There is an inherent uniformity to cabin crew – from the composed expressions [VT1] to the synchronised safety demonstrations down to the uniform itself; effected with almost computational accuracy. Qatar Airways seems to have had a similar train of thought with its generative artificial intelligence – or AI – virtual cabin crew, Sama.
Showcased at Web Summit Qatar 2025, Sama – meaning ‘sky’ in Arabic – was displayed in her holographic glory. While her pearly whites do little to disarm the uncanniness of her 3D-rendered eyes, the programme she represents is the key factor. Qatar’s generative ‘AI cabin crew’ allows customers to book through voice or text chat with the programme. Perhaps veering into the gimmicky nature of technology – fully utilising the excitement driving AI’s development – there is no denying the recent AI explosion.
While ‘AI’ has been used as a catch-all expression, it actually comes in many forms – such as predictive and interpretive AI – for various use cases. Furthermore, it is not a new phenomenon and certainly not to the aviation industry. With the sector being steeped in technology and complex procedures, AI has propped up operations for years. However, machine learning [VT2] – a method that enables AI to learn from data and improve automatically through experience – is developing exponentially and the excitement surrounding it may drive further implementation of the technology in aviation.
“There is a lot of hype, which certainly helps us – this is the next internet,” said Assaia CEO Christiaan Hen. The company, founded in 2017, uses AI-driven computer vision and video analytics to provide real-time monitoring and analysis of aircraft turnaround operations. In addition, other modules allow for improved on-time departures, and emissions controls.
Hen continued: “When we started, there was more scepticism… Over the years, we’ve shown again and again that computer vision actually works and creates value. The industry’s stance towards a solution like ours has really changed, and we see many airports fully skipping any trial phase and moving directly towards full implementation.”
Aerogility, which launched in 2009, provides predictive analysis through its AI tools. The company initially supported clients such as Lockheed Martin before eventually assisting airlines in the analysis of their systems such as engines, landing gear, and airframes. The company is able to create a virtual representation or replica of a company’s assets, operations or systems, designed to simulate their real-world counterparts – a digital twin. The AI-powered technologies provide ‘what-if’ scenarios to airlines, allowing them to support decision making.
Aerogility head of AI Simon Miles said: “More broadly, Aerogility does planning and forecasting amongst other critical things. For airlines in particular, it is about finding the optimal plan for how you’re going to maintain your aircraft.”
Aerogility completes tasks within minutes that could take hours by human hand. “If you’ve got 300 aircraft all with a C-check every two years, and you want to do a 10-year forecast in Microsoft Excel – changing all of the cells from seven days to five days is one hell of a task,” explained Phil Cole, business manager for civil aviation at Aerogility. “That would take you less than five minutes with Aerogility.”
Cole said a “large percentage” of its airline customers are low-cost carriers with its first being easyJet in 2017. “easyJet has in excess of 340 aircraft in their fleet,” he continued. “You wouldn't be able to plan maintenance events for a huge number of assets on an Excel spreadsheet.”
easyJet, clearly recognising the benefits of AI, opened the doors to its new AI-equipped integrated control centre (ICC) in Luton in May 2024, to manage its daily flight programme. At the heart of this centre is its new in-house generative AI tool, Jetstream, which contains over 3,000 pages from eight of the airline’s operational manuals.
“Providing our people with generative AI solutions at their fingertips helps to speed up decision making to solve operational issues as they occur,” said easyJet director of network control Gill Baudot.
The key aspect of generative AI is its ability to create something new – text, images, music, or videos – from extracted data.
“On a relative scale, it is generative AI that is new and has the potential to disrupt the status quo,” International Data Centre (IDC) said in its October 2024 whitepaper on AI.
However, as generative AI programme ChatGPT explained itself: “While generative AI can mimic creativity, it does not possess true understanding or reasoning, keeping it within the bounds of narrow AI rather than general AI.”
General AI would possess almost human-like cognitive skills and is still believed a distant achievement, though recent developments have fuelled debate over whether the concept may be closer than initially thought. The technology, though, is still far, far away from the science fiction concept of super AI – surpassing human intelligence.
With greater computational power and labelled data availability, the training of generative AI has become not only more efficient but cost effective.
