AI in Aviation: A Blessing or a Curse?

Artificial Intelligence or AI… You’ve probably been bombarded with this term over the last few years.

There are people who can’t wait to see it implemented everywhere, and others who are very apprehensive about what it could mean for the world 🌍

AI in aviation is certainly a spicy topic. While it can offer many benefits, it definitely comes with its own set of threats as well.

AI is starting to be researched for lots of different areas of aviation. Things like managing air traffic, optimising airline fuel consumption, airway design, and making the use of airways more efficient.

The trend is only showing more and more utilisation of AI, with major aircraft manufacturers aiming to implement AI on the flight deck as well! It clearly has the potential to change the way we fly modern aircraft.

The million dollar question is: is this a good or a bad thing? 🔮

Like our content? Subscribe to Pilots Who Ask Why and be the first to get our best aviation insights, guides, news, and must-read stories. Join our crew today! 🚀

What is AI?

Artificial Intelligence (AI) is a program’s ability to do things that usually require human intelligence. It’s like teaching computers to think and learn on their own.

EASA uses this definition:

For instance, AI systems today can already understand speech and images, make predictions, solve problems, and do lots of other tasks with pretty insane accuracy. Stuff that us puny humans would do a lot less accurately, and a LOT slower.

So what does it consist of? Well, there are 5 key components of AI:

1) Machine learning. This is about creating algorithms and models that help computers learn and improve without it being explicitly programmed. Think of our own brains for instance. If we enter a new situation that we haven’t seen before, we can still use processes in our own mind to make sense of what we’re observing. AI is essentially trying to mimic this ability with machine learning.

2) Natural Language Processing: This is the ability to interpet what we as humans are saying or typing. You might be familiar with tools like Chat-GPT. It can interpret what you’re asking it to do, sometimes with an almost scary amount of accuracy. This is very different to a Google search for instance. Google simply takes your keywords and tries to match it in a clever way with whatever is already on the internet.

3) Computer Vision : This is the AI’s ability to interpret not just speech and text, but also other types of information models like photos and videos. For instance, you could upload 5 hours worth of video to an AI model and ask you how many trees it has counted throughout the footage. Something that’s not only tedious for a human, but we’d all probably fail horribly at it as well, based on the time and concentration it takes. Using this technology, you might be able to imagine what this could be used for in aviation (more on this later).

4) Robotics: Combined with actual hardware, AI has the ability to independently function as a machine, within a limited set of variables (which is based on how you program it). It would require access to any hardware that needs operating, with the aim to be used for specific purposes. In an aircraft, this could be control surfaces or even engine control systems.

5) Expert Systems: This is where things get interesting. AI can be used in very specific areas to actually mimic human decision-making. No single AI is advanced enough (yet) to be used on a high level in any specific industry. However, AI can be trained, programmed, and guided to a point where it can actually deal with very dynamic but specific environments.

The thing that makes AI tricky to assess, is that it’s always evolving. Researchers are constantly finding new ways to make it even more advanced. It’s all developing at a ridiculously fast pace. So what’s relevant today, will be horribly outdated tomorrow. Keep this in mind with any article you read, including this one!

The Impact of AI on Aviation

AI will inevitably start impacting the aviation industry on a global scale, whether we’re ready for that or not. Companies like airlines will always pick the route of efficiency if the risks can be ‘managed’ and profits can be made. Of course, perspectives on this will be wildly different, but both regulators and companies will have to come to some sort of consensus, at some point.

There are a 3 main levels of how much influence AI has on any situation:

For aviation, it’s currently expected that AI will evolve from Level 1, where AI is there to assist humans – to potentially reach level 3, where AI makes most of the decisions, as time goes on and technology improves.

At level 3, AI decisions could either be overridable, or non-overridable by humans. If this will ever happen, and when, remains to be see though. It’s already very controversial and will require the trust of the general public.

The next question though, is what are the main ways that AI is starting to impact aviation? Let’s break it down:

Aircraft Design

Yep, Boeing, Airbus, and many others are already well on their way with integrating AI into lots of projects. Airbus is specifically focussing on 6 key aspects:

• Knowledge extraction (efficiently scanning and gain valuable information quickly for complex documents
• Computer vision (as we said before, scanning non-text based sources)
• Anomaly detection (find patterns in data that could indicate a failure)
• Conversational assistance (systems that can interact with humans using speech)
• Decision making (aiding pilots during complex problems)

And finally: Autonomous flight (this one speaks for itself..)

On top of this, a lot has already been developed. Projects such as autonomous drones, and air-to-air refuelling equipment that is fully operated by AI.

Boeing has said it is already using AI to sift through lots of data to identify hazards within the airline industry and the aircraft themselves. This is something that could yield results that humans aren’t capable of finding. They already have the MQ28 project that features an AI powered drone.

