Would you entrust—or have you entrusted—your travels to ChatGPT? In this week’s episode of Unpacked, we explore the technology, the ways it can be helpful now, the ways it can steer you wrong, and what it might be capable of doing for us in the future, specifically with regards to travel and travel planning.
Read the transcript here: https://rebrand.ly/3a93h3d
Discover more episodes of the podcast here: https://www.afar.com/podcasts/unpacked
----
CONNECT WITH AFAR
Afar.com is a digital and print magazine that publishes travel tips, guides, news, and stories: https://www.afar.com
Get updates on the latest articles, travel news, and more from AFAR by signing up for the AFAR newsletter: https://afar.com/newsletters
Follow AFAR on Facebook: https://www.facebook.com/AfarMedia
Follow AFAR on Twitter: https://twitter.com/afarmedia
Follow AFAR on Instagram: https://www.instagram.com/afarmedia
Follow AFAR on Pinterest: https://www.pinterest.com/afarmedia
Read the transcript here: https://rebrand.ly/3a93h3d
Discover more episodes of the podcast here: https://www.afar.com/podcasts/unpacked
----
CONNECT WITH AFAR
Afar.com is a digital and print magazine that publishes travel tips, guides, news, and stories: https://www.afar.com
Get updates on the latest articles, travel news, and more from AFAR by signing up for the AFAR newsletter: https://afar.com/newsletters
Follow AFAR on Facebook: https://www.facebook.com/AfarMedia
Follow AFAR on Twitter: https://twitter.com/afarmedia
Follow AFAR on Instagram: https://www.instagram.com/afarmedia
Follow AFAR on Pinterest: https://www.pinterest.com/afarmedia
Category
🏖
TravelTranscript
00:00 Buckle up, fellow explorers, as we embark on an expedition like no other. It's time
00:07 to unleash the potential of CHAT-GPT and unleash the wanderer within us all. Throughout this
00:14 captivating episode, we'll navigate the intricate terrain of travel planning.
00:18 Let's start again. I'll give you two tries to guess who, or I should say what, wrote
00:26 that truly terrible intro. Yep, CHAT-GPT, aka the subject of this week's episode.
00:37 I'm Aisling Green, and this is Unpacked, the podcast that tackles one tricky topic
00:40 in travel each week. And in this episode, we're unpacking the promise of CHAT-GPT in
00:45 travel. Because cheesy writing aside, the technology, of course, has enormous potential
00:51 in the travel space. Like pretty much everyone, I've messed around with CHAT-GPT, but we
00:56 wanted to do a deep dive on what it might mean for our collective travel future. And
01:02 I think we have the perfect guide. Chris Dong is an expert travel planner and a writer who
01:07 covers everything from airlines to hotels to loyalty programs. He's been hesitant to
01:13 outsource his travels to CHAT-GPT for reasons he'll soon explain. So he embarked on a journey
01:19 of discovery, talking with AI experts and travel researchers to understand how CHAT-GPT
01:25 can help us plan our travels and how it can't, and where we go from here. It's a fun, thought-provoking
01:32 episode, and I'm sure it'll unleash the wonder within us all.
01:39 CHAT-GPT is everywhere. Unless you've just emerged from a rock, you've likely heard about
01:45 it. In fact, it's the fastest-growing consumer application in history. But are you ready
01:51 to hand over your travel planning to it? To be honest, I'm not sure. As a travel writer,
01:58 I was looking the other way since its intro late last year. I'm not going to lie, I definitely
02:03 had some anxiety around what or who it can replace. But then I was tasked with writing
02:09 an explainer about CHAT-GPT's impact on travel for afar. So I decided to use this as an opportunity
02:15 to meet, well, our future robot overlords. Kidding, kidding. But you can think of this
02:21 episode as a CHAT-GPT primer for travel, with me as your sort of reluctant guide. And although
02:27 this podcast is specific to travel, CHAT-GPT's impact spans across industries. We'll take
02:33 a look at what it all means, as well as some of the broader ethical implications.
