A Chatbot Reacts to a Book About Tech  ## Metadata - Author: [[Fresh Air]] - Full Title: A Chatbot Reacts to a Book About Tech - Category: #podcasts - URL: https://share.snipd.com/episode/a16d38ab-0e65-4733-8ee2-faf92b43f716 ## Document Notes ## Highlights - **Uncertain AI Bias Origins** - AI's positive bias may stem from training data or creators' influence, but exact cause is unknown. - Lack of transparency about this bias is a significant problem in AI. Transcript: Vauhini Vara It to be very financially successful. And the way to be financially successful is to get a lot of people using a product. The way to get a lot of people using a product is to make people feel comfortable with the product, to trust the product. And one device that a company might use in order to do that is to have the product use language that makes you feel a bit like you're talking to another person. Terry Gross So in reading ChatGPT's responses to your chapters, one of the biases you noticed was that it suggested you add more about the positive side of AI and its creators. I thought that was interesting. Did that say to you that it was revealing the chatbots bias or pointing out your negative bias or both? Vauhini Vara It's such an interesting question, and it gets to the heart, I think, of what is problematic about these technologies, because I can't claim to have any way of knowing why it said what It did. So basically, I fed it these chapters about big technology companies. And I said, what feedback do you have for me? And it said, you could be more positive. And then later, it goes on to provide these sample paragraphs it thinks I should include in the book about how Sam Altman, the CEO of OpenAI, is a visionary leader who's also a pragmatist, Like this really glowing stuff. It would be fun to be able, and it would support a strong critique to be able to say, oh, clearly OpenAI has built this product in such a way that it's deliberately having the products about This propaganda about its CEO that's positive. It's certainly possible that that's the case, but there are all kinds of other explanations for it, too. It's possible that the language that the technology behind the chatbot absorbed in order to learn, quote unquote, how to produce language happens to be somewhat biased toward people Like Sam Altman. ([Time 0:05:25](https://share.snipd.com/snip/4768d2a6-390b-4396-8861-1d7a58831ce6)) - **The Fiction of Chatbot 'I'** - ChatGPT uses comforting first-person language to build trust, but it lacks true self or awareness. - This language style can be both helpful and misleading to users. Transcript: Terry Gross It seemed programmed to flatter and to sound empathetic and kind. And it responded, the way I communicate is designed to foster trust and confidence in my responses, which can be both helpful and potentially misleading. And I thought, that's so true. It can be helpful and misleading. And I thought, maybe the chatbot is actually very transparent. Vauhini Vara So OpenAI has said that ChatGBT and its other products are designed to foster trust, to appear helpful. So it says those things explicitly. And right, by saying this is how I function, yeah, I guess chat GPT is being as transparent as it can be. I don't know that I would trust it to always be as transparent as that or even to know how to be transparent, right? ([Time 0:09:11](https://share.snipd.com/snip/0e27456b-22b7-45d6-8174-a8218957059a)) - **AI Struggles with Grief** - Vauhini Vara used early AI to try writing about her sister's death and found it produced false, unrelated stories. - Repeated attempts brought AI-generated grief closer to her reality but still lacked authentic personal expression. Transcript: Terry Gross So in 2021, when you wrote your essay Ghosts, which went viral, it was published initially in The Believer. It was adapted into This American Life story. And the premise of this was you had wanted to write about the death of your sister. She was diagnosed with cancer, Ewing sarcoma, when you were a freshman in high school and she was a junior and she died when you were both in college. And you felt like you just didn't have the words to describe how bereft you were, what a life-changing event this was in every way. And so in 2021, while you were playing around with a predecessor of ChatGPT, you asked it to help you write the essay. And I want you to describe what your process was. Vauhini Vara This was before ChatGPT was out. This was before AI was a big part of our lives. But I got early access to this AI model. And the way it worked was that there was this web page and it had a big white box and you could type in that box and then press a button and it would seem to complete the thought for you. It'd complete the text for you. And so I'd been playing around with it for a while. I would just like I put in the beginning of Moby Dick, which is my favorite novel, and hit the button just to see what would happen. I did all kinds of stuff like that. And then I started to think about what the promise was that a technology like this was making. And it seemed to me that the promise was that it could produce words for you when you were at a loss for words. And because I'm a writer, I tend to want to make the effort to come up with my own words to describe my experiences or things I've observed. But there was this thing and continues to be this thing that I have a really hard time finding words for, which is the death of my sister and my grief over it. Experienced death or any other kind of loss will be familiar with that feeling of not knowing how to come up with the words to describe this experience that was so profound. And so I kind of took this technology somewhat at face value. I thought to myself, well, if this is the promise this technology is making, let me try to get it to communicate for me about my sister's death. And so I sat down and I wrote this sentence, which was when I was in my freshman year of high school and my sister was in her junior year, she was diagnosed with Ewing sarcoma. And then I hit the button and it produced this story that really has had very little to do with my actual experience and my actual sister. And the last line of that little, you know, three or four paragraph story it produced was, she's doing great now. Chatbot to do, right? Like it was producing