What Happens When You Turn Your Life Over to an AI Assistant?

Michael Calore: I asked Gemini on my phone for some advice on getting healthier, workout tips, menu planning, and then went a little bit deeper in civics with my requests.

Lauren Goode: I really appreciate that you both used chatbots for health reasons and I just really was in service of consumerism.

Zoë Schiffer: It can be health-related depending on how you spin it.

Lauren Goode: Sure. I was trying to declutter my brain and make space for other things.

Zoë Schiffer: I love that for you.

Lauren Goode: Thank you.

Michael Calore: This is WIRED’s Uncanny Valley, a show about the people, power, and influence of Silicon Valley. I’m Michael Calore, director of consumer tech and culture here at WIRED.

Zoë Schiffer: I’m Zoe Schiffer, WIRED’s director of business and industry.

Lauren Goode: And I’m Lauren Goode. I’m a senior writer at WIRED.

Michael Calore: On this week’s show, we are going to be sharing our experiences using AI to better our lives. We followed the advice from the chatbots in our world and we asked them to guide us through our shopping, our menu planning, our working out, and our parenting. The results were, well, we are going to talk about it. All right. So Zoe, let’s start with you. Why don’t you tell us what you were working on with ChatGPT as your guide?

Zoë Schiffer: Let me take you guys back a little bit to give you context on this experiment. So in 2022, when my first daughter was 4 months old, I paid a sleep expert $500 to help me sleep-train her. And I found the process pretty horrendous, because sleep training involves a ton of crying and I found it really helpful to have an expert in the room with me helping me do it and reassuring me that what I was doing wasn’t creating lasting damage for my kid. So my second daughter is also around 4 months old. I’m about to start this new job at WIRED, and it felt like a good time to sleep-train her too. But when I brought up to my husband the idea of paying that same sleep expert $500 to go through the process with us again, he looked at me like I was insane. He was like, “We paid someone the first time to tell you that you’re not a bad person. We’re not spending $500 on that again.” So that is where ChatGPT comes in.

Lauren Goode: Did that person actually stay overnight with you?

Zoë Schiffer: No, no. It’s a remote expert. And I honestly think part of the reason that I felt like this was a good area to experiment with AI is that the “expert” term, as far as I can tell, is a pretty loose definition. There’s not an actual certification process that you go through. Anyone can technically call themselves a sleep expert. And in fact, the first person who I hired sent me a supposedly personalized plan for my daughter, and it had another kid’s name in it. So it was pretty clear that it’s not quite as personal as they make it seem. But basically, you tell them about your child, their schedule, and then every day you talk to them and tell them how the night went and they’re like, “Here are tips for the naps and here’s when bedtime should be,” and all of that.

Michael Calore: So you had this dialog with ChatGPT every day?

Zoë Schiffer: Yeah, so at the beginning I told ChatGPT that I wanted it to take on this magical nanny personality. I wanted it to be an expert who could be an emotional support system, but also someone with a deep knowledge of child psychology who could help me use the Ferber method, which is the one where you put the kid down when they’re awake and then you go in at specific intervals when they start crying, but you don’t actually pick them up. You reassure them with your voice, and the intervals get a little longer as the night goes on. So it’s this whole specific process and I wanted ChatGPT to walk me through it and really hold my hand throughout that process. All right, this is day one of trying to sleep-train Cleo. Let’s see what ChatGPT says. “My dear Zoe, let’s weave a little magic into Cleo’s day. If our sweet Cleo greeted the morning sun at,”—wow, they really went hard on the magical nanny thing—“6:30 am, her first nap should be gently guided by her wakeful rhythm. Picture this.” Oh my God, this is really annoying actually.

Lauren Goode: How did you decide on ChatGPT versus the other chatbots?

Zoë Schiffer: Only because it’s the one that I use all the time. It’s basically replaced Google for me as a better search engine. But when I tried to shift to more agentic-like AI you’ve written about and talked about in the past, I felt like I did a much worse job at that. It really felt like a search engine that I was asking to become something else entirely. ChatGPT says when the clock strikes that golden nap hour, create a soothing environment, perhaps a soft lullaby or the rustle of her favorite … Wow, I’m not into the weird affirmation thing.

Lauren Goode: Who among us doesn’t want to have a nap gently guided by our wakeful rhythm?