“Business leaders have certainly paid attention,” said Stanford University in its 2024 artificial intelligence index report. “Global private investment in generative AI skyrocketed, increasing from roughly $3bn in 2022 to $23bn in 2023… Nearly 80% of Fortune 500 earnings calls mentioned AI [in 2023], more than ever before.”
At the JP Morgan Industrials Conference on March 11, 2025, JetBlue CEO Joanna Geraghty said the company was looking to invest more in AI-powered predictive maintenance solutions to manage gates, but also to roll out AI solutions to drive “self-service” for its crew members and call centre, as well as for customers.
American Airlines noted in January 2025 that AI will “present even more opportunities” to improve customer experience, as well as running more efficient operations, though it would be taking on a “central approach” and is establishing a governance framework for its usage.
Delta Air Lines CEO Ed Bastian said in its third quarter 2024 results: “There’s no question that there are some really interesting applications to drive better predictive modelling and opportunities.” At the start of 2025, Delta launched its generative AI assistant Delta Concierge, signalling its venture into AI.
These are only a few of the countless airlines expressing interest or venturing into AI.
On the OEM side, Boeing chief technology officer Todd Citron said during a panel discussion in 2023 that the company is using generative AI to optimise aircraft design.
“When a machine does it – because it can put more complexity into its electronic brain than one human can – it can optimise over a broader space,” said Citron.
He said that these AI-optimised aerostructures “almost look like an alien spaceship” because they break away from the human tendency to follow ingrained patterns when designing something. Instead, it uses aerodynamic and engineering data to craft the design, which can be fed into new human-led designs.
Airbus expressed its own interest in generative AI in May 2024. “While it is unlikely that GenAI will be able to design future Airbus products from scratch, its ability to assist humans by enabling them to better manage complex and technical documents is proving promising,” said Airbus head of AI and advanced analytics Fabrice Valentin.
These evolving use cases mark a considerable shift from AI’s previous limitations, with the ‘intelligence’ of these technologies rapidly evolving far beyond what perhaps famed mathematician and computer scientist Alan Turing – who conceived the earliest benchmark for AI – could have imagined.
Explain the ‘Turing Test’, a question read in this reporter’s computing exam in May 2014. The test – originally coined ‘the imitation game’ – was created by Turing in 1950; a tester will have separate text-based conversations with a human and a machine – the identity of both remaining anonymous. The person would then evaluate and determine the identity behind both conversations. If the judge could not properly distinguish between the two, the machine passed the test.
This reporter was, at the time, correct in saying that no machine had passed the test. However, one month later, the Turing Test would be passed for the first time by a computer programme mimicking a 13-year-old boy – essentially using the more limited knowledge of youth to shroud its constrained algorithmic make-up.
“A decade ago, the world’s best AI systems were incapable of classifying images at a human level,” Stanford’s report read. “They could not understand language, struggled with visual reasoning, and flunked the most basic reading comprehension tests.”
Hen said: “Companies had tried for a very long time to make computers understand images. You could programme a computer to understand what a cat looks like by writing a lot of rules and then the computer can detect it. But if you get the same cat and just turn the head the other way, the computer had no idea what it is because it doesn’t fit into those pre-written rules.”
This is where AI comes in – able to continuously absorb data to learn these variations. Computer vision is a subfield of AI that helps software understand things like images and videos by recognising patterns.
“It’s like raising a child,” continued Hen. “A young child might just learn what a cat is and mistake a small dog for a different type of cat because they don’t have the underlying knowledge. And that is what all these AI companies are after: data.”
Aerogility added that it works with its airline customers to ensure the data it extracts is “true and accurate” and can require some tidying up to ensure this.
“We can point out inconsistencies in the data too,” said Miles. “We had a customer, which wasn’t an airline, but they provided data that said the same engine was installed on two different aircraft at the same time. They hadn’t realised that’s what their database said.”
This example also highlights the fact that a business can’t simply throw data at an AI programme and run with the solution. There remains a human element to these processes.
AI frees up the industry’s workforce – the technology acting as an additional arm rather than a digital replacement. As company’s operations and data scales up, the need for such technology amplifies.
