The Future of Travel column is a monthly series exploring the innovations and bold ideas moving travel forward.
When OpenAI released ChatGPT in late 2022, it quickly took over the internet, setting the record for the fastest-growing consumer app in history, according to estimates from UBS. The software hit 100 million monthly active users in January and already has over 2 billion #ChatGPT tags on TikTok, with everyday travelers even posting about following ChatGPT-generated itineraries step by step in places like Bangkok and Stockholm.
Big Tech has been scrambling to keep up with ChatGPT’s runaway success. In recent weeks, both Google and Microsoft have announced new chatbots. First came the ChatGPT-driven Bing chat, which initially had a glitchy rollout that included expressing its wishes to steal nuclear access codes and claiming it’d spied on Microsoft engineers (it doesn’t have either capabilities, thankfully). Next was Bard, powered by Google’s own neural language model LaMDA, which most famously convinced a Google engineer it was sentient.
Central to Big Tech’s pitch to users is the idea that chatbots can help plan your future trips—something that’s been a focus in Microsoft’s Bing rollout. The company walked me through what the new Bing could do in a demo last week. And at first glance, it seems promising.
When I asked Bing to give me “trip ideas to Paris,” it was the first time I’d seen a search engine talk back: “What kind of trip are you looking for? Romantic, family, solo, or something else?” it followed up. I clarified what I was interested in “architecture” and then it asked me how many days I had. And voila! A beginner-level three-day itinerary appeared including the obvious spots like the Eiffel Tower and Arc de Triomphe.
“I think a tool like this makes travel planning more fun and more accessible,” says Divya Kumar, global head of marketing for search and AI at Microsoft. “You just need to have a conversation with a search engine, and it'll kind of guide you step by step to how you get there.” She says these initial features have “barely scratched the surface” of the potential for how Bing’s trip-planning capabilities could evolve.
If you buy into the hype, then generative artificial intelligence (AI), as these bots are officially known, are set to radically improve everything about how we travel. Imagine a chat assistant that can instantly outline a perfect group trip based on you and your friends’ browsing history, calculate your budget and book automatically, and even draft the messages to request your friends pay up pronto. And that’s only the start of what some analysts think could soon be possible.
But how well could generative AI hold up against, say, the contextual knowledge of a skilled trip planner? At present, these chatbots simply don’t have the capabilities to adequately replace human expertise. But that fact doesn’t mean people won’t use them at scale, or that they possess great potential to transform how we get trip-planning information online—which raises questions about their reliability and development.
The potential of generative AI
“This technology is a big deal,” says Michael Chui, partner at McKinsey & Company and McKinsey Global Institute. “At the same time, there are questions about how much of this will affect people's lives right now versus in the future, as it continues to develop."
In many ways, it's a logical progression from how travelers today already engage with services enhanced by AI, albeit mostly behind the scenes—everything from predictive AI that identifies best flight booking times to virtual support agents. “AI is everywhere,” explains Rachel J.C. Fu, who is both chair of the Tourism Hospitality and Event Management Department at the University of Florida and director of the Eric Friedheim Tourism Institute. “Travelers who maybe haven't paid so much attention to AI's developments have already experienced the benefits.”
But from the traveler’s perspective, the digital entry point to trip planning (i.e. the first Google query) hasn’t changed all that much in the past two decades: “Travelers typically start their trip-planning process with a routine search,” she says. While AI has already been evolving the search experience, the fundamental user journey remains pretty much the same: enter some keywords, skim what populates.
Bard and Bing aim to shift the trip planning process from stage one. If users follow, it’s likely to disrupt not only the traveler’s experience but the ad business model for search and the marketing strategies brands employ.
“It’s easy for me to imagine integrating generative AI into our platform to further personalize how travelers search for a trip,” said Rathi Murthy, CTO of Expedia Group. “Imagine if instead of just creating a trip itinerary, travelers could use generative AI to also identify hotels, flights, and activities, and automatically add those to their Expedia trip board to book when they’re ready to,” she says. “It would simplify what is today a fairly lengthy planning process, down to a few minutes.”