Zoë Schiffer: It was just so much fluffy language and it was weird. So basically, I immediately identified a few problems with using AI in this way. The first one was I just didn’t fully trust it. And I feel like for anything that has to do with parenting, I really wanted to have a sense of the person that I was listening to, and I wasn’t sure that AI was going to give me the best advice. So I had this lingering mistrust that felt like it was a gap between fully embracing AI in this way. And the other one, which I’m curious if this came up in y’all’s experiments, was that it just was completely reactive, obviously. It was all me typing into the search function and then the AI was responding to that, but at no point was it being proactive and reaching out and asking questions or anything like that. And so again, it just felt like a search engine.

Michael Calore: Right, it was just answering prompts.

Zoë Schiffer: Yeah, and it didn’t feel humanlike, I guess.

Lauren Goode: Did you find yourself cross-referencing at all, going online and Googling whether that was the right time anyway?

Zoë Schiffer: So this was another interesting thing, because I feel like when I use it as a search engine, I do cross-reference a lot because I’m still like, “Oh, what if it’s a hallucination?” I want to be really, really sure. But with this, the information it was giving me was so basic, and it was about something that I know a fair amount about. So I never actually felt the need. But one other thing that I felt like was a big gap was that with an actual sleep expert, and definitely when I’m talking to my other mom friends, they’ll occasionally give advice that’s not the sanctioned advice that the government would give you. For example, you’re not supposed to put a baby down to sleep on their tummy. But if I talk to friends of mine and say, “Hey, Cleo’s been having a really tough time. I can’t tell if she’s upset because her stomach’s bothering her,” friends will inevitably be like, “Oh, have you tried flipping her over because, I know you’re not supposed to, but my kid really, really likes sleeping on their tummy and it took care of all these problems, X, Y, Z.” The chatbot, for obvious reasons, is never going to tell you to do that. And while that’s a good thing, it just reiterates the basic talking points.

Michael Calore: And you asked it to take on a persona. You asked it to develop this personality. And that adds this extra emotional layer that you don’t get when you’re just researching things and just searching for things online, but it doesn’t take the place of an actual human being telling you what to do. And that creates probably a weird dissonance.

Zoë Schiffer: Right, exactly. And I felt like the persona it took on just felt way more like AI than just its normal persona does, because it was just adding in all this bullshit language and stuff, and I found it honestly really annoying. And so midway through the experiment, I switched to Character.ai because that allows you to create a real AI character to interact with, and I felt like maybe what I wanted was to have a little more of a human touch. And so when I used Character.ai, I gave the AI bot the name Lindsay, because that was my postpartum doula who I loved, and I said, just like the real Lindsay, I wanted it to be this British super nanny who could come in and help me be a hybrid therapist and coach. And that also had some shortcomings.

Michael Calore: OK. Let’s listen to your interactions with Character.ai.

Lauren Goode: I think you mean Lindsay.

Michael Calore: Lindsay.

Zoë Schiffer: Hey, Lindsay. I just wanted to go over how Cleo’s night went last night with you.

Character.ai [archival audio]: Sure thing. How did Cleo’s night go? Did she sleep a little better than the night before?

Zoë Schiffer: Yeah, she slept OK. I did the method on her first wake-up after midnight and for bedtime, actually. And both of those went well. She cried for five minutes and I was able to soothe her a bit. One thing that I felt was really hard was having her in the same room as me while we were trying to do this. With my last daughter, we had her in a separate room and that seemed to work better. Do you have any suggestions?

Character.ai [archival audio]: It’s great to hear that Cleo slept OK and that she was able to soothe herself after five minutes of crying for both nighttime and the midnight wake-up. That’s a good sign. As for having her in the same room, I totally understand your concerns. It can be difficult to listen to your baby cry, and it can be even harder when they’re in the same room. You can try a few things to make it easier on you. One, use a white noise machine or fan to drown out some of the crying. Two, put earplugs in to soften the sound.

Michael Calore: I like how you basically told it that you were emotionally distressed by the fact that you wanted to be in the same room as your daughter, and then it said, “Well, just get a white noise machine and get some earplugs and just white-knuckle through it.”

