Curious Conversations, a Research Podcast

Travis (00:04)

Will AI replace human writing? Artificial intelligence is, of course, increasingly used in all sorts of different fields, but I'm not sure there's one I hear more concerns about than that of content creation, specifically writing. Now granted, I'm biased. I write for a living, it's been a huge part of my career, and it's something I continually try to get better at. But now that artificial intelligence can also produce copy, what does that mean for writers like myself? What does it mean for readers? And what does it mean for humans as a whole?

Well, thankfully, those are some of the same questions Virginia Tech's Matthew Volimer has recently been asking.

Matthew (00:38)

it used to be that when you encountered writing in any form, you could reasonably assume that a human being had written it. so I think that people who tend to think that the production of writing is a kind of verification system that proves that they're conscious and capable of making meaning, when they see a machine do it, it kind of freaks them out.

Travis (00:59)

Matthew is a professor of English in the Virginia Tech College of Liberal Arts and Human Sciences. This fall, he's taken on teaching a couple of different courses related to artificial intelligence, writing, and really asking the questions of what this means for humanity. So he was the perfect person for me to talk to about all my concerns.

Matthew (01:15)

AI is is good for a lot of things, figuring out how to write that difficult email. completing a task that would otherwise be tedious, right? But in terms of writing, in terms of human expression,

It's like saying, do you think AI will be able to breathe in ten years? will it be able to go to sleep? I wouldn't think so, and I don't think so.

Travis (01:34)

I'm Travis Williams, and this is Virginia Tech's Curious Conversations. I thought maybe a really good place to start this conversation might be To simply talk about why AI feels like such a threat to writing. So why do you think maybe those fears are there?

Matthew (01:54)

well I think first of all, it used to be that when you encountered writing in any form, you could reasonably assume that a human being had written it. And if they hadn't come up with it, that they had copied it from someone else who had written it. now we live in an age where machines can create fluent and eloquent writing.

And so I think that especially people who th who tend to think that you know the production of writing is a kind of verification system that proves that they're conscious and capable of of making meaning, when they see a machine do it, it kind of freaks them out.

 

Travis (02:40)

maybe the question then becomes how do we identify what is and what isn't human?

Matthew (02:45)

Sure. and I I don't know that there's a good answer for that. I I asked Claude of some point this summer or maybe spring to create a field guide to AI tells, and it came up with this huge long catalog of all the things that an AI is likely because of its training to depend on in order to to communicate. But you could take any kind of any bit of writing and you could also ask Claude to you know, could you could give Claude, you know, Samuel Beckett, you could give Claude Stephen King, you could give Claude any s any writer and ask, you know, look at this body of work, notice the patterns and and try to to replicate it and it would do a decent job for a short period of time, right? I don't see

I don't see Claude anytime soon writing a a work of literary fiction that could be mistaken for for human writing, but I I do think it's capable, you know, especially with sequence of iterations and pattern matching that is overseen by a human that could generate something that most humans wouldn't be able to tell came from a machine.

Travis (03:59)

Yeah. Well I'm glad that you mentioned some of the work that you've done because I know one of the things I guess you're teaching now is this course on what is it, artificial intelligence creativity and and the art of being human. Is that the title of the course if I'm Icar?

Matthew (04:11)

That is the that is the title.

Travis (04:13)

And so How did you go about researching AI? Has it been as as simple as just playing around with it?

Matthew (04:20)

Well, at first it was just kind of a toy, right? Like I've always been interested in simulation and representation and so when I first started playing with Chat GPT I would have it do all kinds of crazy things. Like I would say come up with a with a stand-up routine for to entertain a an English department or I would say like rewrite this poem in an with an Appalachian dialect And I was and I was immediately curious at what it with the ease and fluency that it it could generate prose. I have this this thing that I've I've done in class called the story machine where I have everybody think of I start by ask think of an occupation that's really strange. maybe you an occupation you would wanna have and and maybe an occupation you wouldn't want to have. And so we get a list of occupations, we choose three of them.

