When ChatGPT emerged on the scene in late 2022, it put artificial intelligence, or AI, “in the zeitgeist.” While still in its infancy in the aviation industry, AI is now at the forefront of many people’s minds, said Rob Mather, v-p of aerospace and defense industries at software intelligence corporation IFS.
Less is known about what AI is and what it can do, but many agree it has the power to revolutionize how the industry does business.
“This is potentially the biggest technology transformation that we’ve ever seen, bigger than the computer itself,” said Greg Jarrett, CEO of aviation business operations systems provider Stack.aero.
While some discussion has centered on aircraft applications, AI’s potential runs the gamut for the industry with a range of possibilities on the ground from operating software, supply chain management, manufacturing efficiencies, and safety-enhancing technologies to charter management, human resources management, and maintenance diagnostics, among many others.
However, similar to the mantra for the evolution of electric vehicles, AI must take a crawl, walk, run approach—even though it is evolving so quickly that companies can barely keep up. This is because it must be proven secure and accurate, and—for now—it requires a human interface before its full potential can be unleashed.
AI is not new, said Mather, whose company is a $1 billion firm that provides cloud-based enterprise software for companies globally across industries, including aerospace and defense. Companies have used and still use early forms and predecessors. And more broad-based applications, such as Siri or Alexa, have their foundations in AI, consultancy McKinsey & Company pointed out.
Stepping into AI was a natural shift for IFS, which has focused on software solutions for decades. “We’re a legacy system,” Mather said. But now, “From IFS’ view, the future is all about AI,” he said, adding that his company is developing AI capabilities that build on the company’s core software offerings but also is looking to acquire other specialists in the field to expand its breadth of capabilities.
What is AI?
According to McKinsey: “AI is a machine’s ability to perform the cognitive functions we associate with human minds, such as perceiving, reasoning, learning, interacting with an environment, problem-solving, and even exercising creativity.”
For Mather, AI transcends from a typical algorithm to the point when the software is essentially capable of learning “so that it can adapt and change without having to recode it.”
Typically, coders must input the information and the potential outcomes. This meant developers had to continuously build out scenarios to expand outcomes. But with AI, instead of coding the expected outcomes, the technology can bring the outcomes to the user based on available information.
“That’s the fundamental difference in technology up until today,” Jarrett said. “You would typically know the exact outcome that you are looking for and the objective would be for the system to achieve that outcome. AI is flipping that around.”
Mather added: “With artificial intelligence, you want to get to a point where we set up the structure and then it says, ‘Oh, I know what should happen here.' It should do this without you having to code it up front and foreseeing every possible scenario that you could ever come across.”
Keeping up with the scenarios upfront “takes ages and it takes a large investment, which means cost,” Mather continued. “If you set up a structure where something can figure out the new cases, you aren’t absorbing all of that upfront labor to develop this incredibly complex algorithm. At that point, you shifted the burden…and companies could do a lot of these things that they hadn’t before.”
He added that this “levels the playing field” for organizations that didn’t have the resources to invest in costly management software systems. “It’s almost like AI has the potential for a democratization of capabilities.”
AI in Aviation
AI is not widely adopted yet in aviation—at least on a large scale—although many companies are exploring options or discussing it. “It’s interesting,” remarked Mather. “I meet organizations that are all in on AI and similar types of organizations that are really fear-based relative to AI.”
IFS has started to bring AI to large defense companies, airlines, MROs, and even some of the largest business aviation operations, such as NetJets.
Joe Sambiase, director of maintenance and airworthiness for the General Aviation Manufacturers Association (GAMA), said most of the membership has not yet indicated their use of it on any scale, although there are discussions around it. And he does see potential applications for organizations such as the FAA.
Jarrett’s company, meanwhile, is beta testing it and working behind the scenes with one potential client, but he said it may be 2025 before Stack.aero is ready to roll it out.
Aviation expense management platform MySky has launched AI-based programs and claims that customers are losing thousands for not using such a technology.
On the FBO front, Signature Flight Support sees substantial possibilities: “Signature has long employed traditional machine learning techniques and is excited about the possibilities of artificial intelligence going forward,” the company stated.
And while it is still conceptual for many companies, Mather believes that this is going to change, and likely rapidly. For a long time, he explained, organizations using AI needed to build out the structure or spend “a whole bunch of time training a learning model…or investing time in sorting and labeling your data to have it consumed by artificial intelligence in an effective manner,” he said. “There have been solutions in place around this for a long time, but they’ve been usually bespoke and pretty expensive.”
But AI is changing, he said. “We’re starting to get to a place where those solutions are much more available and much more cost-effective.”
The Applications
IFS is developing AI on multiple fronts, and Mather sees possibilities on many more. The key is AI’s potential to manage big data. Over the past decade or so, aviation companies have embarked on amassing large quantities of data, from health monitoring and flight operations to charter management and client databases. That doesn’t get into the vast amounts of data at the regulatory agencies.