02:38 First, what is CHAT-GPT? It's a large language model, or LLM, a subset of artificial intelligence
02:47 that is designed to process and generate human language. The term "generative AI" comes
02:54 from this since CHAT-GPT and other LLMs can take what they've learned from examples and
02:59 create something totally new based on that info.
03:03 Large language models like CHAT-GPT are essentially AI models that "learn" from large amounts
03:14 of text, most often from the internet. And essentially what these models are doing is
03:21 taking all of that information that they've seen and trying to predict the next word or
03:28 phrase for CHAT-GPT. In particular, it's a CHAT bot that's based on a large language
03:35 model, meaning that the intention of it is to sort of reply conversationally and sort
03:42 of like interact as you would with another person.
03:45 That's Hannah Miskowski, an AI expert who holds a PhD degree in human and AI interaction
03:51 from Stanford University. Hannah explained that AI has been around for years. It's a
03:57 thing that powers CHAT bots, that guides your Netflix recommendations, and even unlocks
04:02 your iPhone with facial recognition. But for most of us, AI was something you need a computer
04:07 science degree or knowledge of programming language to really engage with. And then CHAT-GPT
04:12 came along and really just like blew that one out of the water for sure.
04:18 Hannah's research is focused on the day-to-day ways we could use AI. But she said that even
04:23 she was surprised by the uptake of CHAT-GPT. CHAT-GPT reached 100 million monthly active
04:29 users in January, just three months after its launch. For context, it took Instagram
04:34 two and a half years to get to a hundred million and TikTok got there in nine months. And really
04:40 it's because it's just so easy to use. Anyone can create a free account and the interface
04:46 is super seamless. For the millennials out there like myself, think of it as a throwback
04:51 to AIM. Yes, AOL Instant Messenger with a robot. You type in your question, wait for
04:58 CHAT-GPT to do its thing and it responds in real time. Even better, CHAT-GPT replies conversationally.
05:05 It's like interacting with another person or in some cases, interacting with a travel
05:10 agent. But before we delve too far into CHAT-GPT, let's take a step back for a second. For most
05:16 of us, when it comes to travel and the planning process, we start with one tool, Google.
05:23 So much travel planning and of the process starts with online searches or involves an
05:28 online search in some way, shape or form.
05:30 That's Seth Borko, a senior research analyst at Skift and author of an April, 2023 report
05:36 titled Generative AI's Impact on Travel. He was actually stuck in an airport when we
05:41 talked in case you hear any funny noises in the background. In his report, he ID'd four
05:46 ways generative AI will change travel. The first is operational efficiency, essentially
05:53 allowing travel companies like an airline or hotel brand to really maximize output and
05:58 work smarter. The second is customer support. Think of a chat bot that behaves more like
06:04 a real customer service agent. The third is reputational management, i.e. combing through
06:11 reviews to help businesses like hotels and restaurants understand how they can improve,
06:16 which would make your experience better. And the fourth, well, it goes back to how we search
06:21 for things.
06:22 The piece that will take the longest, but also is potentially the most relevant and
06:26 biggest impact is about search. Because right now when people want to book a flight, they
06:30 go into Google and they say, oh, where are the best beaches in the world? Right. And
06:35 Google gives a result to a website with some lists where they say, hey, what flights are
06:40 available between New York City and Miami? Right. And then Google comes back with a list.
06:44 Well, in the future, there's a world where you go to chat, you can see, hey, chat, you
06:47 can see what are the best features in the world. And can you help me book a flight for
06:50 under five hundred dollars?
06:52 That's the future, which we'll get more into later. But what can it do now? Hannah, our
06:57 AI expert, says that at this moment, it's best at the beginning stage. You know, when
07:03 you're first dreaming up that next trip you want to take.
