AI for Kids

How AI Learns with Cat Pictures But Still Doesn’t Know What a Cat Is (Middle+)

• Amber Ivey (AI) • Season 2 • Episode 19

Share episode

Send us a text

Ever wondered why AI can be simultaneously brilliant and bewilderingly clueless? Reed Coke, Director of Engineering at KUNGFU.AI, offers a beautifully simple explanation: AI is like singing along to songs in a language you don't know. You recognize patterns and can predict what comes next without truly understanding the meaning.

Reed brings his unique background as a linguist-turned-AI engineer to explain natural language processing in terms kids (and adults) can grasp. Growing up bilingual in Dutch and English sparked his passion for languages, eventually leading him to discover programming as a powerful tool for studying human communication. This intersection became his specialty - building AI systems that interact with language.

The conversation explores fascinating territory, from Reed's work on an AI system that could detect cancer risks five years before diagnosis to the amusing challenges of teaching computers to understand context (like distinguishing complaints about counterfeit products from reviews of Halloween "fake doctor" costumes). Reed breaks down complex concepts like supervised learning without sacrificing accuracy, making them accessible for young listeners.

For aspiring young technologists, Reed offers practical advice on getting started with AI through resources like Code Academy, Code Combat, kidspython.com, and Black Girls Hack, emphasizing that community-based learning through classes and camps often provides the best support. Most importantly, he encourages approaching AI development with empathy - considering who created the training data, what experiences it represents, and how systems will affect real people.

Ready to help your child understand the technology shaping their future? Subscribe to AI for Kids and join us for conversations that make complex concepts approachable while inspiring curiosity and critical thinking about the tools transforming our world.

Ready to dive deeper into the world of AI? Subscri

Support the show

Help us become the #1 podcast for AI for Kids.

Buy our new book "Let Kids Be Kids, Not Robots!: Embracing Childhood in an Age of AI"

Social Media & Contact:

Listen, rate, and subscribe! Stay updated with our latest episodes by subscribing to AI for Kids on your favorite podcast platform.

Like our content, subscribe or feel free to donate to our Patreon here: patreon.com/AiDigiTales...

Amber Ivey (AI):

Welcome to the AI for Kids podcast, where playtime, learning and creating collide bit by bit. Ever wonder how your phone recognizes your face. How does a game learn to get harder as you get better? This is AI. This podcast is designed for kids like you and your human parents, making the complex world of AI easy to understand and, most importantly, fun. So are you ready to unlock the mysteries of artificial intelligence? Subscribe and join us on AI for Kids. Hi everyone, welcome back to AI for Kids. Today we have an amazing guest, Reed Coke, director of Engineering at KungFuai. Reed, can you tell us about yourself and what you do at KungFuai?

Reed Coke:

Thanks for having me. I'm from Austin, Texas. I also spend time living in Amsterdam. In other words, up through a fourth grade. I've been doing AI and, specifically, machine learning for more than a decade now, so I've gotten to see a lot of the different changes that have been happening, a lot of the recent changes, but there were also waves before and beyond that.

Reed Coke:

I really love teaching. I like to play soccer. Before and beyond that, I really love teaching. I like to play soccer. I like to bake desserts. I like to solve puzzles. Yeah, that's really me in a nutshell, I think. As far as at work at Kung Fu, we are a consulting company, so we don't have products or run a website. We actually basically just go help other groups who want to build things like that, and they'll often come to us with maybe an idea of an AI project that they think they want to do but don't know how, or even just knowing they probably should start doing some AI but don't even know where to begin. We help them on code writing, data gathering, model training, all these kinds of things, but also just like the planning and the strategy behind where to get started, how to get started, what to think about beforehand what to roll with as we go, all that kind of stuff.

Amber Ivey (AI):

Sounds like a fun job one, and I've had a brief encounter with Amsterdam. My father was in the army and, on our way to Germany.

Reed Coke:

We had a layover in Amsterdam, I'm sure it was amazing growing up there as a kid.

Amber Ivey (AI):

Were your parents military or worked there, or why were you there?

Reed Coke:

Yeah, my dad's a jazz musician. He had a band, came to Austin to perform, played with them a little and decided he wanted to keep doing that. That's so cool. No kidding, it was a very cool opportunity. I have tremendous nostalgia for the Amsterdam airport, so I'm glad you got to experience it.