“There were big data activities, but you can only get so far and you had to be able to hire data scientists, which are scarce resources and cost a lot,” he said.
AI can help figure out the salient data that bring efficiencies, safety of flight, and lower costs, he said. As an example, he pointed to maintenance diagnostics such as anomaly detection. In traditional models, a programmer would input what sensors should read and what faults they should find. Then the programmer would input what the faults may mean. This may be time-consuming and require extensive research once the sensors find those faults. With AI, “you can take a live sensor feed and it can tell you when something is off right away instead of having to go into the data analytics after the fact,” Mather said.
Building on that, he added, is something called “unsupervised learning models,” (which he called a terrible name because “letting AI be unsupervised is a concept that is super scary to me”).
But what "unsupervised" actually means in the context of a learning model, Mather added, is that a person doesn’t need to tell the AI what they are looking at. “Basically, you plug the AI in, and it figures itself out. That works really well in the domain of anomaly detection because previously you would have to take all these sensor feeds and say, ‘Okay, this data means this.’”
Now, AI interacts based on what “normal” looks like and determines whether something is normal. “You are able to do that in real time.”
AI, he further said, has “almost untapped potential in predictive maintenance.” Again, predictive maintenance has been around for some time, he noted, but “has been slow to penetrate broadly within the industry.” A few big players have led the charge—those that can afford it.
“AI dramatically lowers the barrier to entry to being able to utilize predictive maintenance,” Mather said. “You can not only just do diagnostics, but you can then do the predictions on what’s going to happen in the future. So not just what’s wrong right now, but, 'Now I’m trending in the wrong direction. That means something is going to happen down the road. I’ve seen this pattern before.'”
Along that vein, he continued, is flight operations patterns, with the ability to provide insight on aircraft and even pilot performance, although he cautioned that the latter comes with privacy concerns. But it can also have sustainable applications providing insight on reducing an aircraft’s carbon footprint—“Is it better to go over than to go around? Will we save X amount of fuel by doing that?”
Also, Mather added, it has safety applications such as the potential for providing real-time information on dealing with storm reports and the most efficient way to handle it.
Then there is “a whole other conversation” on how it can be used to improve manufacturing processes and the supply chain. From the start, manufacturers can find the bottlenecks and manage them.
AI can help optimize the positioning of inventory to make sure a manufacturer or an airline can meet demands without incurring delays or spending substantial money on an emergency AOG procurement, he said. It further can point to how much inventory an organization should have on hand.
Further, it can be used on the procurement side, including vendor evaluation. “Just like you’re performing real-time evaluation on sensor feeds, you can do real-time performance evaluation on vendors,” he added.
Jarrett also believes that predictive maintenance holds one of the biggest potentials for AI. For Stack.aero, though, he is exploring possibilities surrounding how it can leverage its business operations platforms to build on customer communications and relations.
Stack has been involved in an AI pilot with one of its clients. “We’re seeing AI just help to develop the customer relationship,” he said. “It’s helping us to generate the content for those conservations, and it's making the customer feel appreciated rather than as just another client who’s paying money. You can have a genuine authentic conversation with a customer.”
It could be about a company’s operational patterns, he agreed, but more than that, “it’s also about the things that the company is doing. General things that are public information.”
For instance, AI can give insight when a company is going through the process of change. As a hypothetical, he said a big enterprise, such as Coca-Cola, may be recruiting for certain roles. AI can look at that recruitment and find ads on the internet. If the client is involved with Coca-Cola or part of it, that may lead to conversations on how that may affect them.
More specific to aviation, it can look up trip histories involving destinations, ranges of aircraft, and occupancies, and predict travel in the future based on those patterns.
“We can start a conversation about ‘We think you’re going to be flying this way in over the next 12 months. What do you think about that?’” Jarrett said, reiterating, “Predictive analytics is something it’s very good at.”
And while it may be used for flight optimization, Jarrett believes that predictive analysis is where its best advantage is. “I think machine learning has a much greater potential for optimization, and AI is much more about analysis of text and unstructured data,” he said. “AI can very easily look at unstructured data and turn it into something that appears valuable. Machine learning has to have structured data, and you have to tell it what the outcome is that you’re looking to achieve.”
Signature Flight Support also sees opportunities on the customer front. “Like most companies, we believe technology can improve the customer experience and generative AI presents many innovative possibilities,” the company said. “From better understanding our customers’ needs to design better solutions to meet those needs, there is extraordinary potential for how we can leverage AI.”
Sambiase, meanwhile, points to the potential at regulatory agencies, citing service difficulty reports as an example. “These were always issued via paper or email. It’s really hard to pull a trend off of all this data in an efficient way,” he said. With AI, “you can do this within 30 seconds.”