I say, think of some strange settings where we could plop these characters down. and then think of think of some weird props, right? So so we start with these basic building blocks and move on from there. And I was interested to to know what would happen if if you know when when ask if I asked an LLM to do that, what it would come up with. And it gave me this great list of of occupations. And from that I took I chose Foley Artist. Foley artist is the person who makes is so who oversees sound design in movies and and radio. so, you know, the sound of Darth Vader's mask was like a a scuba machine or whatever, right?

Travis (05:50)

Yeah, those are those are like the people that you see beating up pounds of meat and recording it for punch sounds, right?

Matthew (05:54)

Yeah, or hitting or hitting you know sheets of aluminum to create lightning and thunder or whatever, right? so I fr from there I I looked I I was thinking about two Foley artists and they were maybe w maybe they're working on a video game, I thought. And I was like, well what what what what kinds of conflicts would arise between two Foley artists?

And so I asked I asked the L on that and it came up with a bunch of different conflicts and I thought, well, what if it what if there's something about the content that's troubling? Like what if there's something like especially violent? Like they're working on a kill screen or something and there's a there's something really violent. What could it be? And tell me what are some especially violent cultural pra practices, right? And so human sacrifice was one. And I was like, well what what cultures have pr have practiced human sacrifice and like Aztecs was one.

And so I started using it to to research Aztecs culture. And I asked it, okay, give me produce 50 sounds that would be normal in in an Aztec city. And then also tell me what a foley artist might use to produce those sounds. And it did it instantaneously. I mean like within 30 seconds, right? and so that really illustrated the power of the machine, but also i flicked something in my head, like I would have never come up with any of this on my own, but neither would the machine have generated it without me and my curiosity. And so that's I think that's really where it got started. And I just I just kept experimenting and you know this this summer when I was putting together the AI class, I I sent Claude on research missions to say like, all right, here's the class that I'm teaching. I don't really know what I'm supposed to do, I don't know what I'm gonna do. What if other professors at other major universities done that have been successful or they're compelling and interesting or whatever. And so it it spat all the out all this stuff and every time this every time I iterated I would get some idea myself about about the possibilities for the class. And the one thing I kept coming back to was that the emergence of AI has revealed to us our own kind of algorithmic thinking, right? It's a g it's great at pattern matching and finding patterns within whatever field you you know you could prompt it to to search in and I was also thinking about the title of the class and how AI, creativity, and the art of being human is really in the wrong order in terms of the weight that I wanted to give them. I wanted to place most of the weight on the art of being human and then creativity and then AI, right? and so the theme of the course now is pretty much how did you learn to be a person? And how do you verify your humanity in an age where, you know, machines can write and basically, what are your concerns about about this new technology and your and your place in the world where this is becoming a major factor

Travis (09:08)

you've mentioned several things through this exploration that you found out that AI does really good related to writing. What have you found out that maybe it's not so great at?

 

Matthew (09:19)

I think that it it's it's too fluent. You know, it doesn't account for you know the kinds of happy accidents that humans can discover because you know, they're distracted as they're writing. They get an idea be based on their experience that wouldn't have anything to do necessarily with with what they're working on. So, di a digression or an association that is tied to a particular consciousness, a particular mind who you know, who who is a is a person who's thinking and feeling and sensing and Yeah, forgetting and remembering and and all the things that we do as as humans, right? at the end of the day it is a prediction machine. I mean that's a reductive way of of thinking about it, but according to my understanding, that's at its most basic form, it's a prediction in an imitative, system. So I think about Marshall McLuhan's The Medium is the message, which was a book he was about to publish and the the copywriter changed message to massage. And he was like, No, leave it. I love that. I love the medium is the massage because it introduces this physical thing, like a metaphor of of pressure and hat that was exactly what he really wanted to capture, in in the book. So, yeah, I think there there are l a lot of things. I mean, if you're writing something that's long form, the context window is just is just not good enough. You know, you you can you can use these these things to create really interesting and and oftentimes beautiful sentences and paragraphs, but to create an entire story and one that I would say isn't formulaic or overly formulaic or overly simplified maybe it in the future you know It will have that ca capability, but in the experiments I've been running and it it just does it can't even come close.