Amber Ivey (AI):

Beautiful airport. I was younger, but I remember that airport vividly. Of course I'll get back there as an adult. What was your favorite subject when you were a kid? Did you always enjoy learning about computers and technology?

Reed Coke:

It's interesting. I took a roundabout path. My favorite thing was, and it still is, foreign languages. I grew up speaking Dutch and English. When I moved back to the US, I got to take Spanish in school and later I took a bunch of French in school and eventually I got into linguistics the study of language in general as opposed to any specific language. What are patterns across languages, and this kind of stuff is really interesting to me. What got me into programming was I realized that I could use computers to learn more about human language.

Reed Coke:

So, I ended up specializing in a type of AI called natural language processing or NLP, which I'll probably accidentally say later in our conversation, so we'll define it now which is a lot of Siri-like technology or even ChatGPT or things like this, that kind of interact with human language to do stuff.

Amber Ivey (AI):

That's such a cool connection to interact with human language, to do stuff, that's such a cool connection.

Reed Coke:

Yeah, I originally was studying language teaching and learning. In college. I did neuroscience, psychology education, linguistics. It turns out if you want to learn about language, you need to study people. Studying people is hard and expensive, especially language learning. You usually need to study babies. The laptop is cheaper so you can do more experiments quickly and learn fast on the digital side.

Amber Ivey (AI):

That's hilarious but true. If you could build a cool AI tool or an AI tool to solve a problem, what would it do and why?

Reed Coke:

I would love an AI system that could build a personalized healthy meal plan. I don't have any problem shopping. I don't have any problem cooking. I enjoy both of those. I do not like deciding what I need to make and figuring out the healthiest choice is going to be. That is going to be really tasty and nice, so I would love to have a system take care of that.

Amber Ivey (AI):

I'm with you on that, because I read all the labels and I'm tired of knowing the sugar intake on all my food. I would love for AI to let me know these are things you're going to buy, because it fits in your goals of health this week for your food, and I would love that. If you ever build one, let me know I'll be a customer.

Reed Coke:

Definitely. I'm sure someone's going to.

Amber Ivey (AI):

At this point. If you're listening, please do it. Kids, it can also be you.

Reed Coke:

And I know you've worked on hundreds of AI projects, so can you tell us about one of the coolest projects you've worked on? Yeah, absolutely. A few years ago, one of my earlier projects when I joined Kung Fu AI, I got to work on an early detection system for cancer. This project would look at x-rays of a patient in a hospital and we'd use AI to look at pieces of this x-ray that sort of indicated that there might be a risk of cancer in a particular area. Several systems out there can look at an x-ray of someone who currently has cancer and sort of point to where it is or how severe it might be or things like that.

Reed Coke:

We were looking at x-rays of mostly people diagnosed in the next five years. That was exciting because you always want to catch it early. You can treat it more effectively, more easily, more cheaply. It was incredible to work on something important for making the world a better place. The puzzle solving side was just hard to do well, and it was cool to work with a team of programmers and doctors. I learned a ton about the problem and solving things like that.

Amber Ivey (AI):

That's amazing. When you were first talking about it, I thought it was to identify people who already had it. The fact that you are using it to predict and help people catch it early, like that, is one of the biggest things I've heard from people. Catching it early can help increase the likelihood of treating it, so that's amazing that you've done that work. You've also taught AI to kids. What was the funnest part about teaching AI to young people? What was the funnest part about teaching AI to young people?

Reed Coke:

That's a great question. I got to do that for about five years and I really focused on teaching Python and AI, and I don't have a single this day. This thing happened and that was great, but it's more of. There's a moment that a student will realize that this is just something they can do now. This is a tool in their toolkit, this is the power that they have that they can use whenever they want and it's so cool to see like just these doors open up in their eyes all of a sudden and, honestly, a lot of times it's only maybe two months of practice every week to get to this point. It happens really fast, which is great for me because it means I get to see it all the time.

Amber Ivey (AI):

It's great for the kids listening, because if it's something you could do in two months versus 10 years, everyone should think about trying that. Are there tools for kids in middle and high school to introduce themselves to things like Python?

Reed Coke:

There's getting started with Python and getting started with AI. I think there are increasingly ways now to get started with AI and skip the Python part or focus on the different side of it that relies on Python specifically less. However, I think the side where you also learn Python and have this new superpower you can use whenever you want is probably the most rewarding. So just to show my hand, I'm very much on that. Okay, great AI. Python will help you learn AI, but Python also will help you do all sorts of other stuff. I fell in love with Python a long time before I fell in love with AI. For me, there are a lot of cool websites out there that are free, or at least the beginning is free. There's a big one called Code Academy, perfectly kid friendly. There's this website called Code Combat, which is like an adventure game, but you're writing code to control your character rather than just clicking or moving around. I also have occasionally come across people who've told me that they really liked kidspythoncom.

