Audio Podcast

In this episode, Matt sits down with Ken Calhoun and Meryl Perlson to discuss the impact of artificial intelligence on higher education. What are the implications for teaching, for ownership of content, for privacy, and for creating, teaching, and learning in the world today?

Related Resources

DALL-E "Painting": Rembrandt of a surly duck demanding a drink from a cowering bartender

What does Artificial Intelligence Do Well?

AI in Higher Education

AI and Design

AI Design Jam

ChatGPT, Artificial Intelligence, and Academic Integrity.docx

Navigating the Brave New World of Generative AI in Higher Education

Top 9 ethical issues in artificial intelligence

An A.I. Pioneer on What We Should Really Fear

Implications of Artificial Intelligence for Filmmaking, Journalism, & Public Relations

Lev Manovich on Facebook

7 Ways to Use ChatGPT

Introduction: Artificial Intelligence, Image Generators, and Automatic Ducks

Matt Boyle: So welcome back. With me today is Ken Calhoun, the interim dean of the School of Communication and the Arts and Program Chair for graphic design here at Lasell. And Meryl Perlson, who is professor and program chair in the School of Communication and the Arts.

We're going to be talking today about artificial intelligence, what it means for higher education, what it means for creativity, what it means for things like privacy, what it means just about anything that I haven't anticipated in that list. Artificial intelligence has [been] seen a lot more recently, like, for example, in this description of DALL-E. DALL-E is a network-based image and text generation model developed by OpenAI. It can generate unique images from text descriptions and can generate text descriptions from images.

Basically, you tell it to make a thing and it makes the thing. I told it to make me a picture of a duck like it was painting Rembrandt, a baroque style. I just made up stuff. "Send a duck into a bar and have it yell at an angry bartender" or something like that. And it produced four images. Two of them were deranged and looked insane. Two of them seem perfectly fine. One of them, actually, I like quite a bit, and then I just continued to play with it. So, it can do that in any particular style, which is a deeply fascinating thing, to greater and lesser success.

So that's what I know about DALL-E specifically. Ken, Meryl, I assume you know more about that and other artificial intelligence tools? What else should listeners know about A.I. image generators or A.I. in general?

Concerns about Integrity, Creativity, Content Ownership, and More

Ken Calhoun: Well, I think with the image generators, what we're seeing is a, um, a kind of explosion in terms of basic awareness of these tools. And it's really because these tools are only something that were publicly available within the last six months, and within the last six months that we've seen a lot of kind of refinement within those tools because of the exponential aspect of machine learning and people using these tools. And in doing so, they are refining the tools by using them. So they'll just get sharper and sharper and better and better.

So what it does for people like us who teach courses in image generation by humans is it immediately poses a threat. And what I tried to do earlier last semester was understand the depth of that threat and try to make people, my students and other faculty members, aware of the potential hazards from a purely educational perspective. But, of course, the list of concerns regarding these tools goes beyond what we might be more immediately concerned with in an academic setting and goes deeper into really quality of life in the imaginable future.

So, it's a wide-ranging conversation, I think, I guess, for the sake of this discussion today. We'll keep it close to the academic concerns. But I was able to look up certain, you know, lists of these different challenges. We talk about professional challenges. I'll just lay some of these out and let and let Meryl comment. But one of them, uh, you know -- is the obvious question is how will these tools, the image making and writing tools that are being developed, language-based tools going to make human image making and human writing obsolete? That's obviously kind of a knee-jerk fear when you hear about these kinds of tools and then are we contributing to this outcome every time we use it?

Other questions I have are how do we stay relevant as artists and writers in this setting? How can writers and artists whose work, or works, are merely perceived as data be credited and compensated? That's another big issue, and that's why there are artists mobilized right now against A.I. -- Artists against A.I. -- and then real specifically, you know, how and should this, should this even be used as a resource for the designer today?

And if so, that raises a bunch of questions about, you know -- citation attribution, plagiarization, just the notion of cheating. So all of those things are things that we're confronted with in the classroom. If you look at this list from the World Economic Forum about concerns, they're wide-ranging and go beyond that. Everything from employment is an issue to inequality. You know, how do machines affect our behavior in interaction as humans? Artificial stupidity. That's something that needs to be, something we need to be prepared for. Uh, racism in the system is likely. And we can talk about why. Security. Evil genies. We haven't had to worry about evil genies for a while, but it's back.

