A chatbot in front of a computer - Source Mohamed Nohassi on Unsplash
  The potential of GPT models for the travel industry

Amadeus;

It can be difficult to get a firm overview of what GPT (Generative Pre-trained Transformer) models might mean for the future of the travel industry – or, indeed, humanity. While many commentators have an opinion, there are comparatively few experts in the field able to offer definitive insight.

I do not pretend to be one of them, but by being a fierce and “greedy” observer of what is happening in the tech field and in the travel industry, I can see all the potential but also the current limits of the tool.

What we know so far.

Today there are already some key points that can be stated with some certainty.

Firstly, GPT models – and the wider ‘generative artificial intelligence (AI)’ field – all have tremendous possibilities. The technology, although not new, has the potential to transform everyday life and thus create a huge appetite and concerns.

The chatbot (that has been around for a while), by comprehending and generating human-like natural language in such a stunning manner, has made it widely adopted. ChatGPT has indeed taken the world by storm, racing to accumulate 100 million monthly active users in a mere 40 days, with millions more signing up each month. The tool is innovative because of the sheer volume of data it was trained with – it exceeds previous chatbots by orders of magnitude. ChatGPT 3 was trained with 17 billion trainable parameters and 45 terabytes of data, for example. For reference, the Hubble Space Telescope generates ten terabytes of data per year. In other words, Chat GPT has read more text than a million people will read in their lifetime put together! And let’s not forget that ChatGPT 4 is already in the marketplace and from what I have seen so far, it is far bigger than ChatGPT 3.

However, many experts from Yann Le Cun (Meta Chief AI Scientist) to Sam Altman himself (OpenAI CEO) are expressing the limits of the tool, from its nature, the way it is trained, the access to data or its inability to provide proper predictions. Nonetheless, its current possibilities and its capacity to change the way we work remain impressive, and worth the investigation.

What can GPT models do for the travel industry?

It is the distance between this clear potential and current application which is causing a great deal of uncertainty. It is indeed important to remember that commercial uses, in the travel sector and elsewhere, are currently limited.

Most of the use cases in the travel industry are focusing on how GPT models could impact travelers at each stage of the journey.

For example, GPT models could be used in trip planning. Imagine if a traveler could say, “Hey, I want to do nature-based activities in southeast Asia. I have two weeks, and this is my price range. What do you recommend?” and they would receive a complete trip itinerary in just a few seconds, including flights, hotels, train tickets and transfers to book. This could revolutionize how people search for and book travel. Expedia for example, has started to address this.

On trips, AI-powered instant translation services could make it easier for travelers to manage disruptions or negotiate last-minute changes directly with hosts who do not speak the same language. Machine learning models could also automatically inform a host about a late arrival at their hotel due to a delayed flight or allow them to change bookings on the spot.

The tool could also help travel companies enhance the overall shopping experience, allowing people to go from inspiration to booking in a much more streamlined manner. It could provide first level customer support, but also analyze customer feedback and identify trends or patterns, providing valuable insight into customer sentiment and opportunities for personalization.

Beyond the traveler experience, there are many other possibilities GPT models could offer to travel actors or corporations, this can be done by providing a wider source of content, accelerating, and automating some processes. Typically, in business travel, budget control and forecast, policy enforcement, and expense auditing, could all benefit from GPT models.

What is Amadeus doing in this area?

Today, the examples I will describe below remain works in progress. At Amadeus, we strive for innovation, it is part of our DNA, and therefore we are developing a number of pilots to move them closer to reality. As for any innovation project, it is important to stress these projects are not guaranteed to succeed - but they are nonetheless exciting areas of exploration. As American president, Franklin Roosevelt, famously remarked: “It is common sense to take a method and try it. If it fails, admit it frankly and try another. But try something.”

Firstly, in Search & Shopping for airlines for our corporate customers, we are working to build a conversational chatbot to help users complete a booking using natural language. In collaboration with Microsoft and Accenture, a prototype has been built and we are working to integrate the technology with Cytric Easy, with plans to explore the addition of ground transport and hotel bookings.

This gets to the core of the technology – and how Amadeus can use it. GPT models can translate human requests into API requests using live data, which can be actioned into real bookings.

At the same time, we are working together with Microsoft on a co-pilot tool to help generate code across Java and other languages. Amadeus is piloting the tool and working with developers to understand the benefits in this area. 

Finally, we are looking at improving content integration. Cytric aggregates content from many providers, using APIs, which can be a cumbersome process. In experiments we have conducted, ChatGPT was able to reduce individual components of some integration tasks from days to a matter of minutes. 

But are GPT models safe to use?

At Amadeus, while we are on the positive end of the scale, as with everything we must err on the side of caution in this fast-evolving space. In response, we have developed principles that guide AI development and its uses within our company. 

First and foremost, we take the management and usage of this technology very seriously and carefully to ensure the respect of end-users and customers’ data privacy, AI Ethics and internal legal policies. We also manage the acceptance of the deployment on a case-by-case basis and have started educating our employees on the principals to comply with.

To conclude, GPT models are a technology that offer a multitude of possibilities - but much remains to be seen.

Today at Amadeus our teams, in collaboration with customers and partners, like Microsoft, are looking at how to maximize this potential.  Looking to help make the experience of travel better, free our employees from tedious and time-consuming tasks, accelerate our developments and enable corporations to be more efficient. 

With much more to come, I will continue to take a close interest in what is to come.

Rudy Daniello is the Executive Vice President Amadeus Cytric Solutions