AI and Creative Writing – Amit Gupta
“I think it’s up to us as a society to figure out the way we want these tools to be used. At least in my personal experience, I think technology started out in a very optimistic place, where it was so on the fringe – computers even were so on the fringe, that it wasn’t something we ever really paid too much attention to as a society in terms of what are the dangers, and what are the repercussions of this. We just saw what wonders could come out of it.”
Amit Gupta
Amit Gupta is an optimist, a science fiction writer, and co-founder of Sudowrite, the AI-powered creative writing app. He is also an uncle and a son, and a friend to all dogs. He previously founded and sold Photojojo.
In this episode we discuss:
– Some of the challenges of creative writing
– How writing has lagged other creative pursuits in terms of the availability of technology to support creativity (there is no photoshop for writing)
– How AI can be a support for creative writing
– Some of the considerations to prevent the misuse of tools like Sudowrite
– Why flagging too many of the issues with future technologies may hold us back
Host and Producer: Georgie Powell Sentient Digital
Music and audio production: Toccare Philip Amalong
Transcript
Amit: I definitely feel like all creativity in the future is going to have some sort of AI assistance, or I should say the bulk of creativity in the future. Similarly to how today I think the bulk of creative projects have some computer in the process, you can still sculpt and paint a number of other things without computers.
But even some of those painters are probably using images for reference that they found on the internet or other things like that. So, I think that AI is going to have a part, have a hand in a lot of our creativity. I think it’s going to be one of the most powerful tools we’ve invented for creativity.
Georgie: Welcome to Freedom Matters, where we explore the intersection of technology, productivity, and digital well-being. I’m your host, Georgie Powell. And each episode, we’ll be talking to experts in productivity and digital wellness. We’ll be sharing their experiences on how to take back control of technology. We hope you leave feeling inspired, so let’s get to it.
Our guest this week is Amit Gupta, an optimist, science fiction writer and co-founder of Sudowrite, the AI-powered creative writing app. He’s also an uncle and a son and a friend to all dogs. He previously founded and sold Photojojo.
In this last episode, in our series on the tools we use, we discussed the role of AI and writing, its impact on creativity, the positive and the negative use cases, as well as Amit’s optimistic view on the future of technology.
Amit, welcome to the Freedom Matters podcast. It’s really fantastic to have you as a guest. Thank you so much for joining us today.
Amit: Thank you so much for having me. I’m excited to be here.
Georgie: We started by discussing Amit’s career, first as a serial entrepreneur, which ended after a major health scare led him to refocus his attention on his love for creative writing. He wasn’t able to bury desire to build businesses for long, however, he is still writing. I asked him, what’s your passion when it comes to writing?
Amit: Yeah. So, when I got into writing, it was because I love science fiction, and I loved it growing up, and I think it inspired a lot of my kind of professional direction. And I also noticed how so much of science fiction today is very dystopian. It’s all about how the future is going to be terrible. And everything that’s bad today will be even worse. And I think that’s really dangerous, actually.
It’s a good thing because it gives us caution. It helps us see what could be around the next corner and what we should try to avoid or prepare for. But I think it’s also dangerous because this stuff is what inspires kids to go on and do what they’re going to do. They read the stories when they’re young, and they impart the morals and the lessons and the stories. And if we only see a future that’s bleak and dark and dystopian, then that’s the only future we can imagine. Because fiction helps us imagine what could be.
I really came into fiction wanting to write science fiction that was more optimistic, that was more bright, that showed a future where technology could help people and help individuals. I write short stories.
To answer your last question, I’m writing less right now than I had been for a while, because I’ve just gotten so busy with the company. But it’s definitely still a passion of mine. And I published a couple of things.
And my co-founder, James, who I met in my reading group has also published a bunch of stuff. And he also does science fiction, but I don’t think he’s as focused on the optimistic side of things. He’s your optimistic person, but that’s not his [inaudible 03:25]
Georgie: Yeah, but you find yourself as being an optimist, so I guess you have to be an optimistic writer if you are defining yourself as being an optimist.
Amit: Yeah, hopefully.
Georgie: Okay, cool. And then so you met James, and then this is wow, hang on, we are getting blocked. This is a really solitary pathway for us. We’re finding this hard. There must be a tool, we can create a tool that’s going to help writers.
Amit: Yeah. So, we’re in the same writing group and I think for both of us, definitely for me, the reading group is revolutionary. I think before that I was very singularly focused on writing. Now I had a weekly cadence where I had to, like prepare something, give people feedback, and I could see how clearly that was improving my writing and increasing the output.