And then the notion of the singularity and even robot rights are things that the World Economic Forum has identified. So those are all really kind of lofty questions on the more immediate sense, Meryl, What, what thoughts do you have when I read through some of those concerns?

Defining Creativity and Integrity in a World with A.I.

Meryl Perlson: Well, I share a lot of them, Ken. But I also...you described it as a threat, and I think it is a threat to the status quo.

But I also think it's an opportunity, you know, the types of things our students do -- professional communicators, professional designers, people who identify communication problems, visual or verbal or otherwise, and design around that isn't going to change. So I think it really pushes us to consider what our students are going to be doing five years from now. You know, how do we define creativity? Is creativity about the conceptualization of the solution, about having an original idea? How important is the execution to that?

Because even the example that, Matt, you just provided -- a duck drawn in a Rembrandt or Baroque style -- required knowledge of Rembrandt, it required knowledge of what Baroque style meant. And so that's not going away. Whether our students are going to be the ones having to push the buttons is really more the question. And I think, in that regard, we...there is to me the concerns are at the entry level in some ways.

You know, if you can have chatGPT, write simple marketing messages, and you can throw a few photos at a video AI and have it edit a package for you. What are the entry-level positions doing? Do we need the folks who can come in and maybe do some of the execution of the strategic ideas? Or are we really thinking that the entry-level is now going to be coming in with the strategic ideas? So I just think it's moving the bar in terms of where the productivity in the human piece comes in. You know, what can humans do that the machines can't?

I also think in terms of the originality in higher ed, you know, we have such a relationship with integrity and original work and thought, and if you're having to pull a paper together for you, is that acceptable? If you've identified sources for it and go back in and edit afterwards and fact check and whatnot, are you not doing the critical thinking that was required to assemble this group -- this information? So it opens a lot of questions for me.

Matt Boyle: Yeah, it opens them for me too. You could certainly ask ChatGPT to assemble a paper for you. And I have. I have reservations about even doing that because I think by giving it the props that give it the skill, am I giving it the skill to make other individuals obsolete, doing it themselves? But just the same, it feels like you need to actually practice it to understand what it's capable of doing and to understand how to responsibly address it with respect to writing a paper.

It isn't great when it comes to being asked to write a paper. It'll write 300 words and skip along the surface. It'll give you what is almost the logical fallacy of arguing an argument with the premise of the argument. You might say, write me a paper about why the death penalty is unethical, right? Here's your prompt. And it will say the death penalty is unethical because the death penalty raises ethical issues. Like okay, technically, that's right. In the 10,000 miles above the sky. But where are the examples? And that's, I think, what it's terrible that it falls down badly when it's asked to give specific examples, specific evidence, but it provides wonderful templates. for stuff I could ask it to give me an outline for something. It's like "Here's an outline for what you just asked for: Point 1, Point 2, point 3." That I look at and think to what you said. Is it appropriate to ask an individual to use it as a template? We already suggest, for example, graphic organizers for a student. Okay. Is this just going to become a graphic organizer? How do you cite it in that case?

Collaborating with A.I. in Instruction?

Matt: Back to what Ken was bringing up. And there's a, it's just a very big open question about how to do that. And then. Is it organizing your thoughts and are you? Letting yourself down with the soft expectations that what it gives me is okay because you can keep on prompting it to give you more than just 300 words. And I could say, "Write me the introduction. Great. Now write me the first paragraph. Great. Now write me the second paragraph and so on until it's done." And I have, and I can even include an example of myself doing that in the notes for this podcast. The paper isn't great, though. It makes examples up if I push it hard enough. It will make them up, and I looked them up and I was like, "These examples exist nowhere; you've just made these up for myself."

Meryl Perlson: Well, but, you know, we already receive a lot of crappy papers, right?

So, I mean, that's, you know, and I think of students, you know, people have different learning styles and neurodiversity. For some people, organizing that information is a huge obstacle to expressing themselves. You know, we've all had those students that are so articulate verbally, but, you know, when it comes to organizing information on a paper, it becomes overwhelming. So, are there ways that we can harness this? I think, you know, I love the idea that artists are coming up against AI and but I think the gates are open, and we need to think about what our priorities are as educators, what we think education really means and how to harness these tools to advance that.