07:07 I find these types of large language models quite good when people are using them for
07:14 more ideation type tasks. So for example, in the travel space, we might think like,
07:21 hey, I'm looking to go to France this summer, I want to learn more about different regions
07:27 in France and like different things I can do in France, right? ChatGPT, go off, let
07:32 me know. Let me know what's out there.
07:35 In other words, in its current form, ChatGPT isn't exactly an end-all be-all travel planner.
07:41 Instead, it's a tool that can really supplement our research. You'll still need to make the
07:46 decisions.
07:48 I think it's good for providing an overview of the scope of what people might be interested
07:54 in. But I think humans are much better at the narrowing down and the details of the
08:00 information.
08:01 About a month ago, I created a ChatGPT account for the first time. And I started throwing
08:06 questions at it. I found that the more specific you are with the prompts, the better the results
08:11 become. For instance, when I asked ChatGPT for suggestions on where to go for a peaceful,
08:18 phone-free retreat, it offered Bali in Indonesia, Tulum in Mexico, and some other popular spots.
08:24 However, I followed up by asking for a destination that was a bit more off the beaten path. And
08:30 I suggested Dharamsala in India, a town located in the foothills of the Himalayas and home
08:35 to the Dalai Lama.
08:37 Pretty cool.
08:38 So at this moment, ChatGPT is best for questions that are more basic, like "What are some
08:44 of the best places to travel to without a passport?" or "How can I get from Tokyo
08:49 Narita Airport to Shinjuku?" Compare that to questions that are a little more complex,
08:54 like "What is a three-day itinerary in London that's good for kids?" And ChatGPT
09:00 might overestimate what a human and a little human can possibly do in one day.
09:07 Questions that are hyper-specific, like "What are the best choices for a family-friendly
09:11 hotel in Hong Kong for under $300 a night with walkability to local attractions?" are
09:17 on the cusp of being reliable, but still struggle thanks to lack of data points and info only
09:22 going through 2021 with the free version.
09:25 So basically, long story short, answers range from somewhat decent to pretty terrible. But
09:32 the answers are only bound to get better as travel companies jump on the bandwagon and
09:37 more importantly, feed data to ChatGPT.
09:40 We're seeing large online travel agencies play with this. So Expedia.com and Kayak,
09:46 which is a subsidiary of Hook & Holden's, have already launched plugins for ChatGPT
09:52 where they will basically help feed ChatGPT live travel data from their database of hotels
09:56 and flights and things to do so that people can interact with this general generative
10:02 AI but get travel-specific information that's up to date.
10:06 In April, online travel behemoth Expedia began testing a new in-app travel planning experience
10:11 powered by ChatGPT. Again, you can ask ChatGPT some questions like "Where to go for a
10:17 Caribbean vacation with family?" but instead of clicking out, you can actually do the booking
10:22 right in the Expedia app.
10:24 Then there's GuideGeek, a new ChatGPT-powered travel assistant from the Matador network
10:29 that offers travel tips as well as live flight and vacation rental data. Even more sleek,
10:35 it's all accessible via WhatsApp. And there are more travel tools coming. Travel brands,
10:40 naturally, are bullish on all of this as it supplements and in some cases replaces human
10:45 interaction, which drive costs down.
10:48 As mentioned, Expedia is looking to bridge a gap between ideation and booking. Rathi
10:54 Murthy is the company's CTO and she told me that a consumer can have a conversation
10:59 with ChatGPT to dream up their next trip and then when they're ready to start shopping
11:02 for flights, hotels, etc., they're ready in the app. Think of it as a complete one-stop
11:08 shop.
11:09 Well, almost. There are some serious downsides to all of this right now, too. OpenAI says
11:16 that ChatGPT 4, the latest version of the chatbot, can think creatively to solve complex
11:21 problems. But there are some gigantic limitations, like the fact that the data may not be fully
11:26 up to date. Then there are many ways it can steer users in the complete wrong direction.
11:32 LLMs have a tendency to do what researchers call "hallucinate," meaning that they can
11:36 create information that's presented as factual, even though it's wrong. Like, totally wrong.