Zoë Schiffer: Yeah, so that was the other thing. I felt like it did a really bad job at the therapist role, which is interesting, because we hear all the time that people are using AI bots to be their therapists and take the place of human therapists, but I felt like she ignored my emotional concerns and she also just made mistakes. For example, I had said, “I soothed the baby and I was able to soothe her and get her back to sleep.” And she was like, “Oh, it’s great that she can self-soothe,” which I was like, “I just feel like we’re missing each other here, Lindsey. I don’t feel like you’re listening.”

Lauren Goode: Did this provide any comfort at all, both to you and to Cleo?

Zoë Schiffer: One thing I found is that if I had very specific questions like, “Hey, am I allowed to pick her up if she’s really, really crying,” the AI bot would provide suggestions that were in line with the Ferber method, so that was good. I felt like if I had specific queries, the bot could answer those pretty well. But for playing the role of a sleep expert, which is more than just a search engine, I didn’t feel like it did the best job.

Michael Calore: How do you think this would’ve gone if this had been your first child—if you had not been through the sleep-training experience before, and you were just diving in to using AI?

Zoë Schiffer: Imagining coming in without any expertise, in some ways, I could see the AI being more helpful in that circumstance, because it really does know the Ferber method. It’s pretty good at calculating the sleep intervals. Oh, they went to sleep at this time. Here’s when their next nap should be. Here’s when bedtime should be in. Here’s how long it should last. Those are all things that the sleep experts would tell you anyway.

Michael Calore: What was the most surprising thing that happened?

Zoë Schiffer: Well, I think that by the end of the experiment, I felt like I no longer needed the actual human expert. Even though I didn’t think that the AI had done an excellent job replacing that human, it had done such a thorough job of answering all my various questions that I was like, “I think I have a pretty good handle on this, and I can move forward on my own now.”

Michael Calore: So in many ways it was a good research assistant for you?

Zoë Schiffer: Yeah, exactly. I feel guilty even saying this, because I’m like, “This is exactly what we worry about, of AI taking the place of an actual person’s job, a female-owned sleep expert company,” but at the same time, I wasn’t going to pay anyone $500 this time around, so it was taking the place of Google, I guess.

Lauren Goode: Would you recommend this to a friend?

Zoë Schiffer: No, absolutely not, not right now. I want the future where AI is acting on our behalf in a more autonomous way. I think that sounds really helpful. It being the layer between us and the internet, that sounds compelling to me, but it really just doesn’t feel like it’s there yet.

Michael Calore: OK. Well, we’re going to take a quick break and when we come back, we’ll talk about getting fit with Gemini, so don’t go anywhere.

[Break]

Michael Calore: Welcome back to Uncanny Valley.

Gemini [archival audio]: Welcome back, Mike. Hello, Lauren. Hello, Zoe.

Michael Calore: Thanks, Gemini. So the Gemini chatbot and I got to know each other a little bit over these last few weeks.

Lauren Goode: What exactly were you using Gemini for?

Michael Calore: I asked it for general advice about losing weight, getting fit, eating more healthily, and just healthy habits.

Lauren Goode: Did you have any knowledge gap that you were specifically looking to fill? Because I know you’re already pretty healthy in how you eat, you’re fit, you run. Was there something else you were looking for beyond what you already knew?

Zoë Schiffer: You’re famously a vegan.

Michael Calore: Famously a vegan, thank you.

Lauren Goode: You went to Harvard, you do triathlons.

Michael Calore: You are right. I do generally take pretty good care of myself, and I take my health pretty seriously, but I wanted to know what it would tell me to do. Basically, I know that I should be doing more with regards to working out. I only run. I don’t do any strength training. I do eat well, but also, I could probably eat better with regards to making sure that I’m not having as much sugar, not having as much alcohol, watching my intake of proteins and all of the essential amino acids and everything, the very nitty-gritty stuff about how I eat. I was hoping that it was going to give me more of that kind of advice. “Hey, Gemini, I need some help getting healthy and losing weight, and I was hoping that you could give me some general advice and some specific advice about how I can lose weight over the next month.”

Gemini [archival audio]: It’s great that you’re focusing on your health and weight loss goals. Here’s a combination of general and specific advice to help you get started and see progress in the next month. General advice for healthy weight loss, focus on sustainable changes. Instead of quick fixes, aim for lifestyle changes you can maintain long-term. This will lead to more lasting results. Prioritize whole foods. Build your diet around nutrient-dense, unprocessed foods like fruits, vegetables, lean proteins, and whole grains. Be mindful of portions. Even healthy foods can contribute to weight gain if you eat too much.