And I could see, for example, that first, knowing what to ask to put in the outline or knowing or having to fact-check the examples and correct the draft, I see those all as potentially valuable learning activities.

Ken Calhoun: Right. And I think that if you're going to engage, that's the first question because a lot of artists will refuse to. Should we as an institution take that stance? I'm not so sure because, you know, we are a graphic design program. And that program is entangled, you know, by nature with commerce and with commercial design. How, how will this affect that particular field? I feel like if we didn't engage it at this point and make ourselves and our students aware of it, then we are in a kind of state of denial. We're burying our head in the sand, and we're allowing a certain aspect of design tool development to just go unrecognized, which I think would be a disservice to the students.

So one way I've had the students dance close to the fire, let's say, is to think of it as a resource, you know, more so for, um, like a textual resource or a basic resource because it -- one thing for graphic design that it doesn't do well is typography. And typography is a good chunk of what graphic design is about -- understanding typefaces, understanding fonts and how how those work and when those fonts should be applied. There takes a lot of cultural insight into knowing what fonts should be used for a particular messaging.

I do a piece called, um, Emotional Type, right? And so it's about how typefaces have certain baggage with them, cultural baggage. So I show students a wedding invitation, which is very frilly and ornate and has a kind of calligraphy type of typeface. And then I show them a flier for a punk gig in the eighties, which is cut together. And, you know, why is one more appropriate than the other? Those kinds of things are hard -- those cultural things are hard for AI to understand --

One, one last stronghold, it seems of humanness in this mix is common sense. And I've been reading about the early, you know, A.I. pioneers always, you know, struggled with this idea how well, how can an A.I. absorb common sense? Well, that common sense is something that you can see in choosing a typeface. You know -- what, what culturally do I know about how a wedding should be presented? What were these, these certain typefaces conjure or invite into the, into the thoughts and feelings and emotions of the viewer? So, in that case, we can use we could use, say, an AI to maybe create an image or a background, but we still have to address it in terms of composition, applying a typeface, making good choices about, the about the things that the A.I. has cranked out and using them. So, just I know this is maybe more of an audio presentation, but I will talk through some of the things and make these things available.

One assignment I gave was for my students to redesign the cover for Italo Calvino Invisible Cities. There's the actual cover of that of the most recent series that you can see. If you can see my screen, it's a very abstract, simple, minimalist image with a gray sky, with a little bit of a skyline represented in the negative space, and just one small, little, almost childlike bird, like a W, flying in the sky. So my students were then asked to redesign this cover. I gave them a little bit of info. A lot of them obviously weren't familiar with this book, which is, you know, really actually a very special book. And I gave them the specs. You know, here's what your mission is to redesign the cover, follow these sizes.

Obviously, DALL-E at the, at the way I'm using it only produces images a certain size. So we had to, we had to lower the DPI because it wasn't -- DALL-E's cranking out images at 72 DPI and obviously a print image or a book cover would need to be at least 300 DPI. So I suppose you can pay for higher, higher end images, high resolution images. But anyway, here's the results and I'll make these available where this podcast is posted, and you can see the students within the class period we had.

They didn't take this home and work on it. This is what I call design jam. They did this in a matter of, you know, 45 minutes to an hour and you can see some of the results. And, you know, they obviously handled the typography themselves, but they layered images, brought images in. Still did, you know, their the design work on it, applying some blend modes, you know, cropping and choosing, you know, the part of the image that they feel -- So all those kind of human decisions are happening in this process.

Ownership of Content

Ken: So it is a kind of collaborative effort between an AI and a human who's then corralling these images and then applying all these other things to it, including this cultural awareness of typefaces to, to create different kinds of -- here we have almost 20 different variations of the book cover. So in that case and working at that level, I feel like it's a tool they should be aware of. The thing that they're going to have to grapple with in their own, you know, assessment of whether or not they should engage this more deeply are the ethical concerns and also things like ownership and rights.