Air Traffic Management

Let’s have a look in which ways AI can be used to optimise overall air traffic management:

Scheduling and Dispatching: Optimising air traffic management by analysing lots of different types of data, such as weather conditions, flight schedules, and airspace capacity, even based on the time of day and year! You could even let it process this information in real-time to provide more accurate and efficient routing suggestions.

Baggage handling: AI algorithms can optimise baggage handling operations by tracking and managing baggage flows more efficiently. It can also help identify potential bottlenecks before they happen, automate sorting processes, and improve baggage routing accuracy, reducing the likelihood of misplaced or delayed luggage.

Fuel Optimisation: Currently it takes lots of people working together to optimise fuel consumption. AI might not be able to replace these entire teams, but it can definitely assist them. By looking at data on weather conditions, flight routes, and aircraft performance, AI algorithms can suggest more fuel-efficient flight paths, all in real time! This could not only reduce costs, but also contribute to environmental sustainability.

Noise Management: AI can assist in managing airport noise levels. Noise policies are usually constructed in a way that allows a certain amount of cumulative noise per time per location.

By looking at flight data and noise measurements, AI algorithms can help optimise flight paths, runway usage, and aircraft operations to minimize noise impact on local communities.

Drones

We are already in an environment where companies are heavily focussed on AI powered UAS, like the MQ28 that we mentioned earlier. This will only grow, as more companies get in on this technology.

Airbus is on a similar trajectory with projects that aim to improve their AI powered drones as well.

Aircraft Maintenance

Yes, even maintenance can massively benefit from AI. Let’s have a look at the main ways it could help make our lives easier:

Predictive Maintenance: We can make AI look at data so it can predict when components can be expected to fail, and therefore when to replace or repair them.

Condition Monitoring: AI can be used to continuously monitor the condition of critical components and systems in real-time, especially now that health monitoring is becoming more and more mainstream, even in the helicopter industry.

Fault Diagnosis: AI can assist in diagnosing complex aircraft faults by analysing historical data and comparing it to the current state of the aircraft.

Data-driven Decision Making: Basically identifying trends, patterns, and maintenance practices based on huge amounts of data that us humans would need multiple years for.

Inventory Management: AI can make inventory management more efficient by predicting spare part requirements based on historical data, maintenance schedules, and usage patterns.

Cybersecurity

When we talk about cybersecurity in the aviation industry, we can usually spilt it into 3 main elements:

1. The system that can be digitally exploited
2. The potential threats that could cause harm to that system
3. The tools that help us mitigate those threats

The good and bad thing about AI, is that it can enhance any of these 3.

Risk Management

AI can help identify where the highest risks are within the current aviation industry. Just imagine throwing 234598 binders full of info on some poor risk managers lap and telling him to come up with the 3 highest risks.

AI can be of huge assistance here. It can scan huge documents relatively easily and spit out summaries. Whether those summaries then get interpreted by humans or not, it will provide a massive benefit to risk management on an industry scale.

Sustainability

We mentioned sustainability earlier, but it’s not one to brush over. All of the use-cases mentioned above all indirectly contribute to a more sustainable aviation industry. Airway optimisation, noise optimisation, determining whether or not winglets or other aircraft mods are worth pursuing, etc. There are countless ways in which AI can help us make the industry more sustainable.

How Trustworthy is AI?

Trustworthy is probably the wrong word. When we as humans talk about how trustworthy a person is, we usually refer to someone’s real or hidden intentions, and how they differ to what they appear like, right?

Well, with AI, things get complicated. Its trustworthiness depends on so many things, and will still be graded differently depending on who you ask. And this is probably a good thing. It will require multiple people with different perspectives to come to a conclusion, depending on the operation.

In general though, it requires a certain amount of items to be considered ‘trustworthy’. There are 7 main items that will need to be addressed before we can even begin to trust any form of AI:

• Human Agency: Someone needs to be in charge of the AI
• Technical Robustness: AI will need to consistently prove that its reliable without glitches or safety concerns
• Privacy: AI is data-hungry. AI should only be fed data that’s actually useful to the end goal, and AI should not get access to user data without user consent.
• Transparency: Operators and users should be able to see and understand how the AI operates, and be subject to inspections.
• Fairness: Bias should be eliminated as much as possible.
• Environmental Well-being: The impact of AI on the environment should at the very least, be considered and made transparent.
• Accountability: It should be clear who is responsible for AI decisions.

What about you? Would you be able to trust AI if these boxes are ticked? Let us know!

Can AI Act as the Pilot?

Well, at the moment it depends on who you ask. There’s understandably a lot of resistance from people inside and outside the industry when it comes to adopting AI in the cockpit, whether as an aid or a pilot replacement.