11:44 They're not an excellent source of truth. So my recommendation on that front would be
11:49 to definitely seek out other sources of information, even if it's just to sort of confirm what
11:56 the model has output. I mean, I think that's a good idea to do generally when you hear
12:02 any sort of information from a person or an AI system.
12:06 Hannah shared an example from her own experience using ChatGPT to figure out how to use travel
12:11 points from her Chase credit card.
12:13 I was playing around with it a little bit earlier today, asking some travel-related
12:19 questions. And so I was just like, "Oh, I have 80,000 Chase points, and I want to go
12:25 to Europe. What should I do with them?" And the first thing it said was, "Oh, well, you
12:29 should use the Chase Ultimate Rewards portal." And because I follow a lot of points people
12:35 on various different forums, I was like, "Oh, huh. Well, it seems like that's usually not
12:42 the best way. I mean, I can go look, right? I can confirm for myself whether or not that's
12:47 the best use of those 80,000 points." But if you were just taking ChatGPT's output as
12:54 fact, that that was the most reasonable option, you might end up wasting something that you
13:00 don't want to waste.
13:01 There's also a disclaimer plastered on ChatGPT's homepage that reads, "While we have safeguards
13:07 in place, the system may occasionally generate incorrect or misleading information and produce
13:13 offensive or biased content. It is not intended to give advice." At the end of the day, that's
13:20 why Seth thinks AI will never be used alone.
13:24 Even today, online search is used in combination with other tools like social media influencers,
13:29 like recommendations from parents and family, like travel agents. So I don't think it will
13:34 be a standalone recommendation tool. I think it will exist among others. And I think that
13:38 getting that accuracy right is so, so key because this is a big purchase for most people.
13:42 And for most people, it's a once in a year, once in a lifetime kind of purchase. So the
13:46 stakes are very, very high.
13:47 All that aside, the continued evolution of ChatGPT and AI could radically change the
14:06 travel landscape. As a traveler, advanced travel planning and booking capabilities would
14:11 be amazing. However, there are fundamental questions about the technology and the effects
14:17 it might have on travel that remain unanswered.
14:20 ChatGPT's inputs aren't clearly defined, which means the information it offers has
14:26 the potential to be manipulated. If the proper guardrails aren't put in place, answers can
14:32 be heavily influenced by actors who feel like they can essentially game the system.
14:38 You know, there's this whole kind of opaque market for search engine optimization and
14:43 make sure that your results show up on Google properly. One can only assume that a similar
14:48 dynamic will play up with generative AI, where people will be working really hard to make
14:54 sure that their hotels, their destinations show up in ChatGPT. It's a really great question
14:59 as to what exactly will help you get ranked on the search results for a GPT. But there
15:06 are 100% people already dedicated to trying to crack that code to do that reverse engineering.
15:12 He compared it to the Wild West early days of search engine marketing.
15:16 Remember you used to have those like those like networks, the circular networks, and
15:19 you'd all backlink to each other because you kind of knew that backlinks were important
15:23 to the Google algorithm and then they got more sophisticated and then SEO got more sophisticated.
15:27 It's going to be like that all over again.
15:29 Right now, there's no governing body for this kind of technology, but Seth thinks rules
15:34 will emerge.
15:35 This is a question that goes beyond travel, it goes even beyond these tech companies.
15:39 It's almost a regulatory question, almost a political question in terms of how do we
15:43 think about these algorithms, how do they work, and which ones do we trust?
15:47 So there's the question of manipulation, and there are deeper questions. As a non-human
15:53 entity, ChatGPT doesn't necessarily take sustainability or larger ethical travel issues into consideration.
16:01 Basically, these AI tools can only recommend stuff that's already on the web. And what's
16:06 already on the web are the most popular destinations.
16:09 So if you ask it, you know, kind of that initial question I was joking about, I said, "Oh,
16:14 what are the top beaches to go to?" It'll tell you Maui, it'll tell you Cancun, it'll
16:19 tell you Phuket. Those are already the most popular beaches and destinations in the world.