So, I think when some of these large AI models started to come out a few years ago, especially for writing, James started playing around with it first, and he built just a fun playground for the writing group to experiment with.
For people who do podcasts, for people who do video, for people who do 3D modeling, any kind of creative pursuit you can think of, there’s probably a very advanced software to help you make that thing. Like, you’re not going to edit using scissors and whatever or cutting up audio tape. You’re going to use this multitrack editor to cut and paste stuff and move it around and do interesting effects and stuff.
But writers haven’t had anything like that. We haven’t had the photoshop of writing moment before. We’ve had grammar check, we’ve had spellcheck, but that’s about it. Otherwise, we’re basically just typing letters one at a time to get our stories down and then editing them painstakingly one word at a time.
So, it was clear to us when we saw these large language models. I’m out that this was a photoshop moment, this was the moment where we could actually apply the computational power of computers to helping us be more creative with reading. So, we got really excited and we started playing around nights and weekends just like seeing what we could do with it.
And it was pretty astonishing, very quickly, obviously, it took a lot of work to get where we are today. We’re very quickly at the start, you could see that it could work. And we showed it to a bunch of writers who respect it and we’ve got a really great reaction. And neither one of us I think we’re intending to start a company at the time. We had both done that before and we were writers.
So, for the next, I’d say, almost a year, we just had it as a side project, as a hobby. We would let people in who we knew, and they would use it, and they would try it out. But I don’t think we realized we were starting a company until quite a ways into it.
Georgie: And not giving away too many trade secrets, can you tell us actually how it works?
Amit: Yeah. So, at its very base layer, it’s using a text transformer model. So, this is a very large text model that’s trained on a huge corpus of text. So, it’s looked at basically the entire internet, lots of public domain literature, other sources that the company who trained it was able to get a hold of.
And in this training, it’s basically learning, given this series of words, what word could come next. So, if I give it a sentence, it’s trying to predict the first word of the next sentence, if I give it three pages, it’s trying to predict the very next word that comes on the next page.
And so we use that to do a lot of different things. We help writers get unstuck by predicting and showing them what could come next in their story. So, actually write like the next few paragraphs, five different variants of it using the writer’s style, their prose, their voice, following their narrative arc, following their plot arcs, all that kind of stuff.
But writers have problems with different types of stuff. So, some writers don’t get stuck, they just have trouble in the editing process when they want to make something more descriptive to make it come alive. So, we can do that too. You can just select like, a phrase or a word or sentence. And we’ll tell you in that — for that word, like what it would smell like, what it would taste like, what it would sound like.
And we can give you lots of different options. So, we can help you describe scenes with vivid detail. We can help you with plotting, we can help you with finding the ending, we can give you feedback. And that’s actually super interesting because when we talk to writers who’ve been using this for a year now, one of the things they tell us is that they didn’t expect this.
But the reason they like Sudowrite is because it’s become this always available partner to them. It’s like a reading buddy. And they have people in their lives who they occasionally send work to get feedback.
But when they do that, they know that they’re asking a lot. That person has to read this novel, or read this chapter, or whatever, and take the time to write feedback. And if someone’s doing that for me, I have to do that for them too and it takes time. So, they may not be getting feedback very often, maybe it’s a few times a year.
But with Sudowrite, you can get feedback, literally by the push of a button anytime you want. And they find this revolutionary, it feels like they’re no longer waiting to see what to do next. They immediately get that kind of feedback.
Kind of the way, I don’t know if you’re a writer, but often writers will set aside what they’re writing for like a few weeks or a month even, and that’s another way to get feedback. Because when you come back and look at it again, you’re looking at it with fresh eyes. And writers tell us that’s what they’re getting from Sudowrite. They see their own work with fresh eyes, but they don’t have to wait a month.
Georgie: Is there a way in which these AI tools are actually limiting creativity because they are repurposing what has previously been ingested and understood?
Amit: So, it’s not repurposing, like none of the phrases suggested are phrases that have ever been written before. None of the — the words obviously are because it’s not inventing new words, but the combinations of words are completely fresh and new.
And I think that it’s really up to you about how fresh and new this is going to feel to you, how innovative because it’s mimicking your style. It’s learning from you, how you write what your voice is, where you go with these types of things, and it’s following in that train. So, it’s really good at understanding and then continuing. It’s not good at inventing or coming up with a new concept or whatever.
So, you still need to come up with the story, who the characters are, all that kind of stuff. Sudowrite can help with pieces of that. But in the end, like you’re making the decisions and make the story yours — the decisions that make this story human.
And I think that, I think it’s flexible enough having seen enough material across its training set that it can probably follow pretty much any style. If you have some sort of style that it just can’t grok and can’t help with, then perhaps you’ve discovered something new that no one’s ever done before.