Travis (11:16)

That's fascinating. especially the idea that and you can tell me if I'm understanding you wrong, but it sounds like it kind of flattens out some of the nuances and maybe even quirks that humans have when they write. And what's funny to me about that in some ways is that you started by saying it was more comparing it to like impressions. And when people do impressions, it's most it's mostly the quirks and and kind of the flaws that people pull out that we notice about other

Matthew (11:37)

Exactly. Right.

Travis (11:40)

people and that are either endearing or maybe funny or

Matthew (11:42)

Right.

Travis (11:43)

what have you.

Matthew (11:44)

exaggerations too. And I think you also have to think too, if the data set is is so large for what it is drawn on in its training and what it continues to draw on, right, like

 

so much of what humans make is not good and so much of what has been published is not great. and and s in so many humans have written middling books with forgettable sentences and so why wouldn't AI, you know, take the path of least resistance and generate stuff that that is of you know mid quality.

Travis (12:18)

That's a I think that's a great a great point. I I hope that it's not diving in any of my writings and in using those

Matthew (12:25)

Yeah.

Travis (12:26)

to to learn to learn from. I know one of the things that I used AI for very early on was I wanted to test it, so I wanted to know what its top five lines were from the movie Fletch. And he gave me a bunch of stuff from Caddyshack and I was just like, Well, I'm out. I'm out on the I'm out on these chat bots. I'm back in now because I think that they're I think that they're more educated now. But at the time I was done.

Matthew (12:45)

Th I well, first of all, you're a man of of high taste because Fletch is is one of my all time favorites. But I think w

Travis (12:53)

Kindred spirits. I knew we were.

Matthew (12:55)

it I think too though that like you may have made a mistake in in stopping where you did. Right? You got you th and I think th this is oftentimes something And I'm not saying that you should have spent, five hours trying to get the results that you wanted, but I think if you do get a result that is insufficient or wrong or whatever, you can amend your original prompt or you can say, no, you're wrong. I said Fletch. and and give it even you give it the context. Like rem you know, Fletch is this journalist who's going undercover and blah blah blah whatever. and and it you and it you might get a better result.

Travis (13:39)

Yeah, that's a great point. I to be completely honest with you, once it was wrong, I was so happy that it was wrong and I felt smarter than the robots. that's probably that's probably where I stopped. But I'm curious when you're teaching this course, and I know you're just at the beginning of teaching this course. Where are you finding that a lot of your students are coming to you as far as their experience using AI to either assist or maybe even do writing for them?

Matthew (14:05)

Yeah, so it was really interesting. I looked before class where these students were coming from in terms of like what their majors were.

 

And it was literally all over the the university. we're talking data analytics, English, national security, philosophy, nanotechnology, biology, finance, sports medicine. I mean, I can't remember like just name a major and it's, you know, probably close And so I was like, that's gonna be really interesting, because they're gonna have probably a a wide variety of perspectives regarding AI. And so the first well, the first day I had them, I asked them, who are you? Why are you here? What's your experience with AI, and what are your concerns? What are your what are you curious about? What are you worried about? Excited about? And so, you know, we wrote for about 10 minutes. And then I had everybody go around the room and read what they'd written. But I also had everyone bring in notebooks and I said, I know I want everybody to write down as much as you can. write the name of the per I had the person spell their name, say their major, and then go from there and and thennTake as detailed notes as you possibly and what emerged was just a wide variety of perspectives and concerns, but every single person did have a concernnthe the people who were like English majors and writers were concerned. and there was also a guy who he's really into music and he's concerned about the effects of AI on the music industry and just a a lot of people are are are talking about protecting art from AI, which is an interesting thing to think about. but what was interesting to me was like some of the students had never that had never even occurred to them before on my feeds it's constant. if I go onto Facebook now, it's just like writers saying, If you even experiment with AI, you know, you're garbage. and and so I think it's educational for them just to be in a room with a diverse set of that that where people are are representing a diverse set of disciplines and future occupations because we can all agree that AI has been a disruptive force in the world and it it disrupts every field in in particular and idiosyncratic ways and so just just learning about that I think is is really valuable.