Amber Ivey (AI):

I haven't heard about that one and the name is so simple and easy to ask. The name makes sense.

Reed Coke:

Yeah, I gotta believe it. So that's one way, that's the. I want to just go try this myself in my bedroom or at the dinner table. I also think there are powerful community-based things to do. I was listening to other guests on this show and I heard about Black Girls Hack. It feels like an awesome example of a group excited about this and if you're interested, go check out a group like that. I'm sure that they would all be willing to help you, tell you how they learned, help you learn. I really I think that's the best way. But I also recognize sometimes it could be hard to find Other than that.

Reed Coke:

Take technology classes. At school I didn't do that. I thought I was too cool. When I had the chance, I later found out I love this and I could have been doing it a lot longer. Also, more summer camps. Universities have a lot of community outreach days. I've gotten to be a part of some of these days at the University of Texas and Austin and then also University of Michigan at Ann Arbor. They'll usually do it once a year. Bring a bunch of people in and do some cool activities and just see what this stuff is all about. So I think the community-based ones are powerful. It can be challenging, I think, to just do something on the internet by yourself and really stick with it. So I think having a community helps, but I think there's options for everybody, depending on what you like.

Amber Ivey (AI):

I definitely agree with you, like the ability to have a community and all those resources you mentioned. We will make sure to drop those in the show notes to make sure you work with your parents to check those out. I want to dig into this world that you live in, so, which is natural language processing. Can you explain how computers understand words and how that makes things like Siri and chat, gbc, work?

Reed Coke:

Sure I would be happy to. My dad is a musician. Music has always been a big part of my life. However, my dad is a jazz musician, so very often he's listening to songs and languages that I do not know and he does not know. That has never stopped me from singing along, and a lot of people have had an experience like that too.

Reed Coke:

Ai that deals with language in a lot of ways is like that. Let's say, I listen to every Portuguese Bossa Nova song. I don't speak Portuguese, I just try to sing along and get better at making sounds. As I listen more I might start to recognize patterns. These three words in this song are also in this other song, so if I hear two of them I might be able to guess that third one is coming, because I've heard it in different places. Sometimes it's ba-da and sometimes it's ba-da-da-da-dee. There's some structure. You can mix and match a bit the distinction compared to a person knowing language. If you look at a baby learning language, they have this incredible ability to very quickly map a word or really, for them at that point, probably a sound to something physical in the world around them, like mama and dad.

Reed Coke:

Yeah, Computers don't do that. They're not making that connection to the physical world in the way that people do. It's just yeah. I've heard these sounds or seen these letters together this many times, so the connection isn't quite there. Yeah, they're just learning to say the sounds back, even though they don't know what the sounds mean. Even something like Chachi BT, which is really impressive but isn't necessarily, in my opinion, any more intelligent in a human-like way compared to systems before. It's just something that has read a gigantic collection of language and pictures and it's just really good at singing the next line, Just like me listening to K-pop now.

Amber Ivey (AI):

Thank you for explaining that in a way that's so easy to understand. Hopefully parents are also listening so they can understand. I love that example. Thank you for bringing in music and language into explaining natural language processing in a way kids can resonate with.

Reed Coke:

Just like me imitating the bossa nova. It's always going to be better at the chorus than the verse, and that will probably come up later in our conversation, because it's heard the chorus that many more times.

Amber Ivey (AI):

I definitely want to dig into that. But before we dig into it, are you ready?

Reed Coke:

to play some trivia, absolutely.

Amber Ivey (AI):

Okay, we're going to play tech trivia, AI trivia. I'll ask you some fun questions about AI and technology and you'll try to answer as quickly as you can. Are you ready?

Reed Coke:

I'm ready.

Amber Ivey (AI):

Which AI model is designed to understand and generate human-like texts?

Reed Coke:

A generative, pre-trained transformer. Generative AI has a specific technical meaning. There are AI models that basically make decisions like categorize hey, is that a cat? Is that a dog? Is this Amber's face? Is this Reed's face? Those are classification models or discriminative AI, because they're making a decision. Generative AI is the opposite creating something. They're like building an output and not just sorting an output. So this is where you get systems like Chachi. Bt or Gemini is now a product that has a Google umbrella that has some of this, but also things like Dolly or what's it called Mid Journey, Some of the image generations. Also, generative AI.