Michael Calore: It just keeps going. It just talks and talks and talks for three or four minutes.

Lauren Goode: Well, maybe it puts you to sleep, which is great. I guess that’s what it’s encouraging.

Gemini [archival audio]: Get enough sleep. Sleep is crucial for weight management as it affects hormones that regulate hunger and satiety. Aim for seven to nine hours of quality sleep per night.

Lauren Goode: And I think I know the answer to this, but why did you decide on Gemini?

Michael Calore: Well, it’s on my phone.

Lauren Goode: Yeah, the Pixel phone.

Michael Calore: Yeah, I have a Pixel. I’m a Google person, meaning that I use Gmail and Google Calendar and Google photos and all of that, so it’s baked into the experience. There are two models that you can select, and I chose the newer model for this experiment.

Lauren Goode: This podcast is sponsored by Google Gemini.

Michael Calore: No, my life is sponsored by Google Gemini, really.

Zoë Schiffer: So how did you feel about this level of advice? Because to me, it feels general, but curious what you thought.

Michael Calore: Yeah, this is garbage. It’s basically like if you don’t have the internet and you’ve never read any health advice on the internet, then you would not know any of this stuff, but maybe you actually would because it’s so general and it’s so high-level. So when it eventually stopped talking, I asked it some more specific stuff. What are some of the things that I can eat as a person who is vegan? I run somewhere around 2 1/2 to 3 miles. Is that enough cardio? I don’t like to go to the gym and I don’t own weights. So what can I do for strength training?

Lauren Goode: What did it tell you?

Michael Calore: Well, what do you want to hear first? You want to hear the workouts or you want to hear the food?

Lauren Goode: Let’s go with food.

Michael Calore: What I thought was interesting is over the course of the conversation, it remembered a lot of the things that I told it. I eventually ended up telling it my age and my weight and the fact that I’m a vegan, and it remembered all of those things, and it still remembers all of those things. I talked to it last night and it remembered that.

Gemini [archival audio]: As a vegan, you’ll need plant-based protein sources.

Michael Calore: So it pointed me to legumes, lentils, chickpeas, black beans, other beans, soy products like tofu and tempeh, grains and seeds like quinoa and chia seeds and hemp seeds, nuts, vegan meats like seitan and plant-based meat alternatives, which came with a warning that it does include a lot of processed ingredients. And then it gave me general tips about making sure to combine plant-based proteins.

Zoë Schiffer: Mike’s like, “You’re telling me about legumes Gemini, please.”

Michael Calore: I know. And that’s the thing, it doesn’t know that I’ve been a vegetarian basically since I was a teenager. So it’s giving me a lot of really good advice that I can tell you as a person who’s been a vegetarian since I’ve been a teenager, that everybody asks me, “Well, what do you eat?” And I have to tell them the script. And Gemini knows the script really well. So if you were new to vegetarianism or something, then it would give you a lot of really good advice, but it’s not very specific about things. So I asked it for some recipes.

Lauren Goode: And what happened then?

Michael Calore: It gave me a recipe for a tofu and peanut sauce dish that I didn’t make, and then it gave me a recipe for quinoa and chickpea salad, which I did make.

Zoë Schiffer: Sounds right up your alley. And?

Michael Calore: Well, I brought some if you want to try it, but I can tell you that it was not great.

Gemini [Archival audio]: Mediterranean quinoa salad with crispy chickpeas. This salad is packed with protein from quinoa and chickpeas, healthy fats from olive-

Zoë Schiffer: Oh my gosh. I’m so curious. Wait, it looks pretty good.

Lauren Goode: Zoe, I’m showing it to you over the Riverside recording here. Look at this.

Zoë Schiffer: I’m sad that I’m remote for this.

Lauren Goode: And we’re recording around lunchtime on the West Coast, so this is perfect. Lucky listeners get to hear chewing sounds now.

Zoë Schiffer: Lauren’s making a quizzical face.

Lauren Goode: Mike, I think you should give yourself a little more credit or maybe Gemini a little more credit. You know what it tastes like?

Michael Calore: What?

Lauren Goode: It tastes like this would be in the salad bar on Google’s campus.