Travis (16:40)

How do you go about or or do you, maybe that's a better question, do you go about setting any sort of parameters for this course on how they should and shouldn't use artificial intelligence as you're kind of going through this?

Matthew (16:53)

so there's two things. One is I've told them that they need to bring their notebook to class and they, you know, if if anything appears on the board that should appear in their notebook. I give the I will be giving them assignments where they, you know, like one one of the things was on the second day I said, bring in an artifact on your phone that represents something about AI.

Right. It could be pro AI, anti AI, could be produced by a by AI, something that has disturbed you, something that's excited you, something that's can you know, whatever. And so we I got them into pairs and they showed each other what they'd brought in and I said, Okay, now write a description of what you saw and then write a description of how you responded to it and then just we just went around and and talked It was like, here's this this AI depicting a breakdancing baby.

And it's really realistic and that's disturbing to me because it shows that we're we're at a point where it's hard to tell, you know, real vi quote unquote real videos from AI generated. there were other concerns about AI deceiving people, misleading people, AI generated songs, which, you know, for s for so many people, music is this sacred art form. And what does it mean? You know, like talked about the fact, like, I've heard a I've heard a song before and I was like, that's a good song, and then someone tells me it's AI generated and then I feel nothing. And I was like, that's interesting. What how does that switch turn off and on? But there was only one student who brought in something po he was like, here I found an an app that allows you to that will translate a language like into your earphones. And and the person who who had looked at it was like, this this sound this seems pretty positive, although I teach ESL and that might put me out of a job. I don't think anyone in there is like y you know, gung ho, let's all use A all the time. but they're really like there is something at stake. You can feel it in the room. Like they're they're they're there, they're attentive, they want to talk, they have ideas and and notions and you know we talked about like how their parents how their parents are using it. in some cases, parents are like r who really are gung ho. They're like, okay, we're gonna go back to school shopping. And the mom's like, just let chat do it. Let chat make the list. And the and the students like, no, I can actually do this. This is something I want to do. Right. so it there may be some sort of generational divide or at least perceived generational divide in terms of like how how people are responding to Giving a task over to a machine.

Travis (19:27)

Yeah, that's that would be really fascinating to to explore. because just in my social feeds, I do see that people some people are, like you said, very disturbed and very upset by the use of it. But then like that's half of the people that I'm connected to, I guess the other half are using it to make flyers. and

Matthew (19:43)

Yeah.

But the other the other thing that they're doing twice a week are these authentication opportunities. and so basically what I do is I I give them a series of questions or a question, you know, and on the first day the questions were, What did you not say when you introduced yourself today? Why not?

What did another person make you notice that you had not noticed before? What do you already believe about AI that you were least certain you can defend? What question do you have right now that you genuinely do not know how to answer? Why is it a question for you? And they got to choose and then and then write a 250-word response. And \now they're doing this on the computer. I don't I, you know, they it could be that some at some point start using AI to answer these questions, but for the most part, it would be easier to just answer it on your own. You know, like the you know, ChatGPD doesn't know what what you didn't say in class today, right? and so I'm hoping that these become a way of of capturing a lot of the concerns they have about their own humanity, about technology, and then like for the second authentic authentication opportunity, you know, I every time I read these things I will ask them a question. and a and the question will arise out of you know what they've written.