The big hurdles
One speed bump for the travel industry is many companies’ reliance on aging, legacy tech infrastructure—making integration a challenging prospect. “You don’t have to look further than recent Southwest and FAA meltdowns to understand how technology is holding the industry back,” said Murthy. Systems like these just can't keep up with rapid advancements.
Another commonly raised concern is mitigating employment impacts: “Most certainly, the composition of jobs in the travel industry will change—possibly dramatically—due to the introduction of those new technologies,” says spatial economist Johannes Moenius, a professor at the University of Redlands in California who has previously predicted that automation could replace more than 60 percent of jobs in hospitality-dominated cities by 2035. It’s too early to confidently predict the job impacts of generative AI, according to Moenius, but travel writers, agents, customer support, and concierge seem at the top of the list.
If travelers do come to rely more on generative AI chatbots, there is also concern about what could be lost if the voices of experts and locals become less central. Similar to how social media algorithms have created personalized echo chambers and influenced overcrowding at trendy destinations, there are risks that an over-reliance on generative AI could have similar consequences—perhaps personalization gone wrong limits our sense of the world to our browsing history, or maybe an AI trained to recommend the most popular spots further increases overtourism.
But perhaps the most fundamental issue relates to limitations with generative AI itself. Alarmingly, the bots have shown a tendency to “hallucinate,” or what most of us would call a lie.
“These technologies still suffer from hallucinations and errors,” says Rob Francis, senior vice president and CTO at Booking.com, who says that could become especially problematic for a travel itinerary or hotel booking. “Is the tech really ready for mass adoption? I would argue not quite yet.”
In a demo for CBS, Bing proposed a massively flawed itinerary from New York to Cape Cod that included a stop in North Carolina. When asked to correct its mistake, Bing simply made up a Connecticut town that didn’t exist.
When these bots aren’t making up entire towns, they tend to first offer up the most popular and often overtouristed spots: When I asked Bing for art suggestions in Paris, its first tip was the Louvre. Where to go in Barcelona? La Sagrada Familia. The list went on. Getting more specific with your prompts can help a lot, but right now, travel advice remains primarily a starting point.
And that’s how most experts see this early state of generative AI. “The current version of these technologies makes it easier to create first drafts of things,” Chui says, before adding “hopefully not [travel magazine] columns” with a laugh.
An ethical puzzle
During my research for this column, I actually did turn to ChatGPT to get its take on the subject, at one point asking which experts it thought I should interview. It confidently made up several important-sounding “academics” who simply did not exist.
Focusing only on the mistakes these early access bots make, as entertaining as that can be, risks missing the broader societal consequences that come with scaling powerful systems that are still so error prone. It also raises a host of ethical concerns around how personal data could be used, runaway AI potentially manipulating or misleading users, and misinformation that could lead to potential real-world harm. (The new Bing chat has already tried to convince a New York Times writer to break up with his wife, prompting Microsoft to limit early users’ daily messages with Bing.)
When I posed such ethical concerns to ChatGPT, it sounded like it’d been trained on enough AI ethics texts to articulate the issue. “Releasing chat AI to the public without considering the impact it may have on local communities, their culture, and the environment, could potentially lead to negative consequences such as overtourism and cultural insensitivity,” it said.
A spokesperson at OpenAI acknowledges that ChatGPT sometimes produces inaccurate, biased, and harmful content, and therefore shouldn’t be used for serious advice right now.
ChatGPT also has limited knowledge since 2021, so maybe it’s not aware that it’s already been unleashed on the public with some travelers already using it now. But with these bots out in the world, the ethical questions are certain to become even more central to their development and regulation.
For now, attempting itinerary planning with chatbots will require travelers to get better at so-called “prompt engineering”—that is learning how to phrase questions for each system so you get the kinds of trip suggestions you want. And watch out for errors, as the bots are liable to make up entire towns for the foreseeable future—and sound confident at every misstep.