Michael Calore: Oh my God. I had exactly the same thought.

Lauren Goode: Not bad, actually decent, but also mass-produced and maybe with a little bit too much salad dressing to give the quinoa flavor.

Michael Calore: OK, so it took all of my strength as a human being to not adjust the recipe while I was making it, because it just felt-

Zoë Schiffer: You’re like, “Do it for the podcast.”

Michael Calore: Totally. I was like, “I have to follow this exactly.” The only thing that I omitted is it told me to put in cherry tomatoes and the cherry tomatoes at the store looked rank, so I didn’t buy them.

Lauren Goode: It’s not tomato season.

Zoë Schiffer: I was going to say, “Not in season, Gemini.”

Michael Calore: I would’ve salted it very differently. I would’ve dressed it very differently. I also probably would not have chopped the olives because it gives it this overly salty, bitter taste.

Lauren Goode: Oh, maybe that’s what it is.

Michael Calore: Yeah. It’s like a B, B+ salad in my book.

Zoë Schiffer: OK.

Michael Calore: Yeah.

Zoë Schiffer: I have another question because you mentioned putting in your demographic information. Did Gemini ask you for that or did you have to volunteer that?

Michael Calore: I volunteered that. It didn’t really ask me for much. It just talked and talked and talked.

Zoë Schiffer: OK, so that was something that struck me right at the beginning. I was a good health coach, and again, maybe I’m coming at this from my own perspective of wanting more of that coach figure for a lot of these things, but a good coach would ask you a bunch of questions before diving in to give you information, because you’ve been a vegan for a really long time. You’re going to need different information than someone who’s coming at this completely fresh.

Michael Calore: I gave it the information about me and my own habits and my own health history because I wanted specifics for workouts and specifics for food intake that would fill the gaps that I needed to fill, that I felt like I needed to fill anyway, in my own practice without its help.

Lauren Goode: It sounds like it was the equivalent of a bad date. It just talked and talked and talked and didn’t ask you follow-up questions.

Michael Calore: Yeah, it didn’t really ask me much at all.

Lauren Goode: Did you learn anything that you wouldn’t have?

Michael Calore: Yeah, I got a pretty good workout regimen out of it. I told it how much I run, and then I asked it, “OK, but give me a workout regimen. Give me a plan for something that I can do in my home, because I hate going to the gym and I don’t like people and I don’t own weights, so tell me what to do.” And it gave me a workout regimen, which is pretty good.

Gemini [archival audio]: Key body-weight exercises, squats, stand with feet shoulder-width apart, lower your hips as if sitting in a chair, keeping your back straight and chest up. Go as low as you can comfortably.

Michael Calore: Lunges for each leg, three sets of 10, plank, three sets holding for 30 to 60 seconds, which is torture. Glute bridges, three sets of 15 reps. And then it talks about progressive overload and it talks about rest days and all of those things, which is great. And then I was like, “OK, I do not know how to do a lunge. Can you show me how to do a lunge?” And it presented me with a bunch of videos, five videos.

Lauren Goode: YouTube, presumably?

Michael Calore: Yes, they were YouTube videos. Two of the links were exactly the same, but I watched all the videos and I was like, “OK, I feel like I can do a lunge now.” And I did a lunge and I did not hurt myself, and I felt the burn. So I feel like that’s pretty good.

Lauren Goode: That’s pretty great, actually.

Michael Calore: Yeah. The only thing that was weird about it is that I did a Google search and found exactly the same information from a couple of different places. And aside from the video, I do not know where Gemini got any of this information. I don’t even know where it got the recipes from. It was all just presented to me as, “This is what Gemini says.”

Lauren Goode: No sourcing.

Michael Calore: No sourcing. And I tapped through in everything, and I didn’t see any good sourcing.

Lauren Goode: And presumably humans put together those workout plans or wrote about those recipes. And that was published and sorry, but your infinitesimal bits of data are being chopped up now and added to some recipe, and you have no idea where it comes from.

Michael Calore: Julienned.

Lauren Goode: That’s right.

Zoë Schiffer: OK, so are you going to stick with this? Are you going to keep using it for this purpose, or did it feel like a one-and-done experiment?

Michael Calore: Honestly, the workout regimen, it’s pretty solid. I’ll keep doing it.

