Dr Gary Stager on artificial intelligence – origins, misconceptions and applications

Transcript

Introduction:

You're listening to the Victorian Academy of Teaching and Leadership Podcast, where we showcase conversations with some of the world's biggest thought leaders in education. We also bring you the thoughts and reflections of teachers and school leaders from across Victoria.

Anam Javed:

Hello everyone, my name is Anam Javed, and I'm the master teacher in residence for technologies here at the Academy. Today we're joined by Dr. Gary Stager, whose internationally recognized work has been helping school leaders and teachers embrace technology and amplify the potential of each student in preparation for an uncertain future. Today, we're discussing artificial intelligence, and this is a great episode for school teachers and school leaders to better understand and explore the challenges and opportunities that AI brings. Thank you for joining us, Gary. Let's get started.

Gary Stager:

Great to be here.

Anam Javed:

So Gary, when I speak to teachers and colleagues in education, the vibe very much is that artificial intelligence just popped out of nowhere in 2022. Can you tell us a little bit about the origins of artificial intelligence?

Gary Stager:

Well, in a lot of ways, apparently education was revolutionized over Christmas break. It did pop out of nowhere in the form of this ChatGPT software. Artificial intelligence is a legitimate scientific discipline that has at least 60 years of work behind it. And I should say, I'm not an expert in AI. I have been lucky enough to know some of the pioneers in the field, and spent some time with them, and maybe that's allowed me to have a more rational measured approach to both the hype and the hysteria around ChatGPT, around generative artificial intelligence, as the broader category, and around the implications of it for education.

Anam Javed:

Wonderful. I love that you used the term irrational and measured approach, and keeping on that theme, Gary, could you enlighten us on the differences between artificial intelligence and machine learning? Are they the same? What's the go here?

Gary Stager:

Well, I hope this isn't a test. Sometimes artificial intelligence is just software. We've had a lot of software, and software gets better and gets smarter, and should help us be better at what we do. It should help us be better and more creative and smarter at what we do in our everyday lives. My mentor and friend, Seymour Papert, who is one of the pioneers at artificial intelligence, liked to say, "That everyone needs a prosthetic." If I put my glasses on, no one accuses me of cheating, it just helps me live my life better. And there's been all sorts of technological advantages that are in that spirit, and I think what's happening with artificial intelligence is not dissimilar.

My understanding is machine learning is where the software learns some things. Where the software gets smarter. The simplest example that I can share was one that I saw recently done with the BBC micro:bit:, this $20 little miracle brain board that allows kids to explore electronics and robotics and physical computing. And there's millions of them in circulation. And the example was, they made a hat out of paper, and they said we're going to train the system. And so, they nodded their head up and down repeatedly, and then said, "Stop," and the computer remembered what that motion was like, because there's an accelerometer in the micro bit. Accelerometer is something that knows whether it's been tilted or turned or flipped or dropped or thrown. And they said, "I'm going to call that nod."

And then they shook their head from side to side, and they recorded that for a moment, until the computer said it got enough data. And they said, "I'm going to call that shake." And you could think about, when you're trying to teach your iPhone for facial recognition, you do a little machine learning, where you turn from side to side, so it learns what your face is made up of, so it can recognize it. And then, a kid, or anyone else, could write a computer program that says, when you nod, do something. When you shake, do something. So, you're taking some phenomena and you're naming it, or a list of instructions and you're naming it, and the computer is breaking down whatever the complexities of that seemingly simple action is.

And then it learns them, and making inverted commas. Artificial intelligence was the idea that we can teach computers to think. And there's been very different visions of artificial intelligence, the MIT origins of John McCarthy and Marvin Minsky, and people like Seymour Papert was about... Seymour used to say, "You can't think about thinking without thinking about thinking about something. And that if you're thinking about thinking or thinking about thinking about something, you become a better thinker."

And in the process of getting computers to think about thinking, you become more human. You become a better thinker. And that's a vision of artificial intelligence that is taken a backseat to this get rich quick, gold mining vision that the media and policymakers are agitated about now. I had someone from a state department of education in the United States come up to me at a cocktail party recently, and say, "We're being tasked with developing policies around artificial intelligence, am I qualified to do so?" And I said, "Absolutely not." I often recommend that we heed the advice of the great Pee-wee Herman, and "Chillax." It's just software.

Anam Javed:

Yes.

