two. I needed a social environmental, social scientist. I had one. I needed some students who could read some data. I had those. So I had to strategically place myself and go and find these individuals. So I went to Dr. Richard Tavia and said,

hey, I need some mathematicians. At Arizona State, he said, oh, Carlos is there. Carlos Castillo Chavez. He's there. He's like the Obama, get the gold medal for a Mo'Bano, this big award, right? Carlos is so generous. He gave me two mathematicians because I was looking at contagion effects. So when you talk about social influence, it's the same or similar to like in biology, the contagion effect. So then the next

thing was, okay, I needed to understand about attitudes and environmental issues. And that led me to the Center for Policy Informatics. So going back to your point, I am interested in problems that help to solve something. Meaning even the research that I do, I'm not trying to build a product. I'm not trying to build an application, but I do want the results to be of use. So I'm not interested in creating a math formula that will only go in a book. That's not my interest at all. I knew you were an engineer. Yes, I am the girl, since I was a little girl, who would jump off the roof and then realize that

my leg was broke. So I would try. I would try. So that was me. Got to try all the options. Yeah. Try your options. And so now you do consulting in AI? What is that like? It's really, it's thought leadership, giving people insight into their use cases, giving people insight into ways that they can actually use their data. And it's very interesting. And I get calls from people that that's pretty much the calls that I get. We have this data set and we have this problem and this is how we actually use our processes right now. This is what we do in our company.

How can you help? What are some suggestions? What's available? What's possible? What kind of tools do we use? Do we need Python? Do we need R? Do we need SPSS? Whatever. People are in different places. Oh, should we just buy AWS or should we buy IBM? That type of thing. And then do you usually say you don't have nearly enough data and it isn't clean enough? That always seems to be my answer with AI projects. No, no, no, no, no, no. Everybody has data. Usually somebody, you know, the client, most clients have data. They have data.

They may not have, yeah, they have data. Most of these folks, there is so much data. 80% of all of my data is in unstructured text. 80%. 80% of all data that exists is unstructured text. It's unstructured. We are striving to put structure around it to give us insight. And most of the data that exists is 80% unstructured. That's a lot. That seems like a problem. I always use commas. It is a problem. Yes. And you also are CEO and founder of a non-profit. Is Strong Ties a non-profit? Strong Ties is a non-profit. And so the next question, I don't even know the next question, but I would like to tell you how that name came to be. May I tell you?

Cool. Yes. Okay. So everyone always asks me this. As a matter of fact, when I named Strong Ties, someone said, why didn't you name it something similar to like black women in da, da, da, are working something, something. I was inspired by the work of Grandavetter, which is Strong and Weak Ties. So I was studying network theory and the importance of the work of Grandavetter is that strong in a network, you have weak ties, which is you. That's what you are to me right now. And the hopes are the weak ties have been proven to deliver or be able, that's where you create new bridges of information. That's where you may even get leads to sales.

There isn't a familiarity that people are assumptions about people who are in your weak ties. There are nodes out there that you can connect to. You connected to me through Twitter, you said. And so you're my weak tie. The hopes are that one day you'll be my strong ties. You'll become my friend. So Grand Inventor came up with this model of triads. A friend of a friend is a friend. A friend of an enemy is usually an enemy. And the like, right? And we know this. We may not have known that name, but we know it by like, and what do you call this thing? Facebook and most of the networks that we have, LinkedIn, they use this model.

They use this, you know, this associative list to determine the networks and determine how close people are and determine, you know, your second degree of closeness. Right. And so that's where Strong Ties came from. That's where the name came from. And so I assume it has to do with building Strong Ties. What does your nonprofit do? So I work with youth, primarily 13 to 18 year olds. And now I'm working with undergraduates as well. But my work is primarily with youth, and I introduce and engage them in STEAM education. The art is big. And what that looks like, and the reason I do that is to, of course, broaden participation in the field of STEM through STEAM and so forth. And I use culturally relevant tools.

For instance, I use hip hop. It's one of the big things that my flagship program called Turn Up for STEAM. We use hip hop integration with coding and music composition and art, computational art. And then another program that I have is called STEAM and Global Citizenship Program. And that is where I can introduce, that's an all girls program.