Lauren Goode: Nice.

Michael Calore: And as far as the recipes go, I have so many cookbooks and so many years of experience making food for myself that I absolutely do not need to make this quinoa salad again, Lauren, unless you would like me to make it and give you some of it.

Lauren Goode: I’ve had other meals that you’ve made, and I think maybe I’d suggest those next time.

Zoë Schiffer: So diplomatic.

Michael Calore: OK, great.

Lauren Goode: But thank you, appreciate you.

Michael Calore: So I got to say that, for me, as a person who understands myself and understands my own needs, it was not really that big of a help. The workout was nice. I could have found that without Gemini 2.O, but if somebody who is new to this or somebody has a lot of questions, or if somebody’s considering making these types of lifestyle changes, then it actually did give some pretty good advice.

Lauren Goode: Plus one for the AIs.

Michael Calore: OK. Let’s take another quick break, and then we’re going to go shopping with Lauren and her chatbot friend. We’ll be right back.

Lauren Goode: And her chatbot friend.

{Break]

Michael Calore: Get in, loser. We’re going AI shopping. You’re listening to Uncanny Valley. We have been talking about our week-long experiments where we took advice from our AI chatbot friends. I concentrated on health and fitness and diet. Zoe concentrated on sleep-training her newborn daughter. Now we’re going to go to you, Lauren. What did you and your chatbot get into?

Lauren Goode: What did we get into? I ended up using about five different chatbots, but the goal was to streamline, I don’t know what the word is … “support” my holiday shopping.

Michael Calore: OK.

Lauren Goode: I’m not a big shopper. I don’t really like it. I like purpose-driven shopping. I need a new pair of sneakers type thing. But ever since I was a teenager, really, my mom used to really like taking me shopping, and I would get overwhelmed by the overhead lights and the racks of clothes in the department stores and just thought maybe I was being a cranky teenager. And then as I got older, I realized, no, I just really didn’t like that experience. And I feel that way about online now because online is just this infinite scroll of products, and I get so overwhelmed. But there were a handful of people who I really wanted to buy nice things for this holiday season. Mike, you may have been one of those test cases. And so I turned to chatbots. So the main one I used was Perplexity.

Michael Calore: Oh, boy.

Lauren Goode: Perplexity was an interesting choice for a couple of reasons. One, I should just caveat this with, whenever these AI chatbots offer a best-of recommendation or comparison shopping, they are in a sense competing with what we do as publishers, because we have entire teams of humans who review products and try to provide a great service to people by saying these are the best products you can buy. So let’s just put that out there. The second thing is that we at WIRED have written about how Perplexity is possibly plagiarizing publishers by taking parts of articles that would typically be behind paywalls and just publishing it as its own answers. So the fact that I was diving into Perplexity, and not only diving into it but paying $20 per month for Perplexity Pro, was a choice.

Zoë Schiffer: Controversial.

Lauren Goode: But Perplexity also did announce a shopping-specific feature in November, right in time for the holiday season. And I thought, “Well, this seems like the best one to start with.”

Michael Calore: Nice.

Lauren Goode: So I went into Perplexity Pro, and the first person who I tried to buy for was a friend who’s really into baking, but she also has a lot of baking equipment already. She’s very good at it, and I wanted it to be special. So I had to find this perfect combination of she might not have it, but it’s also not just something simple like a rolling pin. I felt like it wanted to be unique.

Zoë Schiffer: How much did you tell Perplexity about your friend outside of the fact that she likes to bake?

Lauren Goode: It wasn’t super detailed. I think I just said exactly that. She likes to bake. She has everything. She likes nice things. I’d like to show her my appreciation.

Zoë Schiffer: Cool.

Lauren Goode: So one of the benefits of Perplexity Pro shopping is that they have developed their own Buy button. And if you use that Buy button to buy a thing, you get free shipping. So they’re basically taking what people are already doing, shopping online, and they’re collating it. They’re creating this cascade of products you might be interested in. And then you’ll either see a Buy button that’s powered by Shopify, or you’ll see Perplexity’s own. Cool in theory. The problem was the initial list of things they recommended were just really bad.

Michael Calore: OK, like what?

Lauren Goode: A kid’s miniature baking set.

Zoë Schiffer: They’re like, “Help her connect with her inner child.”