Gary Stager:

And I've actually engaged some of the leading tech journalists, one in particular, in the United States, Kara Swisher is a monster tech journalist, who gets scoops on everything, and has been incredibly entrepreneurial, and makes people like Mark Zuckerberg sweat through his shirt. But when she talks to people about generative AI and this software like ChatGPT, there's this moment where they talk about all the upside, all the wonderful things it can do, and the great promise that it offers for society. And then, almost on cue, you will all then have to bow your heads and say, but of course there's a dark side. And everyone just nods in agreement. And I reached out to her and I said, such as? That's the basic journalistic follow-up question.

What's the dark side exactly? Now, I know there are already dystopian science fiction visions of machines taking over the world, but at least for the foreseeable future, the software we're talking about, at best, will write a bad five paragraph essay for a class you don't care, about on a topic you're disinterested in. One of the reasons why I think educators are so agitated about it is because one of the things that ChatGPT does very well is the nonsense that we make kids do in school. So, I used to say that any teacher who could be replaced by a computer probably should be, and we might amend that to say, any assignment that can be submitted by ChatGPT probably shouldn't be an assignment.

Anam Javed:

Love that, Gary. In fact, I feel quite reassured, personally, because not only have you demystified ChatGPT for me, you've actually addressed a lot of the hype train. So, is it going to replace teachers, Gary?

Gary Stager:

Probably not. But probably not for the right reasons. And this might be just tangential, but during the pandemic, if we recognize that schools serve three primary functions, of child minding or childcare, socialization, and delivery of instruction, schools could have done two of the three safely, and we decided to do the one that was impossible to do, which was the delivery of instruction. If you had just said, drop the kids off in a field somewhere and they'll run around all day, you might not have had the mental health outcomes that people seem to think that kids are suffering from, and they probably would've learned and stayed on pace. But instead, we had to invent all kinds of Draconian ways to deliver arbitrary bits of knowledge into unwilling kids' heads while they were sitting in their bedroom.

And I think that if you think about the role of school, I often say that school has an obligation to introduce children to things they don't yet know they love, that you want to create rich experiences where you gain benefit from being co-located in the same space at the same time. Then those things can't be replaced by ChatGPT, or by generative AI. But if you think it's preparing for a standardized exam and that's the primary role of the teacher, to dispense information to children, that absolutely can be replaced by a computer program.

Anam Javed:

Yeah. And so, in a lot of ways there is an onus on teachers to harness the use of ChatGPT and try to move beyond the rhetoric, and as you mentioned, the dark side, that's pandered by media, and move towards using it in ways that are better than, write a prompt for an essay in English, et cetera.

Gary Stager:

Yeah. I'm horrified by the number of people in my LinkedIn feed who have, expert in AI and education, in their bio now. Now, this could be the result of AI and a bad algorithm, that's just making me have to endure all this-

Anam Javed:

Your LinkedIn feed. Yeah.

Gary Stager:

... nonsense. But it seems to me that, first of all, you should have to write a five paragraph essay about AI before you declare yourself an expert in it. You might actually have to read a book that was written more than before yesterday. But beyond that, if you wouldn't announce that you were an expert in genetics or physics or biology, why do you have the audacity to claim that you're an expert in AI? I think there's a lot of-

Anam Javed:

And who's granted that degree and certificate?

Gary Stager:

Yeah. Right. And there's that problem, there is problems with bias and facial recognition and race, and anything that's going to be using large bodies of data are dependent on where the data source originally comes from. And there are unintended consequences, if Australia says to Google and Facebook, hey, we won't let you use our news data anymore, then by definition, the system will become less Australia-centric. But having said that, I think one of the things that kids and teachers need to start thinking a lot more about is, there's a term that I heard introduced in some of these discussions about how AI is going to impact on society and jobs and such, is that you need to own your digital twin.

You need to own your likeness, yourself, your work, your ideas. And that's a notion that has been far into schools. And there's an earlier precedent for it, which is there's a project out of a university in the United States, I believe the University of Mary Washington started something called... I could be wrong about that, but I think it is. Called a Domain of One's Own. That it's preposterous for anyone to have a Gmail account, unless they're using it as their burner email. It's equally ludicrous for the way you communicate to be doing so via your school's email address. Teachers have suffered under the random proclivities of school internet policies long enough. In 2023, every kid doesn't have their own space on the web and may not have email access, it's unacceptable. So, suggesting shifting the burden, saying no, everyone should have their space, their place, their identity, is a really good way of ensuring that.

Anam Javed:

You leave your imprint on the internet.