The other program is for boys and girls. And that program, I'm able to introduce girls to everything from storytelling, animation, and as well to what I'm focusing on now is allowing them to create experiences based on the stories that they create. And we're focusing on the sustainable development goals as well as AI.

So I'm introducing them to the first phase of what I know is in AI, which are to take a data set and try to make some sense of it. And first by creating, giving them a sense of a vector space. Okay. There's so much there. Let's go all the way back to the beginning and do STEM versus STEAM and say what those acronyms mean so science technology engineering and mathematics yes and art a is for art yes that's for the steam yeah art access art is the lens art is the the actual lens if you think of a painting, the painter paints, and it gives the person who's looking at that then the artist in her case, she's the artist and she writes very openly and open-ended that allows anyone who's reading that text and reading that artwork, those books, to actually envision and let their minds imagine what's going to happen to Beloved in that case. And so that's the beauty of the art. It really opens up. It's the window into

a lot of things. And so we use art a lot because it allows kids to open up, to explore, to express, to be creative in a way that will allow them to see STEAM without having to worry about the linear way of writing on a chalkboard. That's very boring to me. You get there, but that's, to me, not the best approach to introduce students to a subject that oftentimes they don't get enough of in school. I understand that. I've always been a little skeptical of the STEAM movement because I am so in favor of STEM, but it's really hard to do science, but it's not that hard to do art and science. That's correct. You want to talk about plants or biology or chemistry, it's far clearer to do that with respect to art.

Yeah. And you think about like, I think in the most easiest way, I would use my dad, for instance. My father was a laborer. He retired from the railroad. When he retired, he started to build furniture. He was not a mathematician, but are experiencing science, technology, engineering, and mathematics, but we don't know what that is. Definitely when you, this mic that I'm speaking in includes vibration, it includes waveforms, it includes a whole bunch of things that have a name on it that's in STEAM. And everybody has some level of mic on their earbuds, earbuds, earphones, and so forth. But they interface and interact with these mediums. And it's a good launch pad for students to actually enter that space. And I heard this gentleman, I think his name is Marcus Roberts. And he said, you take what a

student knows to introduce them to what they don't know. And that's the easiest way that- Oh, yeah. That's great. don't know. And that's the easiest way that, you know, that's the value of the heart. Every kid that I know, you know, they like, a lot of kids like hip hop. If they don't like hip hop, you can give them cardboard tunes. It doesn't matter. Whatever they like, you give that to them and then you use that and say, hey, let me show you the science behind that. I'll show you the waveforms. I'll show you the sound. I will show you even the patterns in the composition. I will show you this thing and I'll help you to program that. Okay, let's look at a game. Most kids love games.

If it's from lollipop playing to actual shoot-em-up games, the actual warrior games that they play online. Or like my husband, he plays pool all day, virtual pool. So in that case, then, yeah, so you take what they know and you put it in their hands. I think that's easier. That's much more easier to digest. Like for instance, yesterday I started my mentoring program, the STEAM and Global Citizenship Program. I had them to play just type out five and then brought them to a list and then led them on. Okay, now add an X in the list and then let's create a pattern, a diagonal pattern. That was so easy to do, right? It was much easier to do. And so those

kinds of things, it's just, you know, using simple things that people already know to introduce them to a subject instead of making it very complicated. Okay. I want to introduce you to vectorization. I don't think so. The kid would sink in their seat, right? But if you combine it with a pen or a drawing tool. Yeah. Before you can learn how to make some cool beats, we've got to discuss Fourier transforms. Yes. So yeah, try that. Let's see. I don't think it will work very well. So yeah. So that's what I do. That's what I do. And I do that all over the world. How do you balance machine learning, AI consulting with running what sounds like a pretty big outreach program?

It is. What I am doing, aggressively doing, is combining my research in AI with the work that I do with the students. That's what I am endeavoring to do now. So it's pretty stretching me thin to actually do both right now. However, both are important because I can't afford to not be in AI because it's so few of us doing this work. And then I can't afford for our kids to not appreciate AI. So my charter right now is that's why I'm introducing students to AI and data science so that I can figure out a way to actually have, when I research, actually show the programming portion, what I'm doing on the ground, as well as meaning the proof in the pudding, as well as the models that are working. How are you going to combine hip hop and AI? Oh, I won't. I'm not even thinking about it.

