We've always been on a quest for knowledge as people, and we've always wanted to understand ourselves better — how we think, how we make decisions, how we improve our world. And I think the origins of AI are part of that quest.
Today we're going back to the beginning to try to understand just what AI is and how it works. It's AI 101. Welcome back to a new episode of Raising Kids in the Age of AI, the podcast from aiEDU Studios in collaboration with Google.
I'm Dr. Aliza Pressman, developmental psychologist and host of the podcast Raising Good Humans.
I'm Alex Kotran, founder and CEO of aiEDU, a nonprofit that helps students get ready to live, work, and thrive in a world where AI is everywhere. Today in the pod, we have two AI experts and parents with us to help: Maya Kulycky, VP of strategy and operations for Google Research, and chief operating officer of Google DeepMind, Lila Ibrahim. They're going to help us see the shape of AI from its origins to its future, the best practices, and going over some of the different types of tools that are available.
This is just what I need, honestly. But before we hear from Maya and Lila, I have a few questions that I think are so basic, but we never really define them, and I like to have things defined. So I'm going to start with just: what actually do we mean when we say AI?
So AI can mean a lot of different things. It's a relatively broad term. In its simplest form, it's technology that allows machines and computer systems to perform tasks that typically require human intelligence — things like learning, problem solving, decision-making, perception, and understanding language. Okay?
And these are terms I hear, but I have no idea what they are. LLM.
Okay, so an LLM is a large language model, and it's okay. So basically, it's a computer program. It's trained on massive amounts of data, and we're talking billions and billions of pages from books, articles, websites, forums. And you take that data, and you process it with literally the most powerful supercomputers in the world. You add a little bit of refinement from human feedback, and you get this incredible tool. We call it a model, and it can generate all kinds of content and engage with you in natural conversational English.
Got it. That makes sense. But then what's generative AI?
Generative AI is a little bit broader. It includes large language models, but it also includes tools that can create art, images, video from text. You can kind of use them interchangeably. If someone's talking about generative AI or large language models, more or less they're talking about, you know, the same broad set of capabilities.
Okay, so now I feel ready to dive in. Who are our AI 101 instructors?
So lucky for us, we have two folks who sit close enough to the hard science to see everything up close, but they're also coming at this on the side of the humanities and can help translate some of this complexity to non-experts who are trying to peel back the curtain. So we're going to hear from DeepMind's Lila Ibrahim later on, but first we're going to hear from Maya Kulycky.
My name is Maya Kulycky. I lead strategy, operations, and outreach for Google Research.
Part of Maya's role is working with research labs to help shape the next iteration of technological breakthroughs that we hope are going to change the world.
I also had a hand recently in introducing Google AI Essentials, a course which was designed to help answer some of the most basic questions that people have about AI. When we think about AI and its origins, it's really tied to the human brain and our ability to try to understand how do we think, how does the human brain work, and to replicate some of the activity that we do as people in order to assist us thinking, understanding, learning. These are different cognitive tasks that are part of AI. What AI can do is remarkable. It can't do what we do as people, and it is not as efficient as the human brain. The human brain is a remarkable, beautiful organ that runs off very little energy. It is just that's why people have wanted to find something similar to it, you know, since the dawn of time. It is exceptional and unique. But we've created some tools that have some aspects that are shared with the human brain, and we are very, very grateful for it. They allow us to share information and to understand things as people using our brains much better than we did in the past, and to get to mutual understandings more quickly too.
I always encourage people to go and experiment with large language models and see how they can be used in their lives to make their lives easier. It certainly makes my life easier. I'm a parent and I have two lovely kids, ages 13 and 15, who would be cringing right now if they knew that I was mentioning them. I'll tell you some of the personal miseries that AI has eliminated from a very, very basic point of view. So my kids, for instance, would ask me for a jacket. I want a jacket. I want a red jacket. Okay, I'm going to search for a red jacket, and I have to type in, I'm looking for a red jacket. And then, you know, the conversation goes like, what kind of jacket? Oh, it's a puffer jacket. Okay. And that's not that color red. It's a different color red. Okay. I mean, we're 15 minutes in and we're still looking for this jacket. AI has allowed us to do things like circle to search, right? Or I can give it a picture of someone wearing the jacket that my kids are asking for, and then a few seconds later know exactly where that jacket is, where I can buy it, how much it costs, is it on sale, things like that. So I'm thankful for all that time not wasted.
At the same time, please, please do not take the output of a large language model and think you're just going to take that and roll. No, it's meant to be a collaborative tool. It's meant to be something that helps you get to a certain point. You look at it and you say, okay, this is a great place to start, but I'm going to improve it. Or, oh, this piece is not what I meant. I need to change this. I need to change that. It's something that you need to check the work of. It's not something that is meant to be a freestanding tool without us as people.
It's not just Maya here saying this, but in the previous episode we heard similar ideas. Do have to check your work. Yeah, this is what we mean when we talk about the human in the loop. It's this idea that, you know, AI, if we really want it to be an extension of us and not a replacement, we have to understand our role in that equation. And brings us back to why it's so important to understand AI, because if we understand it, then we have the building blocks to actually begin to harness it and use it, you know, for our purposes and desires as opposed to what the AI thinks that we want.