Gary Stager:

You can do your best work, you can have access to what you need, you're unencumbered by some arbitrary gatekeepers. And when I was 12 years old, we published an underground school newspaper, when it was hard to publish a newspaper because you actually had to have access to a printer. Now, every kid's got a computer in their shoe, and a printer in their house, and a worldwide web, and infinite audience, and streaming, and podcasting, and yet, they've never been more helpless. And their teachers are helpless as well. And I think we ought to be addressing that helplessness of, why are you tolerating this nonsense? It was actually the case that teachers couldn't use a telephone for 120 years. 120 years after Alexander Graham Bell invented the thing, you might still have to call your doctor from outside the canteen at lunchtime. Or take a parent call from the nurse's office. And then, this thing I'm holding up, the cell phone happened, and you were no longer willing to suffer that indignity. However, we're perfectly happy to tell kids that they can't have access to the world in which they live.

Anam Javed:

And I think what you're talking about, Gary, I really appreciate, because the debate around ChatGPT, from what I'm hearing from you, is symptomatic actually of the wicked problems that we talk about in the Australian education system as well, and bigger conversations around ownership and agency as well. And where it sits on a spectrum with policymaking and consistency and structure. So, I guess, on that note, talking about student agency, can you share with me what excites you about the potential that ChatGPT, or even AI, broadly has in that regard?

Gary Stager:

And there's a difference between childish and childlike. I want teachers to be childlike and filled with wonder and curiosity.

Anam Javed:

Beautiful.

Gary Stager:

And childish is petulant and whingy, and we want to create, for kids and teachers alike, we want to empower them. And I learned even from the early childhood educators [inaudible 00:16:05], it's irresponsible to build pens around children. It's our job to find constructive ways of using what they find in their world in a constructive way. So, let's talk about ChatGPT in that regard. You can get answers to questions, it's a front end on an infinite, potentially, library of information. That's all good stuff. Everything I've ever read about writing says you need a copy editor. You're not so clever, you need a copy editor. ChatGPT is a good copy editor.

Things I'm excited about isn't the stuff that everyone's talking about, of, you can ask the software to give you a purple dragon. Who cares? Although, that's fine, I don't find it to be very good at giving me purple dragons, and there are issues around who owns the dragon, and where's the copyright and compensation and metering, where things come from, and all that needs to be resolved. The stuff that I'm excited about is the software that's letting me do my work better. And I'm writing a white paper about this currently because there's a couple titles that I use that I haven't seen anyone else in education talking about. There's a piece of software called Authory, that keeps track of all of my writing online, and puts it in one place, archives it, allows me to share it in a variety of ways. Now, why is that important? Because I wrote for a magazine for a dozen years, they changed servers, and it lost all my work.

So one of the things kids and teachers should learn is, you're responsible for your work. This is related to the digital twin or the Domain of One's Own. And Authory now knows where my work tends to appear, and it's on the lookout for it constantly. And when I write something new, it makes a copy of it. And David Pogue from CBS News, in America, said, "This saved my life," and I know him a little bit, and I saw him say that on social media, and it has saved my life as well. And as a result, several thousand pieces of writing that I've done are now not lost to the ages.

Anam Javed:

Brilliant.

Gary Stager:

You're recording audio in a classroom, or video in a classroom, classrooms are noisy. They have things humming and buzzing and alarms going off and other kids talking. So, there's a software like a thing called LALAL.AI, where you take an audio file and say, clean this up, and it does it. And I only know how to use 1% of the tool, that 1% is sufficient for me. My favorite of all of them, and we're on a podcast now, so I'll share this with you.

Anam Javed:

Yeah.

Gary Stager:

There's a piece of software called Descript. And I've been around computing now, going on 50 years, and every decade or so I see a piece of software, where I think, huh, that's kind of cool. About every 20 years I see something where I crook my neck and go, oh, that's magic. Descript is magic. So, if you have a video file or an audio file, so kids shot some video or they recorded themselves saying something, you upload it to the software, it transcribes it on the fly.

Anam Javed:

Wow.

Gary Stager:

Okay, that's a neat trick, that's a neat trick, but there's other software that does that. Having a transcript of something you've heard or watched is useful for a teacher or for kids sometimes. You want to grab a sentence from it and use it some other form, properly cited, of course, that's great, but that's not the magic. The magic is, you say, find all the filler words, all the ums, huh, stammering, like, like, repeated words, it finds them.

Anam Javed:

Wow.

Gary Stager:

You can manually, if you want, go through and say delete them, or you could just say kill all of them. But it doesn't just kill them all from the transcript, the video or the audio file is edited.

Anam Javed:

Wow.