Well, let me tell you this. actually came up with the first AI machine learning to recognize and come up with a original hip hop composition. Yoshio Bengio received the Turing Award this year, as well as Joffrey Hinton and another gentleman. But yeah, it's possible. It could definitely happen. So the way that that's working is they're looking at different patterns in hip hop to actually come up with new compositions without consulting an artist, which is pretty cool. So it can happen. I've seen some AI generated poetry. Some of it these nice things, which hip hop is a lot of. But when you're thinking about the data set and you're using as input existing data sets, then you can look at different patterns, right? Because we know it looks like it's a waveform, right? And because if you think about hip hop, hip hop bars a lot from African rhythms, right? When you base it down.

So there are some base rhythmic, there's some base patterns that are already embedded within hip hop that people can extract out and use. Now I want to take all of this. Look up here. I'm wondering, I don't know if Yoshia Bengio, if Dr. Yoshia Bengio's, I'm pretty sure it's out there because it's well known that he came up with this hip hop. I call him Dr. Yoshia B. And so, yeah. And the students you involve, is it, tell me about them in general.

Yeah, yeah. So I am particularly interested in reaching out to underserved and underrepresented students. And what that looks like is traditionally African-American or Hispanic students and Native students. And what that looks like is traditionally African-American or Hispanic students and Native students. Native students, it's not a lot to pull from, but we're working on that right now, trying to make inroads with the Natives. But yeah, that's the students that I reach out to, students who may not have the resources or may not have the representation in their community. Okay, this is going to be a very stupid question. And I'm mostly asking it because some people have asked me and I want better answers than I've been given.

Why? Why bother? I mean, they go to school. Why bother? Oh, schools right now don't have programming, STEAM programming or STEM programming for that matter. I think most of the data that I collect, almost 80% of these students don't have this in their schools. So many schools are just still struggling to get pens and pencils. And it's very important that we invest in our 21st century jobs. These are students that in five or six years, they're going to be hitting the job market.

If we haven't invested in them to actually study the fields of STEAM and STEM, then the U.S. would be in a very vulnerable state right now, according to, I think it's UC, what is it, UCD? The indices that talks about how advanced we are in mathematics and how advanced we are in science. America is lagging behind so many countries in these areas, which is correlated with our economic growth. So it behooves us to invest in all hands on deck, invest in students to actually go into these fields.

Cool. That's a different answer than I give. So economic answers are a good one. Yeah. I mean, I love the area, of course. And also another thing I think about, if we're creating technologies that will probably be used by everybody, why not let everybody, some portions of everybody, be represented in creating that technology? That's a big one as well. That's the one I usually go for. Yeah. Story about if you want to sell things to people, you really have to have them there because they're going to want different things.

Yeah, definitely. Loretta, I've enjoyed talking with you so much but we are about out of time so I guess it's time for me to ask you if there are any thoughts you'd like to leave us with yeah, I would like to leave your listening audience with engage as much as we're advancing through technology in particular artificial intelligence, augmentation,

all things AI, representation, augmented VR, XR, MR, and how we see it in spatial computing or through our lens that we work with these devices, engage with that, these technologies, so that we stay connected as a human. We allow ourselves to not become augmented and not become artificial in the process of the website, strongtiesaz.org. That will be in the show notes, of course. And thank you so much for being with us. Thank you for having me. And thank you.

Thank you, everyone. Thanks, Loretta. Thank you. Thank you also to our Patreon supporters for their support and Loretta's mic. Special thanks to IWL, who are currently supporting us at a corporate level. Yay. Thank you also to Christopher for producing and co-hosting and thank you for listening. You can always contact us at showandembedded.fm or hit the contact link on Embedded FM. And now a quote to leave you with from Toni Morrison's beloved. Freeing yourself was one thing.

Claiming ownership of that freed self was another. Embedded is an independently produced radio show that focuses on the many aspects of engineering. It is a production of Logical Elegance, an embedded software consulting company in California. If there are advertisements in the show, we did not put them there and do not receive money from them. At this time, our sponsors are Logical Elegance and listeners like you.