There are travel startups building businesses entirely using artificial intelligence (AI). And there are established businesses who have taken the bold move of rebuilding using AI.
But as an airline, or a group of airlines, that has to make safety the priority and is governed by huge amounts of legislation, it’s a different story.
This is the somewhat unenviable role of Ben Dias, chief AI scientist at British Airways-parent IAG. His challenge is to decide how and where to implement AI within the confines and guardrails it operates under.
Speaking at the Skift Data + AI Summit in London earlier this week, Dias shared some of the group’s approach to AI and how it manages the potential risk.
IAG focuses on challenges at a group level and “domain transformation” such as maintenance, repair and overhaul, according to Dias. It operates a federal approach for the carriers, who have their own AI teams, enabling them to work on their own use cases.
There are also risk committees within each airline, with each carrying risk assessment on those use cases “so it’s a quicker turnaround, they’re closer to the information, etc.”
Dias acknowledged that, in general, governance can be easier because some decisions have to be made by humans. But he added that “there’s a general rule when you augment the human by default and you automate by risk level.”
Recovering when AI goes wrong
When it comes to automation, you ask three questions, he said.
“How critical is it, and if it fails, what is the cost in revenue, safety, customers and reputation? If it fails, what’s the blast radius? Is it self-contained or does it cascade across other systems? What is your fallback if it fails, and AI does fail?”
Dias added that plenty can be automated if the technology or system is quick to recover, but if not it’s down to risk assessment.
He also shared a little of the company’s AI policy, which has 10 principles including elements such as “no bias and discrimination,” human oversight and what Dias described as wanting AI initiatives to be “adequately accurate.”
“Every solution is mapped against these requirements, and if it doesn’t [meet them], we don’t say ‘no.’ We just say ‘What can you put in place to mitigate [the risk]?'”
Others at the event also shared their approach to AI and some of the related challenges.
Freedom to build
Nicolas Maynard, SVP of data and AI for Accor said that the hotel group’s approach is more centralized.
He said the company is trying to contain the rise of vibe coding a little, “reining it in without causing a bottleneck.”
“A lot of folks who try to do development or think they can do development, think that because it works on their laptop it’s going to work everywhere. That’s very dangerous for us because we want to make sure we protect our customers’ data and our IP.”
Keeping initiatives central means they are closer to development knowledge and diminishes shadow AI, Maynard said. It also drives the regions to comply with the necessary regulations such as the EU Artificial Intelligence Act.
“It’s not to slow them down but to ensure they have the information on the risks they are taking. What is the data they need to use and is it safe? They have to accept that to run it full scale, it has to be production grade.”
Kari Anna Fiskvik, chief digital and technology officer at Norway-based hotel group Strawberry, echoed his sentiment and said that with AI “everyone thinks they are a builder.”
“It comes with a lot of compliance issues and bad prompt issues. We try not to shut them down because we want them to learn, we want them to experience. It’s a bit of shifting every week, let people try to use it and monitor costs. If we don’t let them use it, they’re not going to come up with the great ideas.”