Gary Stager:

You know when you're doing a Zoom conversation with someone, and you say, oh, where'd my mouse go? How do I get that screen? No one wants to watch the rerun of that. You can grab that paragraph, delete it, it leaves the video. Let's say you use the wrong word in a sentence, you can overdub the word and the video will change. Or if you're really lazy, you can train it with your voice, type the word, it will pronounce it in your voice. It exports in 1000 different formats. Take the third paragraph, move it before the second paragraph, the video changes, or the audio file changes. Why the hell wouldn't kids use this? Their writing could become crisper, shorter, more compelling, the storytelling is better. The whole point of writing is editing, just like the point of programming is debugging. It's making something better. Why wouldn't you allow kids to use a tool like that that would make their communication better, more powerful, more meaningful, convey a greater understanding of something?

So, that's the stuff that I'm really excited about. And in the mathematics context, I've gone down a rabbit hole of, I asked ChatGPT to give me a list of dollar words. I shared this in a workshop yesterday. So dollar words was a problem that the Great American maths educator, Marilyn Burns created, probably 40 years ago. That says, if A is worth 1 cent and Z is worth 26 cents, can you find words in the English language where the sum of the value of the letters in a word add up to be a dollar?

Anam Javed:

Wow.

Gary Stager:

Now, that's a nice puzzle for second-graders, and kids will really struggle with it, and there's a little picture book where all the words are dollar words. So, I asked ChatGPT to give me a list of dollar words. If it had said to me, I don't know what that is, I would've been done with it, unlike me arguing with it about Maria Montessori until 3:00 in the morning one night. And I said, "Give me a list of dollar words," and it said, "Sure," and it explained who created the problem and how it works. And then it filled my screen with a 100 dollars words. And I saw alfalfa. And I went, that doesn't seem right. And I took out a piece of paper and pencil, and I scribbled, I did the maths, and I'm like, yeah, alfalfa is a 39 cent word.

Anam Javed:

Right.

Gary Stager:

It has a lot of A's in it, and all of the letters are at the first half of the alphabet, there's no way this is a dollar word. I said, "Some of these are wrong." And it said, "Oh, I apologize. Here's some more." And I got cabbage. And another list of 100 words. And then I got madder and madder at it, and told it, "No, you're not doing this correctly," and it got more solicitous, which was offending me even more. And then I thought, well, how do I know which ones are right? I should write a computer program that lets me type in the list of words it gives me, because I don't want to do all the manual calculations.

Anam Javed:

Yeah.

Gary Stager:

Now, I've been programming computers long enough that I was able to take Logo, the programming language designed for kids, and write a new dialect called Links, that would fact check ChatGPT for me. But then I started thinking about, how can I engage kids in this? And I went back to some timeless programming problems that Seymour Papert and Cynthia Solomon began posing to children as early as the late 1960s. So, for example, if you teach the computer a random list of names and a random list of verbs, and you asked them to make a sentence of person, does what, person, it'll say, Gary likes Rachel. Mary punches Susie. And kids giggle, and then, you say, oh, what other kinds of words can we add to that? And they'll make adjectives or adverbs. And then someone will inadvertently put it in the sentence in the wrong order and the computer will say something hilarious, and the kids will go, oh, that's funny. No, it's in the wrong order.

And they start dealing with parts of speech in a really concrete way. And when they were doing this in the late 60s, one of the year five kids, working with Paper and Solomon, said, "Oh, now I know why there are nouns and verbs." Which, you teach all that stuff, but they don't necessarily stay taught. They don't necessarily have understanding of it. But when teaching the computer how to do it, you understand it better. So then we might go from that to, can we write a program that lets me type a word in and it gives me the plural form of that word? And so, if I say, truck, add an S, it's trucks. So I say, boy, it says boys. If I say cherry, it says Cherry... C-H-E-R-R-Y-S. Now, we have to build exceptions into it. What do we do about moose? What do we do about goose?

Anam Javed:

Love that.

Gary Stager:

And as the students are doing that, they're dealing with a lot of logic and problem solving, they're dealing with some computational issues, and they're learning parts of speech and the proclivities of our language. Within minutes of that activity, a kid will spontaneously say, oh, this is like AI. And then I said, okay, let's go one step further. So, ChatGPT couldn't generate dollar words correctly, how can we write a program to check it? So I might show them what I wrote so they can understand how they might check a word. Then I might say, what if we wanted to have the computer randomly generate words? What's the shortest word that could be a dollar word? And they think about it, and they go, four letters. I said, so what would be a four letter word that would be a dollar?