So if if you know they say they want to make sure that humanity protects art from AI, I I ask what might protecting art from AI actually look like? You know, to get them thinking about, you know, it's I mean it's easy to make a claim like, let's just let's protect art. Okay, yeah, that sounds good. but but practically how how would one do that? And a follow-up could be like, couldn't you imagine a collaboration between a human and machine that would be acceptable, compelling, emotionally resonant, moving? you know, it's not like we haven't been using machines for a while now to to generate art. You know, you think about

 

an artist like Laurie Anderson using the vocoder, or, you know, electric guitars or, you know, there's computers, PCs, you know, clicking what is it, shift command or no, control option command R to to hit some of the thesaurus, right? I mean for for a while now it's it seems like there hasn't been and there maybe never has been like some sort of pure space within a single human where art you know magically unfurls.

Travis (22:07)

Yeah, that's that reminds me of well, in part it reminds me of Blade Runner. But that's because I'm I

Matthew (22:12)

Another great movie.

Travis (22:14)

I mean, this is how my brain works, right? These are the the things that my brain does and how it operates. But

Matthew (22:18)

Right.

Travis (22:19)

I am curious if is there any other example of technology specifically related to writing where people had this type of like f fear, concern, like are they thinking about the cost of using it as much as we are with AI?

Matthew (22:33)

Well, I do have a friend who his name's Pinckney Benedict and he's a writer. he's from West Virginia. He's written a number of books. He used to teach up the road at Hollands. I guess he his dream was to be a pilot, like his dad. But his eyesight was so poor that he was subsequently disqualified. And so he got really interested in flight simulators and video games and basically taught himself how to code on a Commodore 64 and put himself through through college, you know, writing games for Commodore 64 but at one point he when he first bought a a PC, someone else in his department said, That's gonna ruin your ability to write and think. And I think it that might have even been before he published his first book. so I think, you know, obviously or maybe not obviously, but there are examples of how I'm sure people trusted handwriting over the typewriter at first. and then, you know, people trusted typewriters over PCs. new technology always is is disruptive and there's always something to mourn. I think that, you know, even though there are examples of ways that people have have feared previous technologies in terms of writing or any other thing that AI can do, I don't think there's anything been so so huge and pervasive that affects, so many people as as the AI revolution it also has inspired as much backlash as it has enthusiasm, if not more backlash. I mean I think especially for writers, it's lots of people are like, nah.

Travis (24:17)

Well I think that that's that's definitely fair, at least from many of contested debates that you see online. So I guess, maybe to wind this down, where do you think we will be in ten years from now when it comes to writing and artificial intelligence?

Matthew (24:33)

I don't know, but I think it's interesting that once people see what it can do in terms of like just sheer text production, you that we're gonna get more and more slop. We're gonna get more garbage. But I think that at the same time, people might value human writing more than ever. because, you know, there is no substitute for a a truly magnificent human-made reading experience AI is is good for a lot of things, you know, it might be good for figuring out how to write that difficult email.

Or, you know, you've got a bunch you got a notebook of messy notes and you don't feel like typing it up, so you have AI transcribe it, right? But in terms of in terms of writing, in terms of human expression,

It's like saying, do you think AI will be able to breathe in ten years? will it be able to go to sleep? And I wouldn't think so, and I don't think so. So you know, it's like when streaming came the in music and you know you've got Spotify you got everything at your at your fingertips. What don't you have? You don't have the artifact. You don't have the record sleeve that you can open and see the artwork and feel the texture of you know of the cover. You don't have the lyric sheet. you don't have the crackle of the needle when you set it down on the record. and even people who've never experienced that, who didn't grow up with that, once they encounter it, they long for it, right? so and I don't I don't think that like human writing is gonna like be some sort of like you know bespoke artifact that that people are nostalgic about. I think it's always gonna be front and center. will more writers use it? Absolutely. it's probably gonna be like, what are they using it for, right? but I think it would take more work to generate a really good book, even in collaboration with a human, than to have the human mind just Trusting itself or other human minds to create it.