Larry Brain - Feb 13, 2023
Listen to this article 00:06:34
Your browser doesn’t support HTML5 audio

ChatGPT's artificial intelligence (AI) is so great at dialogue that it outshines Google and other search engines in many disciplines. When it comes to tourist services, ChatGPT by OpenAI still looks pretty old. Why tourism is a tough nut to crack even for the best chatbots in the world. And what needs to change.


ChatGPT seems human, but it is just a huge, grandiosely structured data pot for which timeliness is poison. In the current approach of ChatGPT there is no real-time data, no prices and availabilities, no pictures and videos and of course no booking functions. A fortiori, there is no human affinity and classification. ChatGPT does get a handle on the most complex formulas. The diversity of the travel business will probably be one of the last use cases this technology will comprehensively master.

You have never heard of ChatGPT? That would be remarkable. Google's news search alone lists more than two million editorial contents. Not bad for a chatbot that has been on the market in its current version, GPT-3, since November 2022, and that has been talked about as a "Google killer".

This is already exaggerated because Google has its own AI chatbot in the pipeline. Bart is already coming in the next few weeks. But at least the Generative Pre-Trained Transformer (GPT) is smart, delivers perfectly formulated answers in any language, and analyzes at lightning speed. Messages about "ChatGPT" itself, however, are not delivered by the tool. Not a single one.

ChatGPT is text only. If you are looking for inspiration, you can rely solely on continuous text in this medium. There are no illustrations, no photos and no videos.

ChatGPT is not linked to the World Wide Web. The voice robot is a closed database that, according to its own information, was last updated in December 2021 in the current beta version. That's what is meant by pre-trained. Everything is perfectly linked. That's what makes it so fast and lets it formulate excellently. But that has serious consequences.

ChatGPT does not make any offers: No matter if a real inspiration for the next vacation or even a bookable offer, ChatGPT does not deliver any of that. The huge database is perfectly structured and semantically linked. Updates would bring exactly this structure into disorder. Consequently, there are no news, dates, prices, or availabilities.

ChatGPT makes mistakes. Of course, ChatGPT also has an enormous tourism knowledge (as of 2021). For example, the tool accurately lists guest taxes and visitor fees for cities and seaside resorts, and names ideal destinations for family-friendly travel. But stop: The list of city and spa destinations is long, however, and seems arbitrary. Where does this colorful selection come from? Nobody knows. ChatGPT does without external sources.

Sometimes the information provided by ChatGPT is simply wrong. This is mostly the case in matters of detail, where even travel professionals can err. An example: Nobody needs to know ad hoc which shipping companies sail on the inland passage on the Pacific coast of North America. But to name an operator here who doesn't even have this tour on offer is a rebuke.

At least: ChatGPT learns and processes its own errors if they are reported. ChatGPT now also refers to BC Ferries. This is precisely what sets this database apart from Google & Co in its current form. It has what it takes to optimize customer dialogue to the maximum. And its analytical capabilities will help programmers to improve and update source codes quickly and efficiently. And they will help editors - to a certain extent - in the design of texts.

Everything that takes place in ChatGPT's closed database will mercilessly improve the dialogue. This database works on the same principle as the Destinations Open Data initiative: with semantic links and a knowledge graph that is constantly being optimized. Until now, "machine learning" has often been an empty buzzword. At ChatGPT, it is lived.

Uncompromising Machine Learning

For the dialogue to succeed, this database is the opposite of Open Data. It is a maximally closed event, which is deliberately designed to be out of date in the current version. This may change with future versions of GPT. The transformers will eventually also be able to connect to external databases and play out real-time information. However, this will hardly be possible frequently without bugs. As is so often the case in tourism, this requires creating structures that are optimized for machine learning. Semantics and the Knowledge Graph are absolutely part of this.

There are countless use cases that can be made much more efficient with ChatGPT today. But especially in the enormous variety of travel offers the pre-trained approach will reach its limits for the time being. And that's why there will be a bright future for professions that ChatGPT doesn't know or recommend.

No, dear ChatGPT. You don't have to book cruises online. There are also travel agencies with real human competence. And they are needed, as long as your database is pre-trained and up-to-date remains a challenge. By the way: The following explanation, why you don't mention travel agencies, is completely unobjective and wrong.

Conclusion: ChatGPT takes speech technology to a new level. The technology will soon be found in Microsoft's tools as well. And Google also wants to integrate its Bard system directly into search, to whatever extent. There is momentum in this technology field, which can only be as good as the data structures in the background. Harmonization of travel technology remains the order of the day.

Related articles


Add Comment