The internet is abuzz with speculations about what the boundaries of AI might be. According to enthusiasts, there are barely any limits at all, and the aim is to eventually help travelers create perfect travel itineraries using a simple keyword search. Will it ever be as simple as asking ChatGPT for a trip to a hot tropical beach and picking up the tickets a few minutes later?
Not according to the intrepid researchers who have been gleefully picking apart such daring claims. We have yet to see an article loaded with praise and no serious criticism. The consensus is that AI can be a useful tool in the future, but it currently has too many weaknesses.
In response, we’ve created a list of reasons why ChatGPT shouldn’t yet be given the keys to the travel kingdom. But we have a lighthearted warning before we start: ChatGPT is astonishingly articulate. It’s very polite, very politically correct, and very apologetic when you catch it making a mistake. We know how easily travelers adapt to other cultures by slipping right into the narrative. You’ll soon start taking extra time to thank ChatGPT effusively after every exchange!
Despite ChatGPT’s confident conversational style, peeks under the hood reveal some serious recurring problems and shortcomings.
ChatGPT has been in the making for years, during which time it was trained on a vast, though unspecified, library of digital resources. The cut-off date was September 2021, which is why it may confidently recommend a hotel or restaurant that has since closed down. The premium version includes access to more updated information, but it’s not error-free.
It provides a reference list for complex answers, but it's often unclear exactly where the information came from. Sometimes, it seems to be quoting verbatim from a single source, and other times it seems to mash together the results from several sources.
The problem is that ChatGPT delivers answers in such a confident tone that users take everything at face value. It’s only when you start checking every source in every conversation that the board starts lighting up in red.
ChatGPT positions itself as a neutral source of information and even prefaces its answers with a disclaimer. For example, if you request a list of the top 5 restaurants in the world, it will first declare that "As an AI language model, I cannot have personal preferences or opinions" before gamely producing a list of names.
But ChatGPT can’t eat, taste, smell, or experience the delights of a fantastic meal, so the list is simply a mash-up of the opinions of real humans pre-2021. Dive a little deeper: click on the reference links below ChatGPT’s answers and check the credibility of the source. But beware, because in some cases, the source may not even exist.
The poor chatbot has been force-fed mountains of (sometimes bland, often florid, and occasionally ridiculous) public content produced by writers of varying skill and competence levels. Therefore, ChatGPT sings the same praises over just about every city, street, hotel, meal, or tourist attraction.
Be prepared to see almost every town being variously described as a "food lover's paradise," “walker’s paradise,” “birdwatcher’s paradise,” or “shopper’s paradise,” etc. You’ll also encounter far too many overused terms like “vibrant,” “chic,” “designer,” and “trendy.”
The recurring pattern is that ChatGPT always first mentions the biggest tourist spots in each popular destination. Even its off-the-beaten-track recommendations are mainstream.
Again, this is mostly a reflection of its dependence on pre-2021 articles and information. In the real world, many of yesteryear’s popular destinations are now trying to cope with the detrimental effects of over-tourism and overly commercial tourist traps.
Interacting with ChatGPT feels like chatting with a human. But, unlike humans, ChatGPT is oblivious to the sub-text that is central to human conversations.
For example, consider a discussion about nightclubs in Thailand. A human might spontaneously throw in some casual tips for getting to your hotel cheaply, quickly, and safely at 3 AM - after a night of too many cocktails and energetic dancing in a hot, high-humidity environment. ChatGPT won’t add such information without prompting.
But the difference is that the chatbot interface feels so ‘real’ that you vaguely expect it to. You kind of assume that it would have added such information if it knew. It’s a letdown when you realize that, unlike a human, it never will add spontaneous tidbits. AI doesn’t “get” humans even if its confident interface makes it seem that it does.
That’s why it’s hard to identify possible gaps in ChatGPT’s answers. Prompt engineering (learning how to phrase questions to cover every eventuality and get the best possible answer) is a skill that currently borders on art.
Do you have the time to learn it?
But just when you get the feeling that you’re starting to ask the right questions, your screen may go blank or display a “disconnected” or “access denied" error message. That’s because free users get a daily limit. Additionally, they may get bumped when the demand on the servers reaches certain levels.
If you get bumped off ChatGPT halfway through your conversation, you’ll lose the entire chat history. You’ll have to start from the beginning when you regain access. That might prove a disaster if you happen to be stuck in a strange city with no money or passport.
ChatGPT is very much a product under development. The problems are glaringly obvious. It must analyze every interaction to improve the product. And, since you need to log in to use it, you can be sure that your chats will be stored and used for research. It raises serious ethical concerns about how your personal data may be used, but how else can they deliver on their promise of planning personalized holidays by starting with one keyword?
Never provide sensitive information. And among the advantages and disadvantages of VPN, there’s a strong argument for using a VPN with a link checker. You’ll need to protect your IP address and other information while you check facts and run down possible hallucinations on links that could, occasionally, lead to dangerous sites.
Currently, ChatGPT doesn’t have the contextual knowledge of a skilled trip planner. It can’t yet replace human expertise. But, the AI landscape is changing so fast that the situation might be entirely different within the next year. There’s great potential, but ChatGPT is not ready to be let loose on unsuspecting travelers yet.