So, you know, if you if you create these images, these images are owned, apparently by, you know, this particular, you know, image generator DALL-E. And, you know, this whole issue of legal rights is really complicated. I was listening to a talk. The way a lot of these companies are able to apply the commercial application of these things and not pay for copyright and things like that from all the different sources that they're scraping this, this content from is because a lot of them are doing a tricky kind of legal maneuver, calling themselves nonprofits, and that this is for research and academic purposes, but then also fashioning some kind of for profit aspect of the same operation, importing these images over.

And there's just no oversight at all. So what we're creating is a real mess in terms of trying to sort out, you know, property rights, intellectual property rights like we've seen with Napster, maybe, you know, people were downloading music willy nilly back in the day, and no one was getting credited or paid for it. And the music industry and other forces kind of came together. And now we have things like Spotify, which is, you know, has its own problems, you know, Apple Music. But people are getting "paid," quote unquote. Maybe something like that will eventually be sorted out with this. But right now, it's just so hard to even fathom where to begin, to try to, you know, wrestle those aspects of it into some kind of ethical framework.

But, you know, I do think there are some uses if you're going to engage, there can be some value in certain aspects of these tools.

Ethics, Equity, and Automating Teaching/Learning

Meryl Perlson: The ethics question is really interesting and especially from an equity perspective, because most of the -- from what I understand, a lot of these tools are doing their machine learning based on publicly available databases, and a lot of it is scraping things from Internet. And so the opportunity to replicate existing inequities and culture is huge. You know, when you ask DALL-E to give me a girl walking up a hill, what you know, what does that girl look like? Is she, you know, what is her race or ethnicity or, you know, her characteristics, physical characteristics?

And so I worry also, on the one hand about that, how we might replicate these inequities. But at the same time, maybe that puts an ethical burden on those of us who are using it to make sure that there is a diverse range of images that it's drawing from. You know, the more we use it, maybe the more we're training it to be equitable. And then the question of ownership, you know, I mean, we have fair use law that's already kind of a mess in the sense of, you know, are you able to use this snippet in this particular new media product? You know, well, is 20 seconds of this snippet too long, you know, because it's going to threaten the commercial value of the original or, you know, can we use 10 seconds?

And that's going to be okay? And so it seems like this is gonna jump right into that pile. But I keep coming back to the thought of Gen-Z, and I keep thinking, Tik Tok, and how there does seem to be this...Our students have a different perspective on their privacy and data, and I think they have a different perspective on quoting each other or sampling each other or mashups, you know, where it really is a collaboration, where I take what you've done and I add something to it or I add a twist and now it's something new. And that's not only perfectly acceptable, but some of the most popular content, you know, my version of your thing or my, you know, twist on your thing.

So I wonder again, you know, how can we harness this for a student who likes to learn or work collaboratively, it might be a wonderful thing to be able to interact with DALL-E to develop that image. You know, and here's my idea, Matt, Let me, you know, run it through DALL-E and see what DALL-E does, you know, I mean, that could be a really interesting collaboration. So I, I guess I'm of the "if you can't beat them, join 'em" phase.

I would flip it also because we're focusing a lot on the student production and what that means -- of student usage and what that means in terms of the products they produce and the learning experience they have. But, you know, there's also a lot written about these tools on the teaching side, like there's, um, I think it's called Jane Watson, maybe out of University of North Carolina, but it's essentially, it's this program that is a teaching assistant for a class, and it took them many, many hours to develop this.

But now there's a way to take this tool and adapt it to any class much more quickly. Like if you give it your syllabus and your readings and you do a short Q&A with this bot, it will actually become a teaching assistant for you, and students can submit questions to it. And instead of you getting that email at midnight, you know, students can interact with this chatbot about your class.

You know, I can think of many ways that I would want to use that, and I could also think of many ways I would panic about what advice this chat bot is giving my students, but it exists. I read one statistic that said that the consulting company McKinsey estimated that 20 to 40% of what educators do could be done by AI.

Matt Boyle: There certainly is a lot of stuff. Right. The low-level things wherin you're just saying the same thing over and over again. That could be automated to some degree.