Reed Coke:

Generative AI has been around for a long time, but not in the way that it is taking things by storm right now. Pre-trained is what I was alluding to in the music example. It has already heard so much music and now you can give it the first line of a song and it will try to finish it based on all the music it's already listened to. The reason it's called pre-trained and not like trained is that you can continue teaching it about new music as a scientist. You can take what it knows so far and then feed it more or tune it to care more specifically about I don't know, heavy metal instead of death metal or instead of other types of metal. You can get these really fine grained distinctions.

Reed Coke:

So, that's a pre-trained idea. You're not starting from scratch. The transformer is just an F word. That's a specific way of setting up a neural network. People have been setting neural networks for 60 years. The transformer architecture is one of the most popular, effective ways to train a nuanced system on a huge body of music in this case, or data more generally.

Amber Ivey (AI):

I see why you're such a great teacher, because you're able just to break these things down in such an easy way. Thank you for doing that. Are you ready for the last question of our trivia?

Reed Coke:

Yes, absolutely.

Amber Ivey (AI):

What is a popular AI-powered voice assistant used in Samsung devices? I tend to lean iPhone.

Reed Coke:

If I'm not mistaken, Samsung is Cortana. Is that right?

Amber Ivey (AI):

So that's not.

Reed Coke:

Microsoft.

Amber Ivey (AI):

I'm thinking Bixby.

Reed Coke:

Sure Russia got me through a lot, oh yeah.

Amber Ivey (AI):

So how can computers be smart and silly at the same time?

Reed Coke:

Yeah, this is the thing I like about AI they do it one way and people do it another way.

Reed Coke:

Computers don't think like people. There's a philosophical difference. I have a body. I live in the world, everything I think and do connects to these ideas. For the most part, computers just have bits and bytes and ones and zeros, ultimately on the inside, and they don't have this deep understanding of the way things connect. A different way of thinking about that same thing is when a computer is reading a book it just sees the letters on the page. When I'm a baby listening to my parents read a story, I'm probably watching my parents' face and listening to the tone of their voice, regardless of what they're even saying, so there's something to be said for it.

Reed Coke:

There's so many more layers of information that people have learned to process over the course of their life and we're just starting to figure out how to give rich teaching materials to a computer. As a result, they're really good at some things that we can represent well and then really lacking in other areas. When I read the word dog, I think about my family pet growing up and I think of all the other dogs that I've met and, more importantly, I think of all the other dogs I have pet during my lifetime. But when a computer reads the word dog, it reads the letters D-O-G, it recalls all the other times it has seen what I'm going to call the letter pattern D-O-G and what other letter patterns it has seen close to D-O-G, and it just it doesn't have this physical connection and so, as a result, it can do incredible things, like D-O-G showed up in these 4.5 trillion books. I'm never going to be able to do that, but there's so many things I do not even me realizing consciously that I can do them that are just fully automatic because I'm a human. I do all these things without thinking.

Reed Coke:

I worked on a project that highlighted this really well In a previous job. I needed to make an automatic system that would read customer reviews and find customer reviews where someone said I got shipped a fake item. I paid for something real, I got shipped a fake, and so I trained a system to look at letters in these reviews and learn which sorts of words show up in fake product reviews. Mostly, it worked pretty well. There was one review that I spent the whole summer trying to get it to handle correctly and it just never did, which was someone sent an angry letter because they had ordered a fake a fake fruit, like a fake apple, yeah, and they were unhappy about the quality of the fake apple.

Amber Ivey (AI):

Ah, okay, but they wanted the fake.

Reed Coke:

They wanted a fake apple, they got a real fake apple. And, amusingly, another thing that got wrong a lot of the time was Halloween costumes. Okay, because a lot of the costumes will be called like fake doctor costume. So it got really mixed up. No person would think that's counterfeit going on, but computers just see the world in a different way.

Amber Ivey (AI):

That's such a good point and I know you agree with me on this. Ai sometimes feels like magic. How does this magic work? Yeah, absolutely.

Reed Coke:

There are a couple types of AI, specifically machine learning, the main engine inside AI. The biggest categories are unsupervised learning, supervised learning and reinforcement learning, but supervised learning is 95% of everything that's actually out there, so I'm just going to talk about that one. There are other things.