But I tend to think a lot of our creativity is really a form of re-mixing. We learn from other people, we read other people and we, at first we mimic them to try to understand how to do it and then we develop our own style. But it’s all standing on the shoulders of the people who have come before. And that’s whether you’re a writer or scientists or technologist or anything else, you’re really just taking what’s been done before and putting it together in new ways to create something new. So, I think that’s what it helps with.
Georgie: And edging it forward. And on that I saw that you are actually experimenting with poetry now too, because I can see that being quite a unique form that’s harder to use AI. So, how’s it working with poetry?
Amit: Yeah, it works great. I think poetry, songs, all those kinds of things are really interesting, because our primary focus is fiction writers. And I think often fiction writers will have a poem or will have a song or something else as part of their novel, but they’re not songwriters, they’re not poetry writers. So, I think — or poets. So, having a tool that can do that for them, or at least help them get started is really helpful. It’s getting better at rhyming.
It took a long time to get it to rhyme well, and it still doesn’t do it great. But it’s still — now it can do it in a way that’s like pretty good. You can go up and touch it up and fix it, which feels magical, because rhyming is hard. And a lot of these things are hard. But somehow, I can write other things. I can’t rhyme very easily. I can’t write poetry. So, having it do that feels magical.
Georgie: And is that just because it’s ingested more and more information that it can understand the rhymes or because users giving it feedback about what works and what doesn’t work?
Amit: I think it’s about the model of this getting bigger. So, being able to understand and synthesize more. And also yeah, getting good reinforcement feedback, telling them what you’re looking for. I think they’ve also gotten better at actually understanding what we’re asking for. I think earlier, the model didn’t have a sense of what rhyming was and now maybe it does a little bit better.
It’s interesting, AI is so different from other types of programming. With programming, traditionally, you’re writing everything as clearly as possible. So, if you look at the code, you can see exactly what’s happening and exactly why it’s happening. And with AI models, it’s really a black box. You’re putting in a lot of information, you’re making some decisions about how to train.
But you don’t know why the next word that came out is the next word that came out, not exactly not the way you do with programming. It’s more similar to human where if I gave you a paragraph and say write the next few words, I don’t know how your brain did that. And we don’t know how these AI models do it either.
Georgie: Because you can’t understand, it’s like understanding its context and its environment, in the same way that you don’t know all that for a human either. So fascinating. So, I can see how useful this is for writers. I guess there’s already quite a lot of work happening with AI for visual arts, and also in music production as well. Do you see all these creative fields progressing in parallel?
Amit: Yeah. I don’t know which ones will progress fastest. Obviously, images have had a lot of excitement this year. There have been some really cool new models developed. And just in the last few weeks, even writing models have gotten very [inaudible 00:12:48] as well.
But yeah, I definitely feel like all creativity in the future is going to have some sort of AI assistance, or I should say the bulk of creativity in the future. Similarly to how today I think the bulk of creative projects have some computer in the process, you can still sculpt and paint a number of other things without computers.
But even some of those painters are probably using images for reference that they found on the internet or other things like that. So, I think that AI is going to have a part, have a hand in a lot of our creativity. I think it’s going to be one of the most powerful tools we’ve invented for creativity.
Georgie: I reflected back on the conversation that we had with John Mack as part of this series. His art exhibition highlighted that whilst technology puts a new lens on our experience of the virtual world, our own existence has already been curated by our lived experience. Did Amit agree?
Amit: Yeah, I do see that. I think the tools definitely influence what we create. And I think we, at least like you said, ourselves, our own experiences influence what we make. And that’s what make art beautiful, like the — we can both paint the same thing or take the same photo and they’re different because of who we are and what we’ve brought to it.
Georgie: Yeah, and AI won’t change that.
Amit: Yeah, I think AI will — I think AI will just continue that. I think just the same way if I take a photo with this camera versus this camera, the photo’s going to be slightly different. But I think if I learn how to use this AI tool versus that AI tool, my writing will be different. I feel like it’s, we’re so early in this, so it’s hard to even imagine what the future will bring to it. I think we have a little bit of a peek into it today. But man, it’s going to be a very exciting next five to 10 years for sure.
Georgie: And how much do you think about that? Because one of the things we talked about quite a lot at Freedom is like the concept of responsible technology and constant scanning and trying to think ahead about some of the negative externalities of your product as well as the positives that you’re building for, and the solutions that you’re providing.
How much do you think about that? Are you cautious in any way about the way that, you know, Sudowrite might be used?