And they'd go, ZZZZ. That's not a word. So, what's the more probabilistic number of letters in a word? Well, let's start with five letter words. Ask the computer randomly generate five letter words, have it run that all night, now we can check all of those to see if any of them are dollar words. And I'll get a list of dollar words that were generated at random. Except none of them are English. Now we've got a different problem. How do we know if they're in English? And that's where we run into a wall, because no one has made a software environment for kids sitting on a large language model, which is what ChatGPT is sitting on. It's sitting on the world's knowledge. I don't even have a dictionary in these programming languages for kids, so I can't compare it to the dictionary. We could then transition to software, like Wolfram Language, created by the Wolfram Company, which is the software that's underneath Wolfram Alpha and Mathematica.

Anam Javed:

Yeah.

Gary Stager:

The software that's used by mathematicians and scientists. And in one expression, I could say, show me all of the words in the English language where the sum of the value of the letters equals 100. And it will do that within a couple of seconds. And then I can say, divide that by the number of words in the English language. And by the way, that's not a static number, that changes over time. And it gives me the percentage of dollar words that are in the English language. And just to wrap up this thing, what I really find exciting about this as an educator, is the children are being mathematicians, they're engaged in a playful childlike inquiry, that when I shared the dollar word problem with Stephen Wolfram, who's arguably one of the world's leading mathematicians and scientists, he's Einstein level, he didn't say, why are you bothering the world's greatest mathematician with some second grade arithmetic nonsense? He said to me, "My, how wonderful that is. And do you mind if we put that on our website?"

Anam Javed:

Wow.

Gary Stager:

And in fact, on the Wolfram website, there's a page about dollar words. And, I can't say that I influenced this in any way, but a few weeks later, ChatGPT got a lot smarter, because there's now a plugin for Wolfram language. So that, now, when you ask it a problem like that, it's not going to give you gibberish, it's going to give you something accurate. So, if you set policy for this stuff, you're going to be locked into that policy, and the software changes every day. So, again, to quote the great philosopher and educator and humanitarian Pee-wee Herman, we should just "Chillax."

Anam Javed:

And I feel very chillaxed, Gary, particularly because you've left us with an incredible array of interdisciplinary applications. Not only have you talked about open access AI tools, you've actually talked about their applications in literacy, numeracy, digital technologies, languages, play-based learning, and a range of other applications in education. And I genuinely feel excited as an educator, but I was wondering, if you think about our classroom teachers in Victoria, teachers at the coalface, who you've talked about so passionately, what are some words about this AI boom, this advent of ChatGPT, what are some words you want to leave them with? Some nuggets of wisdom?

Gary Stager:

Stop talking, stop worrying, start doing something. The more you use ChatGPT, the less horrified you are by it. Except for, unless you're horrified by the fact that it's not that good. I don't use it that much, but I've used it enough to know what it's good at and what it's not good at. And I'm actually co-authoring something with it, which is going to create all sorts of ethical implications, because about 90% of the work work was written by ChatGPT, I was just tweaking it. But if you think about we make countless arbitrary decisions about what we're going to teach kids, what's in the curriculum. Why do we teach kids to write haiku? Were you in a meeting where that was decided?

Why is that more important than something else? And if we think that teaching poetry is important, if you ask ChatGPT to give you a poem about pirates, and then you say, ooh, less violent, I would like every second word to rhyme. I need it to be shorter. I need it to have a female point of view. I need it to be appropriate for seven year olds. You are engaging in all the intellectual processes that one would hope you would engage in by writing poetry. That, in this case, it's just the voice in your head that's helping you think about the poem that you want to write. It's removing some of the friction in how you want to express yourself. So, my best piece of advice is, when everyone's talking about something, leave the conversation and go do something with it, and gain a sense of it, and then see how you can make this part of your repertoire.

Anam Javed:

Thank you, Gary. I'm actually going to do just that, as we leave and wrap up this conversation, I'm actually going to go and have a play, not just with ChatGPT, but some of the wonderful tools you've mentioned. And on that note, I think we'll leave this conversation here. I could talk about this all day with you. A big thank you to Dr. Gary Stager for your incredible insights into mathematics and AI and literacy and numeracy and computational technologies in the classroom. If you've enjoyed this episode, stay tuned, we've got more episodes to come very soon. Thank you.

Gary Stager:

Thank you.

Outro:

We hope you enjoyed this Academy podcast episode. You can find out more about our upcoming professional learning opportunities at academy.vic.gov.au. And follow us on social media to stay up to date.