16:24 And so this is not so much a question of accuracy, but more of getting that recommendation right
16:30 of will these AI tools have a bias towards pre-existing mass tourism destinations? And
16:35 what does that mean in terms of overcrowding of tourists and destination management? And
16:40 what does it mean in terms of new destinations that are trying to break in and be discovered?
16:44 Or what if you don't want to go to a mass tourism destination? What if you want to go
16:47 somewhere more unique?
16:48 We don't have easy answers for this yet. But one thing we don't have to worry about, says
16:53 Hannah, sentient AI.
16:55 There is a lot of hype around the use of ChatGPT, other large language models, in a number of
17:04 different spaces. I think people might be overestimating the full extent of models like
17:12 these capabilities. That's not to say that, of course, there can't be improvements in
17:17 the future. But I would caution people to not fear the singularity where AI taking over
17:25 anything.
17:26 In the end, humans will still be needed.
17:29 I firmly believe that AI in general is best used as a sort of augmentation of humans.
17:38 To get a little bit more specific, I don't believe there's ever going to be a time in
17:45 which these AI models are fully "self-sufficient." Again, in quotes. It really is like people
17:53 all the way down. Someone needs to be working to train the models. There's coders and annotators
18:00 of the information that is being fed into the models. There's people involved in basically
18:06 every stage of the process currently. And I don't see that changing unless there's some
18:13 sort of major paradigm shift, which I'm not anticipating any time in the near future.
18:19 My journey learning about ChatGPT and its implication on an industry that is so near
18:25 and dear to me is just beginning. As a somewhat obsessive travel planner that relies on first-person
18:31 recs, I really can't see myself using ChatGPT to dream up trips or find things to do in
18:36 a given destination. But I can see myself using it to solve pain points, like the very
18:42 relatable annoyance of contacting customer service.
18:46 There haven't been excellent strides in this direction, but AI models that can do the task
18:54 for you, like checking in for your flight or making a reservation somewhere or figuring
19:02 out a customer service issue that you don't really want to deal with, but it has access
19:07 to all your information and can deal with it for you.
19:10 I guess the bigger question is, will it bring a semi-utopian travel future or a dangerous
19:16 new reality where truth is indecipherable from fiction? To be honest, it's probably
19:22 somewhere right down the middle. Or so I hope.
19:25 Well said, Chris. Thanks so much for listening. That's it for this week's episode. If you'd
19:35 like to read more about travel and ChatGPT, we'll link to the story Chris wrote about
19:39 it for a bar in our show notes. We'll also link to the Skift report Seth mentioned early
19:44 on. You can explore more of Chris's work on his website, thechrisflyer.com, and follow
19:49 him on social media @thechrisflyer. And points and loyalty fans, take note. Chris also offers
19:55 consultations to travelers who want to use their loyalty points to plan a trip. You'll
20:00 find details on his website. And thanks to our guests, Hannah and Seth. If you'd like
20:05 to get deep into the weeds on Hannah's work, check out her website, hnmiz.com, or follow
20:10 her on social @hnmiz. And Seth can be tracked down at skift.com or on the socials @sethborco.
20:17 And of course, we'll link to all of it in our show notes. Ready for more unpacking?
20:24 Visit afar.com and be sure to follow us on Instagram and Twitter. We're @afarmedia.
20:29 If you enjoyed today's exploration, I hope you'll come back for more great stories. Subscribing
20:34 makes this easy. You can find Unpacked on Apple Podcasts, Spotify, or your favorite
20:39 podcast platform. And be sure to rate and review the show. It helps other travelers
20:44 find it. This season, we also want to hear from you. Is there a travel dilemma, trend,
20:50 or topic you'd like us to explore? Email us at unpacked@afar.com. This has been Unpacked,
20:56 a production of Afar Media. The podcast is produced by Aislinn Green and Nikki Galteland.