Amit: Yeah, for sure. Definitely in our terms of service, there’s a bunch of things that we don’t allow. And I think that with the writing product in general, it’s a little bit of a tricky situation as it would be with any creative tool. Where we want to allow writers to write about whatever they imagine, but we also don’t want them to do harm.
And so where our responsibility lies is a subject of frequent discussion. I think with these large models, there’s definitely a lot of possibility for harm, there’s possibility of people generating text that harms other people. There’s possibilities of people using AI generated text to create misinformation campaigns or do astroturfing. So, those types of obvious wrongs are prohibited.
And a lot of AI reading tools will actually have a profanity or decency filter attached to them as well. So, if you are using them and you want to write about a sex scene or write about a murder, or something, you actually can’t do it, it will prevent you from doing that. We don’t have those filters.
And that’s a decision we made consciously because we’re aimed at fiction writers. And I think fiction often does involve situations that are offensive, or scary or dangerous, or whatever. And that’s in the service of the arts. We felt it was important to allow that form of expression on the platform.
But if I was making a social platform today where I was having people post writing from the platform, and we are hosting that, then I think it becomes an even — a thornier problem. Because it’s one thing to have you write about a murder or write about some explicit scenes in your play or in your script or in your novel, it’s another to publish it and then be responsible for people getting offended by it.
But yeah, there’s definitely a whole nest of issues there that are, we’re definitely thinking about a lot and will probably continue to have to navigate as we go forward.
Georgie: Yeah, interesting. There’s another relevant conversation we have with Susie Alegre who’s written — She’s a human rights lawyer and she talks about freedom to think and how actually freedom of thought, and even writing down your thoughts if they’re just for you it’s so important.
So, it may actually be that Sudowrite helps people to indulge their — kind of their wildest and darkest thoughts. And in doing so, you realize that they’re not for public consumption, and you don’t want to share them.
But people go through a process of writing or creating or thinking or processing these thoughts. We all have dark thoughts. Right? But I think it’s really interesting that you made that distinction between creating versus publishing.
Amit: Yeah. And just to your point, I think it is like a thinking tool. It is — like writing is a technology we invented to help us both record and transmit information, but also just to express and do our own thinking with this external kind of like memory source. The idea of curtailing what we can think about on paper is a tricky one for sure.
Georgie: It’s so interesting, isn’t it, if you think about the history of stories, and how they started around a campfire, the small group of people, and then how they’ve evolved through time to be now what they are today, but how central they are to what it means to be human.
Amit: Yeah. Well, since I’m an optimist, I’ll also give this optimistic point of view, which is can we use a tool like this, not just to prevent them from writing something terrible, but what can we help them understand why it is, might be terrible, or might be offensive or hurtful. And I think there are opportunities there.
We have a tool that gives writers feedback. And right now it’s really geared towards narrative feedback. So, it helps you understand what’s maybe not working in your story, where you could put more effort, that kind of thing.
But you can certainly imagine the future where an AI could look at a piece of writing, identify and flag things that might be problematic, or might harm people or might come off the wrong way, and let you make the decision to fix them.
But it’s now pointed out just the way a helpful friend might look at a piece of writing and say like, are you sure you want to say this? This might come off the wrong way. So, I think there is a lot of opportunity for AI to help in that process too.
Georgie: I hope so. I just recently read ‘The Secret Garden’ with my daughter. It’s obviously like a classic children’s story tale, and it’s so beautifully written. It’s absolutely magical to read, but it is incredibly racist because of the era that was written in. And it’s not an excuse. And it’s, I guess my fear is that some of these historical texts which have so much beauty in them, but we’re written in a very different time will basically be canceled.
Amit: Yeah, yeah. We do see media being censored today, right. There’s movies that get little scenes snipped out of them when they get put on TV and that kind of thing because they’re no longer politically correct. I don’t know what their answer is. I feel like that’s a question beyond my paygrade. But it’s definitely something we’re grappling with as a society as a whole.
I can see a future where if you were as an author writing something, you could have an AI help you not only just translate it for different readers, but translate it for different audiences.
So, maybe you read a story that is written in a way that maybe is offensive, but it’s okay for an adult audience. And then you translate it to an audience of ninth graders or fifth graders or whatever, and they get a version of that story that’s somehow a little bit different. That’s more appropriate for them.
I don’t know how I feel about that. I think it’s a difficult thing to grapple with because it goes back to censorship, and it goes back to the stuff we were talking about earlier. But I do think it will be possible. And so it will probably be a question we’ll have to grapple with in the future. Like how much do we want to shelter our children from views that we know are no longer what we hold to be true in society, but are still views that people have.