The problem is, manufacturers ARE moving towards this, whether we as pilots like it or not. Of course the regulators and general public will decide in the end whether or not this is considered acceptable. If one of the major airlines decides to go through with this development and manages to reduce ticket costs or increase profitability with the implementation of AI, one of the following things will happen:

• The general public will refuse to fly with the airline, as they don’t feel safe without pilots in the front, other airlines will be discouraged to pursue the same
• The general public will get used to having AI in charge of things, including planes or helicopters, resulting in airlines becoming more profitable for integrating AI, this will encourage other companies to do the same, which will shift the industry towards the use of AI.

Of course, for this to be even a potential reality, regulators and manufacturers will still have to come to some sort of agreement on standards first. But in the end, you as a consumer, not necessarily as a pilot, will influence the eventual reality.

The biggest challenge companies like Airbus and Boeing will have to tackle is creating a system that can actually deal with uncertainty, complexity, and ambiguity, like so many things are in aviation. It’s not all just following flowcharts and SOP’s.

On top of this, regulators have a mountain of work ahead of them for this to ever become a reality, but more on this later.

The Threats of AI in Aviation

Right, so we’ve listed plenty of benefits so far. But we’re not biased towards AI (we promise). So what about the negatives? The stuff that makes this a slippery slope that can be hard to navigate? Let’s have a look:

1) Safety Concerns: The classic ‘what if it fails’ question. Yes, humans fail too. But we are pretty experienced with acknowledging our shortcomings and designing systems to manage this. How can this be done for AI? Trust could easily be misplaced.

2) Cybersecurity Risks: AI relies on data. If it doesn’t have access to data, it won’t be able to function anywhere close to the performance humans provide. Giving 1 system access to all this data comes with security risks. How this will be addressed is a whole topic on its own.

3) Lack of Human Judgment: Humans are much better at assessing situations that are less defined. Yeah, sure you could ask the AI to interpret a METAR or TAF, but aviation is so much bigger than that. How can we ever fill the gaps here?

4) Job Displacement: This might be no-brainer, but AI is of course a giant risk for people’s jobs in lots of different ways. It’s already happening, from fast food to warehouses: AI is slowly taking over. How can we manage this in our industry?

5) Ethical Considerations: As pilots we do sometimes have to handle ethical considerations, even during emergencies. AI may be programmed to prioritise certain actions, which raises questions about what the right thing to do is in certain situations. Setting ethical guidelines and making sure AI decisions are transparent and accountable is a difficult challenge that we are not sure on how to settle yet.

6) Privacy Issues: Like we said earlier, AI systems need a lot of data to work properly, and some of that data might be personal information about passengers and others people involved in aviation. How can you protect people’s privacy, if that same data is sometimes crucial for the AI to function properly?

7) Regulatory and Legal Challenges: Technology has always been way ahead of regulation across the globe. AI is no exception, and in fact is probably even further ahead. Ask yourself how many people you know truly understand what AI is, how it functions, and what the associated risks are! Now imagine politicians that are probably twice our age who might still struggle with opening a PDF, see the problem yet?

How can AI be Regulated?

Regulating AI will become a giant challenge for governments and organisations like ICAO, EASA, the FAA, and CASA.

AI regulation involves creating policies and guidelines to ensure responsible and ethical AI development and deployment. Here are some key ways AI can be regulated:

1. Legislation: What laws need to come into place that don’t exist yet?
2. Ethical Guidelines: What ethical practices should AI adopt?
3. Standards and Certification: What industry standards will be the baseline for certification?
4. Oversight and Compliance: How can we as humans keep on eye on what’s happening and call things off when it gets out of hand?
5. Transparency and Accountability: How will AI be accountable for mistakes or accidents?
6. Impact Assessments: How can companies assess the impact of AI before implementing it?
7. Public Input: How will the public ‘feel’ about interacting with AI without a sense of control over the situation?

EASA has already published an AI roadmap where they highlight how they are going to deal with this. Let’s take a look at what their main focus points are for regulating AI:

• Risk assessment: AI systems will be categorised based on their risk level, and high-risk systems will require all sorts of assessments before being allowed utilisation.
• Ban on unacceptable AI: This regulation prohibits certain AI practices that are considered to be unacceptable or against EU values. Think of AI systems that can manipulate human behaviour for instance.
• Transparency and traceability: This requires AI systems to be designed in a way that ensures transparency and traceability, so users can understand how decisions are made and what data is used to train the system.
• Data governance: This aims to ensure that data used to train AI systems is of high quality, and unbiased. In addition, it also focusses on making sure that privacy and data protection rules are respected.
• Human oversight: This regulation requires AI systems to have appropriate human oversight and control. Especially in high-risk situations such as in healthcare or aviation.

While this is the way EASA looks at it, other regulators might have completely different views. We will hopefully find out very soon!

Conclusion

AI presents a lot of promising benefits, as well as a mountain of challenges and risks that will need to be addressed as soon as possible. The technology is developing at a really fast pace. Regulators will need to stay ahead of the game by starting to think about how this technology can safely be implemented, and assess whether it’s a good idea in the first place.