View Today's Schedule
With so many stories in the news about the new capabilities of artificial intelligence, Emily from the Museum's programs team explains what that term means and how AI works.
This Pulsar podcast is brought to you by #MOSatHome. We ask questions submitted by listeners, so if you have a question you'd like us to ask an expert, send it to us at email@example.com.
ERIC: From the Museum of Science in Boston, this is Pulsar, a podcast where we dive into the answers to the most common questions we get from our visitors. I'm your host, Eric, and our newest permanent exhibit is Exploring AI: Making the Invisible Visible. It includes an AI-enhanced oven from June, a full-scale model of NASA's self-driving Perseverance Mars rover, and more examples of applications for artificial intelligence ranging from healthcare to transportation to commerce. But the most common question we get is still: what is AI? My guest today is one of my fellow educators at the museum, Emily, who has been leading the development of our artificial intelligence live shows for the past few years. Emily, thanks for coming on the podcast.
EMILY: Of course. Thank you, Eric.
ERIC: So how do you answer this question, when you get it from visitors, after your shows? What is artificial intelligence? Like, what counts?
EMILY: So I tend to describe AI as a machine that has the ability to make decisions that normally require human intelligence.
ERIC: So that line can be sort of hard to define, are there any characteristics or methods that are widely used by AI?
EMILY: The first thing that I think of is the use of a neural network, which is sort of like the brain of artificial intelligence. And those neural networks sort of help us to train the AI to do the things that we want it to do. And eventually, you can start making those decisions.
ERIC: You go into the neural networks a lot in pretty much all the shows where you're talking about the different facets of AI, a lot of them are built on this neural network, can you talk a little bit about how it kind of imitates the human brain in a lot of ways.
EMILY: So it definitely involves sort of pathways, and it has to be trained, which means you need to show it a ton of data of whatever you're trying to make it do. So for example, when it comes to computer vision, which is one amazing application of artificial intelligence, in order to make a neural network recognize a cat, you have to show it so many pictures of cats, and then tell it these are cats, so that it knows what a cat looks like in the future.
ERIC: So a program coming up with its own rules of what makes a cat a cat, because you show it so many pictures of cats. And the data matters. You have to have good pictures of cats.
EMILY: Oh, yeah, absolutely. Yeah. And it's also like, you know, you can't just show it pictures of tabby cats and expect it to recognize a calico as a cat, you have to be very careful about diversifying all of that data.
ERIC: And that leads to one thing you're always sure to talk about in these artificial intelligence presentations. And that's the fact that these programs can have bias if they're not trained properly, and they don't have a representative training dataset.
EMILY: We're definitely using things like computer vision on human faces. And we've all heard of facial recognition. Some of us can open our cell phones, with our faces, things like that. If you have an AI program that cannot recognize black skin, for example, then that's a big problem. It that means that you didn't use enough diverse data to train the thing, so that it's not recognizing a huge portion of the humans that are trying to use that. And some cities are using some version of facial recognition to help them solve crimes. But again, if it's not using diverse data, then it's going to lead to the arrest of a lot of innocent people. And it has led to that, by the way.
ERIC: I do like in your shows, when you sort of compare a neural network and a brain sort of step by step, can you go into that a little bit?
EMILY: The neural network is modeled after our brains. And the way that we see images, for example, is that our brain breaks down those images into their most basic parts. And then it keeps looking for more complex ways to understand what it's looking at. So I'm just going to keep talking about cats. Thanks, Eric. If you are looking at a picture of a cat, then you're going to see a circle with two, you know, triangle ears, that's sort of the first process that your brain does. And then it sort of builds on top of that, and it starts seeing little features like eyes, nose whiskers, and that image just keeps getting more and more complicated in your brain. And then eventually, you recognize it as the thing that's in front of you. So neural networks are doing something very similar. It's called deep learning, where it's looking at very simple shapes first, and then those images get more and more complex. And that's how it's breaking down and learning from the images that it's seeing.
ERIC: Another application for AI you've talked about is sort of understanding human language. What's a good example of that?
EMILY: Well, I would say that, you know, our personal assistants, our home assistants, are definitely a big deal. We have them in our pockets, in our phones all the time. We're also talking to them in our homes sometimes. So for me, that's always, like, an ever-present use of that particular technology where you can speak to your artificial intelligent device, and it can understand what you're saying, and then even perform tasks based on what you ask it.
ERIC: And the AI involved is really impressive here, because as you talk about in your show, Conversing with Computers, this is my favorite part: human language is really complicated. You can't just have a computer program analyze what you say word by word, you need something deeper that can understand the subtleties that we take for granted in our everyday conversations.
EMILY: Exactly. And when it comes to things like translation services, those also use artificial intelligence to help them. But before we were using AI, we were doing kind of like what you said, where it was just computer code, it was just translating word by word within a sentence. And anyone who speaks a second language knows that that does not work. So AI really upped the game when it came to translation services.
ERIC: So we see more applications of AI in our lives all the time, what's something coming up that AI could help with that you're excited about?
EMILY: The thing that I'm going to say isn't even going to sound that interesting, but I'm so excited for it. It's like smart traffic lights. I want to cut down on the amount of traffic that I have to sit in every time I drive to work or drive across the state or whatever it might be. So these traffic lights will have the amazing ability to see traffic building up within a grid of traffic lights. And it can change the flow of traffic, it can change the timing of those lights, it can make one section move more than another. And that'll cut down on the time that people are waiting in traffic, which we all know is, you know, exhausting. But it can also help with the environment in a small way, because then fewer cars are going to be idling and pumping fumes into the atmosphere. And I imagine it could also learn patterns that happen at certain times of the day. So it can know what to look for in the future.
ERIC: Alright, Emily, thanks so much for joining me and talking about AI.
EMILY: Of course. Thank you.
ERIC: Don't miss our Exploring AI exhibit on your next visit to the museum. And while you're at home, visit mos.org/artificial-intelligence for a wide variety of videos and resources on AI. Until next time, keep asking questions.
If you liked this episode, be sure to check out:
Theme song by Destin Heilman