Explaining the trend analysis further, he pointed to a tire issue that surfaced at an operator he worked at involving retreads. Most of the retreads were fine, but in some cases, the adhesive would come apart. After an examination of the history, he realized that the adhesives encountered problems when it was hot and the tires did not have the proper pressure. This took some time to trace, he said. “AI could have done that within 30 seconds.”
MySky, meanwhile, said its AI-powered approach eliminates costly, labor-intensive processes by pulling together back-office operations that are typically run separately, such as accounting, reporting, procurement, and business intelligence. Based on workforce and other costs, MySky suggests that charter operators could be losing up to $4,000 per month per aircraft by using the traditional, disparate management approaches.
Navigating through the Concerns
However, while the technologies are maturing, the risks involved still concern companies. Jarrett said privacy issues are a primary reason why it would be 2025 before Stack.aero is ready to roll it out.
“We’re very much in an experimental phase. We are running experiments as the AI system ecosystem evolves because it’s moving so quickly,” he said. “We need to make sure that the outcomes we deliver to our customers are beneficial to them.”
Stack is getting a lot of questions about AI, he said, but it’s not whether the company can implement it but when it can safely implement it. “The biggest concern among them is how these companies keep their data private in all of these AI engines,” he said. “It’s not about, ‘Hey, we want it now. Everyone’s doing this.’ People are much more cautious. They’re very protective of their customers’ information.”
Sambiase also cited a need to be able to build strong cybersecurity protections and ensure the data assumptions are correct. “If we assume that AI is reusing and recycling information that already exists, AI could certainly produce an unintended result if somebody puts in incorrect data,” he said. “AI is just going to look at that data and assume that it’s correct and produce a result based on it.”
Mather added that in general, the aerospace industry is particularly cautious and “for good reason. It’s a very safety conscious [industry] and the costs of failure are way too high.” Aviation is on the leading edge of innovation, he said, but “in other ways, we lagged in a lot of cases around adoption. It’s an interesting mix, and I would say that it manifests here.”
He pointed to maintenance. “We look at the core of how that’s set up as an industry, the main tenet is based around the idea of a human being who is trained and certified taking an action and being responsible for that action,” Mather said. “AI applications that put that principle in danger, I don’t see having adoption in the near term. There’s lots of automation that could be done through AI that we shouldn’t do right now.”
Perhaps in the future, he continued, when AI models are better understood and have a history to back them up, AI can reach its full potential.
But for near-term adoption, Mather pointed to “low-hanging fruits” that can be used now—with human interface.
Big data analysis and predictive actions are among the areas that can be implemented near-term, he said. “You’re not changing the regulations. You’re not changing the maintenance program.” AI may identify the problem but the technician will validate and execute it.
The regulations and the human-machine interface are tightly intertwined, he added. The person “needs to be the one making the decisions.”
Further, there are ways to enhance the data for AI, he said. Mather pointed to the idea of retrieval augmentation generation that involves relying on a specific data repository rather than the large language models used by ChatGPT, for instance. “There are challenges around large language models, and training them is expensive. Keeping them up to date is expensive,” Mather said, adding that there needs to be a level of caution around them.
With a narrower, controlled data repository, AI retrieves information from a specific source of data. “It could be a pool of data like your own reliability data, performance data, or all of your manuals.”
The Jobs Fears
While much discussion generally has focused on the possibility of AI replacing jobs, at least for the time being, most involved with it don’t see that happening. “For right now, we have to maintain the core tenet [of a person making the ultimate decisions],” he said. “There’s lots of applications that work around the periphery that make the human being at the center of that more efficient as opposed to replacing them.”
Again, pointing to maintenance, he noted, “Any organization that employs technicians can benefit from those technicians being more efficient.” This is especially true as the maintenance field encounters a technician shortage.
Jarrett agreed. Looking out 10, 20, or even 50 years, responsibility may shift from the human to the AI machine. But for now, “who takes legal responsibility for a decision?”
Further, he said, AI has the potential to enhance rather than detract from the workplace. “We see it just helping to relieve some redundancy and hopefully allow people to be happier in their work because they’re doing more interesting work. They’re doing the valuable work that humans do, while AI is doing the repeat administration work.”
Further, Jarrett continued, “This conversation happens with every phase of technology—about some sort of new technology taking all my jobs. I don’t see it taking away jobs. I see it changing the employment landscape and allowing people to feel more useful in the tasks they are doing at work.”
The Future
Despite the near-term reservations, Jarrett emphasized its transformative possibilities long term “Attitudes will change as people learn more and people get more experience and more education with what’s possible with AI,” he said. “I think things will change rapidly. Attitudes are going to change rapidly over the next 12 months, three years, and 50 years.
Sambiase further noted that if the industry isn’t heavily using it now, he expects it to become a staple going forward. “It offers us another opportunity to produce some measurable gains in safety. It’s a new thing that we haven’t used to help improve safety,” he said. “That’s always going to be the objective. I do suspect that will be a major contributor to some safety improvements going forward.”