Artificial Intelligence alongside pilots, controllers, airport operators, and passengers: how can these new partnerships function effectively? The HAIKU project, coordinated by Deep Blue, seeks to find out.
Paving the Way for a Synergy: Humans and Digital Assistants Together
Digital assistants – such as Apple’s Siri, Amazon’s Alexa, Microsoft’s Cortana, and Google Assistant – are setting the stage to enter the aviation industry, and their potential applications are varied: in the cockpit or control towers assisting pilots and air traffic controllers, at airports aiding those ensuring runway safety, or assisting transiting passengers.
The concept of a digital assistant goes beyond tools based on machine learning algorithms that provide data and information to the human operator. Instead, it’s more like a colleague that interacts and “converses” with its human counterpart. This introduces the idea of Human-Artificial Intelligence teaming, which raises questions: What values should underpin this partnership: transparency, complementarity? Ongoing research will shape some of the features of Artificial Intelligence itself.
The HAIKU project
Funded under the Horizon Europe program of the European Union for research and innovation, the HAIKU project will develop and test digital assistant prototypes across various aviation domains. “The approach is human-centred: focusing on human values, motivations, skills, and limits in the ‘construction’ of Artificial Intelligence. In other words, we will work on AI solutions starting from the operators – the individuals who will use these tools,” explain Simone Pozzi and Vanessa Arrigoni, respectively CEO and Lead Consultant of Deep Blue, which coordinates the consortium of 15 European partners with Engineering being the only other Italian company involved. “This means that Aviation, Artificial Intelligence, and Human Factors, three distinct realities, must come together, discuss, and agree on what’s acceptable and desirable – an ambitious challenge, but also the project’s strength that embeds innovation within a social impact.
The case studies: what kind of partnership?
The project has six application cases: two involve pilots in cockpits; one concerns air traffic controllers in remote or digital control towers; one focuses on drone traffic managers (a role that doesn’t yet exist); one on operators overseeing taxiing and parking at aircraft gates; and one for transiting passengers.
In shaping digital assistants and their collaboration with human operators, Pozzi wonders, “What aspects of the collaborative relationship do we want to maintain?” Safety remains paramount. In every instance, the digital assistant must prioritise operational safety just as the human operator does. Depending on the task, other factors will vary.
In one case study, the digital assistant aids the pilot during sudden critical situations, such as lightning striking the aircraft during landing. In these circumstances, the pilot might “freeze,” a detrimental reaction that could hinder swift, clear decision-making. To assist its human partner, the digital assistant must first detect the pilot’s state (using sensors to measure physiological parameters like skin conductivity, heart rate, and respiratory rate) and then intervene, possibly using coaching techniques to help the pilot relax and regain mental clarity.
“In this specific case, shared awareness, or the ability to realise the state of other team members, is an indispensable value of the partnership,” the expert explains. Trustworthiness is another critical aspect: during emergencies, pilots relying on their digital assistant won’t have time to question the algorithm’s suggestions; they must trust it. Critical evaluation of the output will occur later, demonstrating that explainability or transparency of Artificial Intelligence isn’t always a priority when designing an algorithm.
The Impact of Digital Assistants on the Future of Aviation
Using the digital assistant’s support to the pilot as an example helps envision AI’s future role in aviation. The role AI will play is primarily based on its responsibilities. For instance, in the scenario where the digital assistant helps pilots to “recover” the aircraft in critical situations, the main issue is safety – a matter that aviation takes very seriously and could delay the introduction of the digital co-pilot. If the assistant merely aids pilots in rerouting, the primary concern shifts from safety to flight efficiency, fuel savings, and minimising passenger delays. Here, the importance of explainability resurfaces.
Explainability is also paramount for those managing air traffic, drones, and aircraft, as well as taxiing operations, for instance, managing vehicles in gates and flying strips. For all these cases, safety is the priority, even if the time for decision-making is not as “tight” – several minutes – as pilots in emergencies, allowing for a critical analysis of the algorithm’s suggestions.
One thing is certain: the algorithm will never make autonomous decisions or be free to “learn” anything it wants.
Managed Learning Process
When discussing machine learning, it’s often emphasised that Artificial Intelligence “evolves” as it learns new data: it constantly changes, an unpredictability that’s also its strength. “This approach isn’t feasible in aviation,” Pozzi clarifies. The learning isn’t online but uses pre-collected and selectively chosen data. The main reason for controlled learning conditions is safety and security, and specific cases must be managed individually, not standardised.
Within the aviation context, algorithms are designed to be ‘robust’ against non-conforming conditions. This means these circumstances must be excluded from the data given to the algorithm, or the algorithm itself must be trained not to react to them.
Personalised and predictive model
There might be some exceptions to an algorithm’s offline learning. For instance, in control towers, certain operators adopt a more “aggressive” approach, while others are more “conservative” in managing air traffic. Based on these individual characteristics, it could – ideally – be possible to develop personalised digital assistants that provide suggestions consistent with an air traffic controller’s working style. “This would assume faster data learning, perhaps in real-time, though still under controlled conditions,” adds Vanessa Arrigoni.
As per the developers’ intentions, the digital assistants in the control towers will also be customised according to the characteristics of the airport and its air traffic. This scenario is under analysis as one of the application cases for HAIKU, aiming to develop a smart assistant proficient in optimising landings and departures at the Alicante airport in Spain, based on the airport’s traffic data. “In this instance, the partnership revolves around the ability to gather specific information and present it in an already processed manner,” explains Arrigoni.
The capability to anticipate needs is what one might expect from a digital assistant for airport operators. “In the case concerning Luton airport in London, we are exploring Artificial Intelligence solutions capable of foreseeing potential critical situations, alerting the operator about risks, and advising them on managing aircraft traffic on runways safely,” says Pozzi. The hypothetical scenario involves sudden fog leading to a reduced airport capacity. Artificial Intelligence should process a series of historical data to provide solutions that avert potential accidents. “In this instance, the most intriguing aspect isn’t just discovering data correlation but understanding why AI identified a particular correlation, like between a parking gate and an incorrect aircraft manoeuvre, so the so-called explainability,” emphasises the expert.
The Evolution of Work Procedures
The added value of a project like HAIKU lies in exploring the “costs and benefits” of Artificial Intelligence in various aviation domains: different application cases, and diverse end users. “The project will continue until August 2025, and currently, those managing the various case studies are proceeding independently,” states Arrigoni. “Pilots are further along in defining the concept of Human-AI teaming, air traffic controllers and airport operators in the technical development of AI, and drones in the operational concept.”
The operational aspect is often overlooked when discussing Artificial Intelligence. Yet, concludes Arrigoni, “The technology will impact both training and work organisation. Tasks will change: in the future, some jobs currently performed by humans will be done by digital assistants, meaning roles will shift (likely the human role from operational to more strategic) and new skills will be required. This change will be reflected organizationally in procedures and the way of working. This facet too needs to be studied and resolved to optimise human-smart machine collaboration.”