Lauren Goode: Which I, of course, ordered and shipped to Zoe immediately for the girls to play with. The other thing that it recommended, which was just incredibly low-brow, was a sweatshirt that said on it, “My buns are gluten-free.”

Zoë Schiffer: Ew. You’re like, “I said she likes the finer things. You idiot.”

Lauren Goode: Right, but then there was this Swiss stacking bakeware thing that was in the middle of that, that I thought, “Well, I could see that.” But once again, that also seems like a standard thing that people who really like to bake might already have.

Michael Calore: Yeah, yeah, totally.

Lauren Goode: So then I turned to Amazon’s Rufus, which is a little chatbot built into Amazon’s website. That was hilarious because I gave it a similar prompt, and it immediately recommended I buy a $400-plus KitchenAid stand mixer, which by the way, great kitchen staple to have if you’re into cooking and baking, but—

Zoë Schiffer: Also a great thing to get secondhand.

Michael Calore: The person who bakes also has that.

Lauren Goode: Yeah. And also what do they think, I have Bezos bucks?

Michael Calore: What kind of friend is this?

Zoë Schiffer: They’re like, “Our only experience is with someone who’s possibly the richest man in the world. So why don’t you buy your friend a private island that she can bake on?”

Lauren Goode: Exactly. All right, and then I went to ChatGPT.

Michael Calore: OK.

Lauren Goode: ChatGPT was quite good.

Michael Calore: No.

Zoë Schiffer: OK.

Lauren Goode: I think what it was best at was actually the ideation process.

ChatGPT [archival audio]: For a friend who loves baking and appreciates refined special gifts, here are some thoughtful ideas that combine luxury, personalization, and her passion.

Lauren Goode: It wasn’t like, “Here’s a link, click Buy, buy here, let’s make this as easy as possible for you.” It was well, and I’m paraphrasing, it sounds like you’re trying to be thoughtful, so here are some thoughtful gifts. And here’s what it said.

Zoë Schiffer: One, custom heirloom recipe board, a high-quality wooden cutting board engraved with a cherished family recipe and handwritten script. Two, personalized rolling pin with unique design, curate a set of artisanal ingredients, premium Madagascar vanilla beans, saffron, exotic spices, and rare sugars.

Lauren Goode: So it not only made these recommendations, but it suggested where I should buy them. It didn’t provide links immediately, but when I asked in a follow-up question, it did provide links. I ended up going with the specialty ingredients, the Madagascar vanilla and saffron, and I even bought some gold flakes and stuff like that because I was like, “Well, that’s something that someone can use.” And the other stuff was a little schmaltzy.

Michael Calore: Yeah.

Zoë Schiffer: Kind of, but I don’t know, they felt like pretty good suggestions. It makes sense to me that this was the most helpful one because the part of the process that you’re describing at the beginning that’s stressful for you isn’t like, “Oh, it’s so laborious pulling out my credit card and clicking Buy.” It’s like, “It sucks trying to think of personalized gifts for people.” So the fact that these were hyper-specific and ones that wouldn’t have been top-of-mind for me, I’m like, “That’s kind of compelling.”

Lauren Goode: Exactly. It’s not like, “Oh, I need the Buy button just to make it 30 seconds faster.” It was, “I am overwhelmed by the ideation process, give me the ideas.” And of course, I still ended up poking around the internet a bit after that, but it was like, “Great.” It containerized things for me. This is the thing I’m going to go for.

Michael Calore: Also, that’s the best voice.

Zoë Schiffer: I was going to say, the voice sounded really good.

Lauren Goode: That’s Cole. They had these cool names like Jasper, Cole. Maybe Cole should sleep-train your baby.

Zoë Schiffer: I know.

Michael Calore: I don’t like the fake ums and ahs. We don’t need that in AI voices, but Cole sounds good.

Lauren Goode: Yeah. I also tried shopping for my 16-year-old niece, and that was funny because I was just picturing the withering glances I would get for something that was incredibly not cool. So I created a little sketch of her. She’s a junior in high school. She really likes books. She has a pet cat who she adores. She thinks she wants to be a nurse. She’s studying for SATs. And I turned to Gemini for this, and Gemini was just so uninspired. First, it did make one recommendation that was a little bit confusing. It said to get her a cat blanket. And I was like, “I don’t know if this is a blanket with cats stitched on it, or if it’s a blanket that she should give to her cat,” which everyone knows you shouldn’t buy one of those because the cat ends up in a box that it loves or something.