Amber Ivey (AI):

Can you do a brief one-liner on each, even the other two? 100% yes, Okay great.

Reed Coke:

We'll start with supervised learning. That's where we'd say here's a million pictures of cats and a million pictures of dogs. I'm going to give you a new picture. Tell me if it's a cat or a dog Unsupervised learning. You don't know which ones are cats and which are dogs ahead of time. Here's two million pictures. Try to split them in a way where it learns to separate out the cats and the dogs. And then reinforcement learning is a lot more like AI in the movies, constantly learning at every moment. In most common situations, you train a supervised model and that's it. It interacts with people, maybe, but it doesn't actually learn anything from those interactions, and if you want to have it change, you have to train it again. The reinforcement learning system is always learning on the fly.

Reed Coke:

Thank you for explaining that one and also the simplest one in a lot of ways. So for supervised learning, let's do the million cats, million dogs example. You get this data set, what's called a training data set. So you get your 2 million pictures and they tell you ahead of time these ones contain a cat, these ones contain a dog, and the idea is to train something that will then look at new pictures and identify cats or dogs. You could see how this might become facial recognition rather than animal detection.

Reed Coke:

The AI takes in all these information in the pictures, which is really just like the colors in each of the pixels in the picture, and it just uses a bunch of maths to figure out which patterns of colors or pixel values are most likely to show up in a cat picture and which different colors and patterns are most likely to show up in a dog picture cat picture, and which different colors and patterns are most likely to show up in a dog picture.

Reed Coke:

And ideally it wants to find things of oh, when I see this pattern, it's usually a cat and not usually a dog, or it's usually a dog and not usually a cat, and then it will latch onto those and when it gets a new, that's called training. So once it's learned those patterns, it's trained and then you can give it a new picture. It just looks for patterns based on the 2 million pictures. It's seen which 2 million pictures you give it, it's potentially going to learn very different things and behave very differently afterward, which is, I think, both the silly part and the smart part of the magic. The AI doesn't know what a dog is or what a cat is. All it knows is these are DOG pictures and these are CAT pictures, and it doesn't know that if you showed it a picture of an animal that's not a dog or a cat, it's not going to know that this is a picture of a mouse. It's probably going to say this is a weird looking dog.

Amber Ivey (AI):

Is that because it wasn't trained on those images? Is that because it wasn't?

Reed Coke:

trained on those images Exactly. If you had put images of mice into the training set, it could potentially learn to differentiate all three. Just because it can always find a dog in a picture doesn't mean it's going to know what to do if your dog won't stop sneezing, or how to stop your cat from destroying your furniture. Those are completely outside of its world. It's interesting because maybe not in the simple cat-dog example, but when you get to a more really impressive system like a chat, gpt or some of this other modern generative AI stuff that's out there now, it can become really hard for people to understand exactly what it can do and what it can't do, but might be able to convincingly look like it's doing Very true. So I think that's the thing I think about a lot in my job and also just in my personal life.

Amber Ivey (AI):

Yeah, because even when you were just talking through that and even just preparing for the interview, one of the things that struck me was the idea that it doesn't know. It doesn't know what a dog means. Even the thought of you saying dog and immediately me thinking about dogs I've had in my life, my current dog and what that looks like. Ai is not able to do that. I know we've gotten to it, but I want to ask a little bit more around this. Ai can copy what is taught, but it doesn't understand the real world like we do. Can you explain a little bit more of that and do you think I'm going to add an extra layer to that? A little bit more of that, and do you think I'm going to add an extra layer to that? Do you think it can ever understand?

Reed Coke:

the world like we do? That's a good question. To answer the first part of the question, where the limits are Today, the way most systems are built, it is very hard to know. Even for smaller scale systems. It depends so much on what's in their training data. For these supervised models and really all of them unsupervised models it depends what's in the data, even though you're not really training in exactly the same way Reinforcement learning models. It depends on what experiences they've had as they've been learning.

Reed Coke:

So what's tough about that is getting the right data is underappreciated because the more data you have, the better it's going to do when you give it a test and score it. But if you give it maybe not the most representative data or the cleanest data or the most correct data, it can learn weird things that you're not necessarily intending. Chat GPT is very interesting because in order to make it as impressive as it is, it had to be trained on. Very broadly speaking, think of it as the whole internet. That's about the scale of input we're talking about. If you need that much input, you don't have the luxury of finding just the parts that are going to give you the best performing system. You just throw it all in and hope for the best it's going to learn.