a lie, a falsehood that was sort of like the worst possible falsehood that it could produce, right? Because my sister, my challenge was trying to communicate the reality of what had happened. And this was like the opposite of that reality. And so I thought, okay, well, I know enough about these technologies to know that if I give it more words and hit the button, then it'll have more to go off of and it might match more closely My experience. And so I erased everything it wrote, kept my first sentence. I wrote a little more, hit the button again. And in some ways it did get a little closer. And I did that over and over, sort of deleting what the chatbot wrote every time. And the strange thing that happened was that as I did that, the technology did get closer to describing something that resembled grief or my experience of it in a way that was like weirdly Moving to me and impressive to me. But ultimately, this machine was not me. And so it couldn't say anything authentic about my actual experience. And so I realized eventually toward the end of writing this thing that there was nothing that the technology could come up with that was actually going to fulfill my desire to be able To communicate, because it wasn't me. It wasn't doing the thing that I wanted to do, which was to communicate myself on my own. And so ultimately, I published that experiment as an essay, and I thought it was so interesting how it showed both the ways in which a product like this can legitimately produce language That somebody can find moving and intellectually stimulating and interesting. And yet be doing something very different from what a human is doing when we're trying to communicate. ([Time 0:12:03](https://share.snipd.com/snip/c652148c-feaf-4a85-81e5-1473b9e2ade1)) - **Beware AI Bias in Writing** - Users risk subconsciously adopting AI's biases when integrating its suggestions into writing. - Always critically evaluate AI-assisted edits for potential unwanted influence. Transcript: Vauhini Vara There was a study out of Cornell a couple of years ago that I found really interesting where they had some people write an essay about social media just on their own. And then they gave these two other groups special AI models. For one group, they gave them an AI model that was predisposed to produce positive opinions about social media. And then they gave this other group an AI model that was predisposed to produce negative opinions about social media. What they found was that when people wrote essays with the help, quote unquote, of these AI models, they were twice as likely to produce essays that reflected the quote unquote opinion Of the AI model. It seems from that research and other research that's emerged since then, that even if we are using these AI companies' products to edit our work or ask for feedback on our work, there's A real danger that the responses that we're going to get are going to change our writing in fundamental ways that we might not even be aware of. ([Time 0:21:04](https://share.snipd.com/snip/a6970643-9850-4bea-beb1-12a232088af5)) - **AI as Family Communicator** - Vara's father uses AI to compose messages expressing complex personal dilemmas. - He relies on AI as a communication aid because he feels he's not a strong writer. Transcript: Vauhini Vara My dad has recently started sending me messages on WhatsApp where the whole text of the message is something he asked ChatGBT to write. So for example, he recently sent me one that was, it's hard to decide whether to retire in Canada, the United States or India. Here are some pros and cons for each option. So he never said to me, I'm wondering whether I should move to India or Canada for my retirement. He just sent me that response. And so the subtext, like what he's communicating through ChatGPT is the thing that's actually unsaid. And so there are a lot of people out there. I think my dad is one of many people who want to communicate something. My dad was explaining to me on the phone the other day that he's not a writer. He can't communicate these things himself. But if he gives ChatGPT enough of a prompt, it can communicate the thing he wants it to communicate. And there are things that I find problematic about that for sure. ([Time 0:22:26](https://share.snipd.com/snip/b6552de0-6f7e-41e5-974e-0b60628d8189)) - **Practical and Private AI Use** - Use ChatGPT for small tasks like retrieving forgotten words by explaining the context. - Avoid logging in to preserve privacy and minimize data association to your identity. Transcript: Vauhini Vara Sentence with a blank in it and kind of explain the gist of what I'm looking for. And one thing it's pretty good at is coming up with what that word was that was on the tip of my tongue. So that's a small example of how I use it. When I do use it, I tend not to log into it. ([Time 0:24:40](https://share.snipd.com/snip/908afdec-2964-4cf7-b2aa-c617fc1f7367)) - **User Complicity in Tech Exploitation** - Tech exploitation and usefulness coexist; users must accept their complicity in data collection. - This ambivalence complicates our feelings about digital products. Transcript: Terry Gross You can access that archive and learn about where you were at different periods of your life based on what you searched for. How did you start thinking of searches as a record of your life? Vauhini Vara So the first thing that I ever wrote that ended up in this book was this chapter made up entirely of my Google searches. I wrote it in 2019. I had been covering tech for a long time by that point. And so I knew that Google kept records of our searches unless we turned off its ability to do that. And I could have sworn I'd turn it off, but I hadn't. And so for the past, you know, 15 years off and on, but mostly on, Google had been collecting all my searches. Realizing that, it freaked me out on one level. But then also I found it fascinating because as a writer, as a journalist, I'm always interested in archives, right? And I used to keep a diary when I was a kid, but I haven't in a long time. And it occurred to me that probably the best possible archive of my life was the archive of everything I'd searched for over the years. And it made me think about the way in which, like, it's sort of too simplistic to say, these companies exploit us, and we have no say in the