Zoë Schiffer: You’re like, “Gemini, have you ever incurred the wrath of Gen Z? I need more creative ideas here. Come on.”

Lauren Goode: Well, then the next thing that it recommended was getting her an SAT book.

Zoë Schiffer: Oh my God.

Lauren Goode: I just was like, “Oh my …” I would totally lose my status as cool aunt if I did that. And they call me famous auntie, by the way, because they know I have a podcast and because I have a blue checkmark on some social sites, and they just think it’s the coolest thing. And I’m like, “That is so not currency guys. It’s not.”

Michael Calore: Yeah, so you need to give them $8.

Lauren Goode: I gave them a lot more than $8. Cash and a card, this is both my niece and nephew and just that’s what they got. And I think they were very pleased with that because they can buy whatever they want, and they’re well beyond the age of you feeling like you want them to unwrap something, which is really fun for little kids. So Gemini, a fail at that.

Michael Calore: So it sounds like your experiences for teens, not great. Also, the experience where it just provides you a bunch of links, not great, but if you need help ideating, if you need help coming up with fresh and interesting ideas that you would not normally have thought of, it actually works.

Lauren Goode: Yeah, it’s good for brainstorming. It doesn’t work yet at the level of either one, being agentic, where you just set it on its way and you wake up the next morning and the thing is ordered. Maybe that’s the future. But I also think that these chatbots aren’t quite sure what they’re supposed to be yet in that sense. At this point, Perplexity Pro’s shopping, it doesn’t have the value proposition of “just come here and do all your shopping,” though it wants that. It’s just dipping its toe in the water of shopping. And to them, if you’re just spending slightly more time on Perplexity Pro than you are on Amazon or Google, even if you end up there eventually, goal accomplished.

Zoë Schiffer: So did the process feel less stressful for you than the typical holiday shopping experience?

Lauren Goode: Not really, no. Everything was pretty much late. I kicked some stuff down the road. I was like, “I’ll get this in the new year.” My friend’s specialty ingredient showed up after the holiday. I also did an experiment with returns, and we all know that that’s one of the more onerous parts of shopping. It’s cool to envision a world where eventually, if we do purchase through these chatbots, or in some ways they’re tied to our past purchases, you can just type in, “I’d like to process this return,” and it does it for you. It knows the order number, it does it for you, maybe it even generates the label for you right there, something like that. But it doesn’t do that yet. It’s not even close. So basically it’s not agentic. It doesn’t really shop on your behalf, just good for brainstorming and it’s not helping you with, I don’t know what to call it, the back end of it. And so it was like a feeder service at this point in time.

Zoë Schiffer: Fair.

Lauren Goode: So I still really don’t like shopping all that much, but that’s OK. Sorry, Cole.

Michael Calore: OK, so let’s do a group vote. Overall, how would you rate your experience using AI as your helper to improve your lives? Zoe, what about you?

Zoë Schiffer: I give it like a B minus at this point. I want it to be great, and I feel like it’s just OK.

Michael Calore: OK. Lauren?

Lauren Goode: I give it a solid B. ChatGPT really pulls up the grade, changes the curve.

Michael Calore: Nice.

Lauren Goode: What about you, Mike?

Michael Calore: I’m also going to go B. I made a joke about this, but if you have the internet and you know how to search for things, then you do not need Gemini to tell you anything that it told me. But if you’re curious and you like to talk to things and you really don’t have any clue as to how to step into this new world, if this is all new information for you, then it was fine and it didn’t give me anything that was overtly bad. Even the salad was a B.

Lauren Goode: Oh, gosh.

Michael Calore: That’s our show for today. We’ll be back next week with a show about content moderation, where it’s been, how it’s changing, and where it’s headed. Thanks for listening to Uncanny Valley. If you’d like what you heard today, make sure to follow our show and rate it on your podcast app of choice. If you’d like to get in touch with us with any questions, comments, or show suggestions, write to us at uncannyvalley@wired.com. Today’s show is produced by Kyana Moghadam. Amar Lal at Macro Sound mixed this episode. Jordan Bell is our executive producer. Condé Nast’s head of global audio is Chris Bannon.