Georgie: Yeah. And also, you could argue that fiction writing is a great opportunity to open, maybe not looking at the child and the adult divide, but thinking about if you tailored a book to different perspectives or different backgrounds and wrote the story, the same story in slightly different ways for different audiences that were used to different approaches.
One of the joys of fiction is it should force us to open our perspectives and our boundaries and think about things in different ways because it’s much easier to do that through a story. And if you start creating a different lens that’s geared towards different audiences, you actually take away that joy and these whole bubble, these curated bubbles that we live in social media, become a kind of existence in the world of reading as well.
Amit: Yeah, totally.
Georgie: That could be tragic.
Amit: Could be tragic.
Georgie: Good. Important to think about. I am normally an optimist as well, but I am — When it comes to technology I just want to make sure we ask lots of questions all the time. Not too late.
Amit: Yeah, for sure.
Georgie: Amazing. Okay. And the other thing — the other optimistic thing I do want to ask you about, ancient languages, languages that are on the brink of extinction; is that a use case as well?
Amit: Yeah. I think that could be a really interesting use case of technology like this, I think that you can certainly build models trained on any language. And right now we train them on everything. But you could focus them in on a specific language that you’re trying to preserve, potentially.
And what would you do with that? I think you could create something that’s a speaking partner in that language. So, if there are people who want to learn that language or want to engage with someone who knows that language, now you’ve got an entity that can speak it fluently, and can converse with you.
I could also see something, this is not exactly language AI. But when we talk about the future might have some sort of virtual environments just like we do today, and video games. And I wonder if in the future, we have a way of preserving culture like that, we have a way to kind of record not just the photograph, but both 3D models and literature and language and sounds and feels and touch and everything else into an environment that you can then visit.
So, if you want to go to visit this place in this time period, which doesn’t exist, because we’ve progressed through time, maybe you still can, and you can learn from it. I think that could be a really interesting way for us to learn about different cultures, but also learn about history instead of just reading about it, but actually experiencing it.
Georgie: And it being so much more immersive. Yeah, amazing. And then I just wanted to ask you personally about the role of technology in your life and how you think about technology as a tool, and how you always have, whether that’s changed over time, if you do have a particular approach to it.
Amit: Yeah, I always think of technology as a tool. It’s a tool that can be used for good or for bad. And I think all powerful tools can be used for good or bad. And that’s from the simplest tool onward. Like a stone can be used for good or for bad, a knife can be used for good or for bad.
So, I think it’s up to us as a society to figure out the way we want these tools to be used. At least in my like, personal experience, I think technology started out in a very optimistic place where it was so on the fringe, computers even were so on the fringe that it wasn’t something we ever really paid too much attention to as a society in terms of what are the dangers, and what are the repercussions of this. We just saw what wonders could come out of it.
And I do like wholeheartedly believe that we’ve gained so much through technology. I can’t even imagine how many technologies are required to do the thing that we’re currently doing. Like it’s mind boggling how many millions of people have told, like how many hours to create all the little things that cobbled together, create just that microphone that you’re using, let alone the video conferencing software and everything else.
So, it’s obviously enabling some pretty amazing stuff, and I’m really happy that we’ve done those things. And I think one of the fears I have going forward is not actually that we won’t pay enough attention to the dangers, because I think it’s very clear we flag dangers really quickly now. Like, we — often before the technology even exists we’re starting to flag the dangers.
Like the VR, which you were talking about before is an area where we’re not — we haven’t really made a real virtual world yet. We haven’t made it in the way that we talk about it. But there’s already a lot of discussion about how that would be a bad thing. And I think the potential harm there is that the discussion is unbalanced. We should be talking about the harms so that we can prevent them, but not so we can prevent the technology in the first place.
I think there’s a lot of technology that’s coming in the next decade or two that seems really scary. AI is, but also CRISPR and like genetic engineering is another one. There’s like whole realms of technology that are really scary, because they seem like they’re going to potentially change everything we know about how we interact with each other or who we even are.
And I don’t think that means we shouldn’t do them. I think it means that we should be careful about doing them and do them for the right reasons. So, I think my views of technology going back to your actual question have changed from this is universally good. I’m so excited about everything too I hope we’re able to make these things happen and not just shut them down because we’re scared of them.
Georgie: I cannot think of a better way to end the series on the tools that we use. Thank you so much for being a guest on the Freedom Matters podcast. It’s been fantastic to talk to you today. I’m really grateful.
Amit: So, happy to meet you and thanks for doing it.
Georgie: Thank you for joining us on Freedom Matters. If you like what you hear, then subscribe on your favorite platform. And until next time, we wish you happy, healthy, and productive days.