Reed Coke:

There's a lot out there on the internet and not all of it is good. It doesn't understand what's good and what's not. Even if people could agree on that, it's tough. So anytime I'm interacting with an AI system, I try to think about. I try to guess which because of my job and background. I can do it, but I try to guess what it was trained on and the ways in which what I'm doing aligns with what it has seen and what I'm doing does not align with what it's seeing. Understanding. It might not be able to do the best job there In terms of your second very interesting question will it ever? There is a hopeful note on that point, which is actually back to ChatGPT, one of the things I find most interesting about how they built that system In a lot of language-oriented models. Beforehand, they were basically trying to get it to predict the next word, like my singing along example. Right, they're trying to figure out what the next sound is going to be, what the next word, what the next line is, and if they get the exact words right, it gets points. If it's wrong, it loses points.

Reed Coke:

When I think about communication, it's not really about the words most of the time. That's fair. There are a lot of words that mean the same thing. Yes, yeah, all right, it all means the same thing. In my opinion, language is more about connecting than the words said, and with Chachi, it's one of the first big systems where, rather than training for the next correct word, they're training for, would a human say, the answer was useful. I think that's philosophically very interesting. I also think it's not a direct way of having the ai understand the world, but I do think it's a more interesting way. Language is also facial expressions, sound, touch and other things. It's a richer signal and that's why chat gbt has been impressive and made such a big leap forward. So there's some hope in that regard. But I don't think we should give every computer a robot body and let it walk around. I've seen enough movies to believe that might not be the right next step.

Amber Ivey (AI):

Yeah, the moment where you said things like touch, I'm like so does that mean the AI can feel through some sensors?

Reed Coke:

I don't know, but humans underappreciate how rich our world is and how incredible we are at understanding it. And anytime we give computers this like thin slice of what we're really experiencing, they're going to do weird stuff. The thin slice will go great and everything outside of it will be strange and misaligned.

Amber Ivey (AI):

That's a very good point, and with that I do want kids to start learning about AI. Can you talk about how they can start learning and get involved, or some fun activities to learn more about the technology?

Reed Coke:

It goes back for me to that split I mentioned earlier of you can take the programming path or you can take the non-programming path, and I think with the non-programming path there's all sorts of interesting things out there. One of the cool things about all these large language models and about ChachiBT is they can give you such different things depending on how you ask the question. And there's this whole field called prompt engineering. That is basically about learning how to best ask the question to get the answer you really want, or the best possible version of your answer, and that's all you need is to be able to log into any llm online and start learning how to do that, and I think that's going to be a very real skill set. It already is, but in the near future especially so. That's, I think, one path. A lot of the stuff is, for the moment, pretty accessible. You can try it out and if you come up with clever ways of doing your prompts asking, asking questions you could discover something that other people haven't stumbled onto.

Reed Coke:

So, that's super exciting. I'm also a strong believer in learning programming and digging into it deeply from that direction.

Amber Ivey (AI):

Thank you for sharing those things and the reminder about prompt engineering. There are times where I go online because someone else has discovered it and they're like oh, if you do this prompt, it gets to this outcome pretty regularly and that's helped me even in my understanding of how to prompt these tools. But I'm with you like early access. Getting in there and learning while we are early in this space is great. I want to shift a little bit to learn something more about you. We have a game called Two Truths and a dream. You'll tell us two true facts about your life and one dream you had as a kid and the goal is for myself as well as the kids who are listening to guess. Are you ready?

Reed Coke:

I am ready.

Amber Ivey (AI):

Okay, go for it. Fix them up and I'll try to figure it out.

Reed Coke:

I'll tell you three dreams that I had as a kid. Okay, I wanted to learn languages, I don't know. For work. I wanted to translate for the UN and be the CEO of the company. Okay, broad dream.

Amber Ivey (AI):

Okay, got it Okay. The first one was you wanted to be able to learn multiple languages. Second was you want to translate for the UN, which is also pretty cool. And third was you want to be a CEO of the company. So I'm going to go through each of them. Kids at home, think about the context, clues and things we learned about Reed through the interview. You want to learn multiple languages. I believe that one's true. You said like you speak Dutch. Yeah, you, of course, speak English. You also were learning Spanish in school and wanted other languages, so I'm going to hold that one.