But then it brings in those questions of what you've mentioned earlier about equity issues and what DALL-E assumed to be the, “Give me a picture of a girl walking up a hill.” Well, what ethnicity is the girl walking up the hill? Well, what has DALL-E been trained on? If the training is assuming to be this is a white girl walking up the hill because that is quote unquote "the norm,” is that the AI's fault, or is it the fault of the content that is already out there.

And it did have a mixture of ethnicities and the like in the prompts that I was feeding it. It wasn't just assuming. But again, they were predominantly white artifacts that it created for me. Also, some of them were just completely bizarre. It wasn't good at faces, basically. It's interesting that DALL-E vs ChatGPT. I said ChatGPT is not good at examples to flesh out and help individuals see the face of an argument. If you want an argument to be understood, you don't just talk about it in theory. You apply it to an example of this thing happening. And there it just sort of falls short by giving this very thin idea of a face of an argument.

And sometimes I give it a prompt in to DALL-E, you know, give me a picture of this. And the faces are just these weird polygons pixelated into each other that look, look almost like Lovecraftian nightmares more than they do individual people.

The Quality of A.I. Generated Content

Matt: But, um, it's weird. The creativity of prompts definitely is something that you were saying, Meryl. It does draw you in a lot. I was like, "Oh, you can make anything, huh? Well, draw me a picture of a duck frying several eggs for a family of hungry porcupines." You can see I have a lot of duck themed things in there. Yeah, but I'm just throwing random stuff at it for fun. Like, what? Can you do this? Can you do this? Can you do this? If it does, you know, it is intoxicating. Like, "Oh, you can do this -- same thing with ChatGPT. I think I had it write me a story about Captain Hook going into the Apple store and becoming embarrassed because he kept knocking over phones and then get made fun of by the guys from the Muppets in the corner who always laugh at individuals [Statler and Waldorf].

I was like, "Write this for me and see what happens." And it was okay. They even made a joke at Captain Hook’s expense.

Meryl Perlson: And it was a funny joke. So, you know...

Matt Boyle: Well, it was okay, :).

Meryl Perlson: It sounds like it's just a lot of averaging, right? Here, you give me these inputs and I'm looking at all this data or I'm looking at all these sample images and here's how it all averages out. And sometimes that's weird vector vector pictures and sometimes it's a slightly funny joke. But Is it ready for prime time? Can you produce that and submit that and it's good to go? Or does it still require human interaction? I mean, really, like what value do we add? We're going to have to keep asking ourselves this as machine learning dominates so many other areas. What do we offer that machines can't do?

Ken Calhoun: I've been following on Facebook. Someone I know named Lev Manovich. He's actually a very famous digital cultural theorist who wrote -- he actually wrote the book called "The Language of New Media" back in the day and has gone on to write extensively about a lot of the emerging technologies. And he's been very captivated by this this whole image generation, you know, development as both a subject of his kind of academic pondering and also as an artist.

And he came up with something pretty interesting that I think is kind of telling, I'm going to read this to you. This is from Lev Manovich's observations on A.I. that he posted on Facebook just the other day. He says, "Perhaps we're discovering something I would call the laws of media art obsolescence. Or, the faster a certain technology is evolving, the sooner the art created with this technology starts looking dated, uninteresting, and embarrassing."

Because he's talking about the speed at which he has seen these tools become refined, but also the speed at which he's become bored and has learned to identify these kinds of images.

He says, "the opposite effect can be also observed if some media technologies is changing very slowly, like writing, drawing, and printing, cultural artifacts created with this technology fare better. This does not mean that all manuscripts that survived from old civilizations or the 1960s video art has great cultural value..."

Sorry Meryl.

Meryl Perlson: It does!

Ken Calhoun: "...but at least they are not embarrassing to look at today. We feel curiosity. Artifacts and limitations of the media. Technologies from long ago look cute rather than repulsive. And this is the feeling I have when I look at many of the artworks from 2017 through 2022 created with these tools. My apologies," he says, "to all the creators."

Meryl Perlson: Well, it sounds like it builds a lot on Walter Benjamin's work, a work of art in the age of mechanical reproduction, which is when suddenly a photo you get to -- a photo and reproduce it infinitely, or record, a recording that the mechanical reproduction devalues the aura of the art of the work. And it sounds like what your theorist is saying is that it does it culturally, that it just becomes boring because everything starts looking the same and it loses that uniqueness.