matter, or to do what the companies say in turn, Which is, you're only using these products because they're useful to you. You could stop using them tomorrow if you really wanted to. I think there are these very binary positions. And I think the reality is that the exploitation and the usefulness totally go hand in hand with all of these products. ([Time 0:28:54](https://share.snipd.com/snip/5e608b7e-4415-412e-8e8b-3454059064ab)) - **Searches Archive Selfhood** - Searches archive a person's life reflecting different periods through their queries. - Users complicate their relationship with tech by both exploiting and benefiting from these archives. Transcript: Terry Gross How did you start thinking of searches as a record of your life? Vauhini Vara So the first thing that I ever wrote that ended up in this book was this chapter made up entirely of my Google searches. I wrote it in 2019. I had been covering tech for a long time by that point. And so I knew that Google kept records of our searches unless we turned off its ability to do that. And I could have sworn I'd turn it off, but I hadn't. And so for the past, you know, 15 years off and on, but mostly on, Google had been collecting all my searches. Realizing that, it freaked me out on one level. But then also I found it fascinating because as a writer, as a journalist, I'm always interested in archives, right? And I used to keep a diary when I was a kid, but I haven't in a long time. And it occurred to me that probably the best possible archive of my life was the archive of everything I'd searched for over the years. And it made me think about the way in which, like, it's sort of too simplistic to say, these companies exploit us, and we have no say in the matter, or to do what the companies say in turn, Which is, you're only using these products because they're useful to you. You could stop using them tomorrow if you really wanted to. I think there are these very binary positions. And I think the reality is that the exploitation and the usefulness totally go hand in hand with all of these products. And I think what makes that really uncomfortable for us as users is that then we have to contend with our own complicity, like our own role in the exploitation that's taking place when These companies, you know, collect our personal information and use it to become more wealthy, to become more powerful, to influence political systems. ([Time 0:29:03](https://share.snipd.com/snip/ed8dbe88-88b8-4aea-971a-a7b2daef10f2)) - **Choice Still Exists on AI Use** - Despite AI hype, most people in the US have not tried ChatGPT even in 2025. - We still have agency to decide AI's role in society and culture moving forward. Transcript: Vauhini Vara So in the US, for example, most people have never tried chat GPT still in 2025. And so we're in this interesting position where we can actually decide as individuals, as communities, as societies, the extent to which we want AI to be a part of the future, the extent To which we want AI generating novels or generating something that is going to substitute a newspaper or magazine article or a radio show. ([Time 0:32:44](https://share.snipd.com/snip/d6126e19-e92e-4db4-8bc6-6be6911b3771)) - **Love of Spelling and Words** - Vara describes her love of spelling despite its declining practical value in the digital age. - She cherishes the joy of learning obscure words, sharing a story about her national spelling bee experience. Transcript: Terry Gross I always wonder, what is the point of asking young people to spell obscure words that no one uses and no one even can define? Can you explain that? Because it makes no sense to me. Vauhini Vara Yes, I have so many thoughts on this, Terry. I continue to love spelling. I love language. I think, you know, to get philosophical about it, I think we could ask that question about anything, honestly. I think we could say, what's the point of trying to run a three-minute mile, right? Or like, I've taken up rock climbing. What's the point of like trying to climb to the top of a rock wall when elevators exist? Right. And I think it speaks to this thing about AI, too. What's the point of trying to write an essay if AI can write it for you? Are like idiosyncratic curious creatures and we created this thing called language that's really important to us and for me personally who knows why I don't know why I love words I love Language I love knowing how sounds fit together and produce meaning like that feels that's always been fascinating and meaningful to me and so like I love it because I love it. And I kind of love about spelling that there is, you're right, there is no point, especially in the age of spellcheck and AI. Like, there's probably even less of a point than there was in the mid-90s. But strong supporter of spelling. I think everybody should be in spelling bees. Terry Gross Yes, but the question, though, was about asking young people to spell words that nobody ever uses, that no one can even define. Really obscure words. Do you take pleasure in spelling words that you didn't even know existed? Vauhini Vara Pleasure in like knowing that they exist when previously I didn't know that they exist. I know I love it. And I think they do, too. I think there's this misconception about spelling bee kids that like they're all doing it because their parents are making them so that they can get into college 10 years later or something. But every I wrote an article about spelling bee kids in 2018, I think it was for the magazine Harper's. And so I got to spend time with like a more recent generation of spelling kids. And the thing that I think they had in common with the spelling kids I knew in the mid 90s is just like this genuine, idiosyncratic, strange love for words and how they're put together. Terry Gross So you placed third in the National Spelling Bee. What was your losing word? Vauhini Vara Oh, Terry, the losing word was paraplus. Paraplus? Can you spell it? Terry Gross Yeah, but again, I've never heard the word before. I have no idea what it means. I would spell it paraplus or paraplus because one sounds like an animal and the other sounds like an amount. Vauhini Vara Okay, so the word is paraplus. There is an alternate pronunciation, which is paraplus. ([Time 0:33:27](https://share.snipd.com/snip/bfedbef8-3822-4fcd-bde6-1d0327840a18))