Reed Coke:

I need to clarify the wording. What?

Amber Ivey (AI):

I said was.

Reed Coke:

I want to learn languages I don't know for work.

Amber Ivey (AI):

Thank you for the cue. So I'm definitely holding that one off with the clarification of those details. Still think that's true, but I need to analyze the for work part. Interpreter. Back to the language pieces. I'm going to hold that one CEO of a company. I think if you wanted to you could definitely do it, but I don't know if that is the one. This is harder than I thought. I'm just going to say in my probably terrible logic so kids use better logic than me. I'm going to say the dream is the CEO and the other two are actually facts about you.

Reed Coke:

That's a good guess. I'll run down the list quickly Learning languages I don't know for work. There have been times where I've been asked to build an AI system to learn about text in a language I don't know, and part of solving that problem was learning some about the language. So that has happened.

Amber Ivey (AI):

Nice.

Reed Coke:

That's pretty cool. It was fun. I had no idea if it was going to work, and it did. It was a blast.

Amber Ivey (AI):

The work part threw me off.

Reed Coke:

I thought that was going to be the dream, but I was like I've already come down the road, so let me not go back. So I'm going to jump to the CEO one. I did start my own consulting company a number of years ago and ran that company for about four years before joining full-time. The UN is just hard to get into and you've got to really know those languages. And around the time when I might've started going in that direction, I discovered programming and decided to take a different path instead.

Amber Ivey (AI):

That's a really cool way of learning more about you and your background. Thank you for sharing that Before we go. Do you have any advice for kids who want to learn more about AI and how to use it safely and responsibly?

Reed Coke:

Sure, AI safety is a tough and blooming question. This is what gets back to the chorus and verse idea. I, in these Portuguese bossa nova songs I keep referencing to, I can imitate the chorus pretty good because I've heard it about seven more times than any other verse. The verses are still hard and in a lot of the same ways, AI is really good at replicating the things that it sees most often and less good at replicating the things it it sees most often and less good at replicating the things it doesn't see as frequently. With my Chef, if it learned from a lot of people's meal plans and there's some unusual foods I really like it might not be recommending those that often If you have a hard time shopping for pants or shopping for shoes.

Reed Coke:

Ai might not make that all that easier for you if it's because there's something really unique and your feet are different sizes, or something that happens, but it doesn't happen all the time and that gets really serious if you have a rare medical condition. Ai in the hospital hasn't had many examples of someone like you go through it. It might not be doing the best job compared to a human being who understands that it needs to think about you differently, and not just in the context of the million songs that they've already listened to, but the doctor version. So back to the question of what can you do? I think that the single best solution is living with empathy, coding with empathy, always thinking about people. People are gonna be ideally.

Reed Coke:

You're building an AI because you want people to use it, so think about how they're gonna use it and what they're going to be. Ideally. You're building an AI because you want people to use it, so think about how they're going to use it and what they're going to need. Think about the data. Who made that data? What were they living? What's their life experience like? Who's taking the pictures of those cats and dogs? If the person taking the picture of the cats and dogs love to take pictures of cats like this, your AI is going to figure out really quickly.

Reed Coke:

The cats are all like vertical faced or horizontally faced creatures. So, true, these are all things that your system, just blindly, is going to learn because, like you said, it doesn't know, it just sees pixels. Empathy is important to understand what's in that data and therefore what the system will learn from it and also what kind of impact it's going to have once you build it. The other small piece of advice I have is if you are interested in experimenting with AI or even thinking about AI as a career, doing AI doesn't mean you can't do anything else.

Reed Coke:

I love language and I got into AI because I can combine it with language and do those things together. If you love baking or music or basketball or stamp collecting bugs whatever it is that you're excited about you can do that with AI together and get more deep into potentially two things that you love and in a way that, let's just say, half as many people will also be able to be experts in. So I think it helps you and I think it lets you do things that are just more exciting day to day. So I would suggest thinking about other things you love and how you can put them together.

Amber Ivey (AI):

Thank you so much for joining us today, reed, and thank you to all the listeners for tuning in. Don't forget to subscribe to the AI for Kids podcast and stay curious. Thank you for joining us as we explore the fascinating world of artificial intelligence. Don't keep this adventure to yourself. Download it, share it with your friends and let everyone else in on the fun. Subscribe wherever you get your podcasts, or on YouTube. See you next time on AI for Kids.

People on this episode

Head Shot

Amber Ivey (AI)

Host