“There isn’t an alternative,” explained Miles. “You have to choose a clever solution to solve these problems. Otherwise, you’re going to be struggling and doing things in a sub-optimal, inefficient way. The appetite for AI is here simply because there is a need for it.”
While Aerogility and Assaia have had time to mature their systems, generative AI is still in its more experimental stage.
“Generative AI is immature in many cases,” explained Miles. “People are still working out how to use it in a safe and reliable way.”
When using Qatar’s AI booking system, the ever-smiling Sama will provide a disclaimer that she is “still finding [her] wings”.
Qatar said that Sama “interacts naturally” with passengers and is “emotionally aware”, continuously learning from these interactions. Upon testing, Sama mostly repeated back emotions – steering far clear of more ‘turbulent’ ones – before presenting various destinations from Qatar’s network.
Alton Aviation Consultancy director Joshua Ng said these limitations may turn off customers – particularly in “emotionally-charged situations” such as travel disruptions, thus preferring “the human touch”. However, United’s AI tool allows for automatic rebooking through its application rather than waiting in line at a service desk. These agents are then freed up to properly manage more complex queries, again harkening back to the balance of human and machine.
This reporter then tested whether Sama would give a complementary ticket or supply discounts upon request to no avail, instead promising to find the “best fares available”.
This was not the case for Air Canada, however, which was held liable in February 2024 for its chatbot giving incorrect advice. A customer had been told by the machine he was eligible for a refund. The airline tried to argue that the machine was responsible for its own actions, which was ultimately rejected.
“That’s a very good example of where AI has gone wrong,” said Watson Farley & Williams partner Alan Polivnick. “For the most part, the law will hold the airline accountable. If I bought a ticket from Air Canada, I have a contract with them. The fact they outsourced the customer service to a third party, doesn’t change the fact I’ve paid Air Canada for my ticket.”
Ng said the Air Canada case, coupled with privacy and security concerns, highlighted the limitations of this technology. He added: “These shortcomings have the downside of making airlines a lot more cautious when rolling out such AI chatbots, delaying any full-scale implementation to the longer term.”
Air Canada’s rogue AI also underpinned the grey area around regulation on this technology.
“We don’t actually have an internationally accepted definition of what AI is in the aviation sector,” commented Polivnick.
In November 2024, the UK Civil Aviation Authority (CAA) unveiled its new ‘AI strategy’, which outlines an “agile approach to regulation”.
The regulator aims to address the challenges that AI could present and marks a progressive step for regulation.
“They’ve also been looking at how to balance the machine learning automation of AI with the need for constant verification and audit,” added Polivnick.
Hypothetically, if an AI machine recommends to a doctor to provide an X amount of Y medicine. The machine has miscalculated and given the wrong advice. Who is held liable? The doctor, the programmer behind the AI, or the AI itself?
Aviation similarly holds safety to the highest regard and will ultimately limit how much processes can be offset to automation. That is why Aerogility and Assaia both highlighted that their AI should be taken as recommendations – an additional safety net and support to decision making.
“Manpower is still required to manage the AI systems, and to act as a failsafe,” Ng said. He added that, with the limited manpower in aviation, AI could rectify or at least support the labour shortage.
The CAA’s strategy also pointed to the range of opportunities that AI can present. The efficiency of the technology could transform air traffic control’s proficiency with real-time data and predictive modelling, as well as supporting sustainability goals with fuel efficient routes. Polivnick added that predictive AI could be used further to support feedstock procurement for sustainable aviation fuel (SAF), with it able to track feedstock levels based on weather patterns, as well as boost the efficiency of its refinement.
Aerogility said that it is drawing from various departments in the aviation sector that interact with each other the most into one digital twin. The company is also adding its fleet optimisation strategy module to support airlines further. “We want to expand that core and expand the functionality,” said Miles.
Looking back at the past decade – even the last couple of years – the evolution of AI has developed at rapid speed. Whilst supporting the aviation sector in the background for many years, the industry looks to capitalise on the opportunities the advancements present. Where will the industry be 10 years from now and to what level will operations be married with AI? Will we see AI regulations akin to Asimov’s fictional rules of robotics? As AI takes off in the present, perhaps a decade from now, Sama and her similar counterparts will have become highly educated AI companions – key assistants to travelling the skies.