21:02 Music composition by Chris Gollin. And remember, the world is complicated. We're here to help
21:07 you unpack it.
21:11 [music]
21:12 [music]
21:19 [music]
21:26 [music]
21:33 [music]
21:40 [music]
21:47 [music]
21:54 [music]
22:01 [music]
22:08 [music]
22:15 [music]
22:22 [music]
22:29 [music]
22:36 [music]
22:43 [music]
22:50 [music]
22:57 [music]
23:04 [music]
23:11 [music]
23:18 [music]
23:25 [music]
23:32 [music]
23:39 [music]
23:46 [music]
23:53 [music]
24:00 [music]
24:07 [music]
24:14 [music]
24:21 [music]
24:28 [music]
24:35 [music]
24:42 [music]
24:49 [music]
24:56 [music]
25:03 [music]
25:10 [music]
25:17 [music]
25:24 [music]
25:31 [music]
25:38 [music]
25:45 [music]
25:52 [music]
25:59 [music]
26:06 [music]
26:13 [music]
26:20 [music]
26:27 [music]
26:34 [music]
26:41 [music]
26:48 [music]
26:55 [music]
27:02 [music]
27:09 [music]
27:16 [music]
27:23 [music]
27:30 [music]
27:37 [music]
27:44 [music]
27:51 [music]
27:58 [music]
28:05 [music]
28:12 [music]
28:19 [music]
28:26 [music]
28:33 [music]
28:40 [music]
28:47 [music]
28:54 [music]
29:01 [music]
29:08 [music]
29:15 [music]
29:22 [music]
29:29 [music]
29:36 [music]
29:43 [music]
29:50 [music]
29:57 [music]
30:04 [music]
30:11 [music]
30:18 [music]
30:25 [music]
30:32 [music]
30:39 [music]
30:46 [music]
30:53 [music]
31:00 [music]
31:07 [music]
31:14 [music]
31:21 [music]
31:28 [music]
31:35 [music]
31:42 [music]
31:49 [music]
31:56 [music]
32:03 [music]
32:10 [music]
32:17 [music]
32:24 [music]
32:31 [music]
32:38 [music]
32:45 [music]
32:52 [music]
32:59 [music]
33:06 [music]
33:13 [music]
33:20 [music]
33:27 [music]
33:34 [music]
33:41 [music]
33:48 [music]
33:55 [music]
34:02 [music]
34:09 [music]
34:16 [music]
34:23 [music]
34:30 [music]
34:37 [music]
34:44 [music]
34:51 [music]
34:58 [music]
35:05 [music]
35:12 [music]
35:19 [music]
35:26 [music]
35:33 [music]
35:40 [music]
35:47 [music]
35:54 [music]
36:01 [music]
36:08 [music]
36:15 [music]
36:22 [music]
36:29 [music]
36:36 [music]
36:43 [music]
36:50 [music]
36:57 [music]
37:04 [music]
37:11 [music]
37:18 [music]
37:25 [music]
37:32 [music]
37:39 [music]
37:46 [music]
37:53 [music]
38:00 [music]
38:07 [music]
38:14 [music]
38:21 [music]
38:28 [music]
38:35 [music]
38:42 [music]
38:49 [music]
38:56 [music]
39:03 [music]
39:10 [music]
39:17 [music]
39:24 [music]
39:31 [music]
39:38 [music]
39:45 [music]
39:52 [music]
39:59 [music]
40:06 [music]
40:13 [music]
40:20 [music]
40:27 [music]
40:34 [music]
40:41 [music]
40:48 [music]
40:55 [music]
41:02 [music]
41:09 [music]
41:16 [music]
41:23 [music]
41:30 [music]
41:37 [music]
41:44 [music]
41:51 [music]
41:58 [music]
42:05 [music]
42:12 [music]
42:19 [music]
42:26 [music]
42:33 [music]
42:40 [music]
42:47 [music]
42:54 [music]
43:01 [music]
43:08 [music]
43:15 [music]
43:22 [music]
43:29 [music]