And I think you and I could certainly think of tools that we've seen come and go. I'm remembering the video toaster era of media production was one of the first computer-based programs that you could do transitions between videos, and it was amazing. You could do these fancy funny wipes and things like that. But then everything that came out in like what was that? The mid-nineties? The early to mid-nineties starting, I can see stuff from that period and identify immediately what it was cut on.

So yeah, maybe some of this could be a passing fad. I am not sure. At least the styles that emerge from it could be passing fad.

Machine Learning and Exponential Growth

Ken Calhoun: I think the thing that might prevent it from becoming a passing fad is that exponential rate at which it learns.

And I think that could stay a step ahead, I guess of wherever our interests lie, you know, and still be tantalizing in some ways, but also just the practical application. Just like when 3-D became a viable tool for filmmakers, there were obviously the projects that are overtly and very self-consciously 3D project, like a Pixar project, but the way it was being used in other ways was the invisible use of 3-D, which is basically to replace matte paintings and stuff for backgrounds. So you could have, you know, 1920s Moscow, you know, in the background because of greenscreen and other tools. And then just do 3D, you know, 3D model of the city and fly a camera through it.

And you do see those trends when they're introduced into something like cinema. Everyone uses them all at once, and they become kind of played out, but they also disappear into the background. And these days, 3D is being used in a lot of ways that we don't even know when we're watching things. And I think that's how this will eventually kind of, you know, sift down into almost kind of an invisible element in things that people are commercially designing.

The Process of Creativity

Ken: The other point I want to make before maybe we sign off here is that art making and writing -- two things that I love to do and have done my entire life -- aren't just about the outcome, right? They're not just about the product. And if you talk to a lot of educators, especially writing teachers, they'll say, well, it's not about the actual paper. In a lot of ways. It's more so about seeing the student get to that paper, the process of them researching and gathering the information and forming an argument and putting things together.

And that's true on the creative side, too. The creative thing, the event that happens when you're sitting down writing or designing that flow period, or that -- if you want to be more mystical, that collaboration with some other thing -- is never going to really be undermined or compromised by these tools if you're not paying total attention to just producing an outcome.

If you're involved in process and you're involved in the experience of art making. I don't think these are encroaching upon that aspect of what it means to be an artist. So I also find that to be a little bit of a light at the, at the end of the tunnel. So I'm going to hold on to that part of the art making for myself. It's the commercial application of these these, you know, artifacts and products and deliverables that will, I think, be pretty severely impacted by this technology.

Meryl Perlson: I'm totally on board with that, Ken. I think, you know, anything that's going to save time and replace manual labor in a commercial setting that can be provided by A.I. is going to be implemented very quickly because why would, why wouldn't you have software do it? But maybe that's going to push us to go to new places of creativity.

You know, the people who are driving these things are still going to matter. And the questions that they're asking and the things that they're conceiving to be produced, even if it's not by their own hand. And I also think that we may see, you know, some new interesting...forms that we haven't even thought of before because some of the manual labor can be expedited so quickly. So I'm, I'm not, I'm cautiously optimistic. We all have to adapt, right? We've seen disruption across so many different industries. You know, the Industrial Revolution. I mean, suddenly things didn't need to be made by hand. But what if some of these marketing materials, these communications, even journalism, you know what if you don't have to have all of this done by hand, what could the reporters and the investigators be doing instead? What information could they be finding? What stories could they be telling?

Matt Boyle: Yeah. But I do like what you brought up, Ken, about it being more, the process, because I agree in writing that is how writing improves -- through creation and collaboration, creativity and what you're talking about, Meryl. If the scut work can go to the A.I., whether or not you're a teacher attempting to deal with a critical mass of work and figuring out how to -- or if you're a student trying to do the same -- if you can stand on the shoulders of something, it's possible you could go much further than you ever would have been able to go alone.

But yeah, the concerns remain there, but we can leave it on a positive note that way I think.

Meryl Perlson: That's always good.

Matt Boyle: Well, thank you so much for joining me today. I really enjoyed it!

Ken Calhoun: Thank you.

Meryl Perlson: Thanks so much for having us. This was an interesting conversation!