Al in aviation - reaping the benefits by navigating the challenges

The conversation at ERA's recent workshop, held in partnership with international law firm Clyde & Co, focused on the increasing impact of Artificial Intelligence (AI) and machine learning on aviation. In this issue of Regional International, Clyde & Co share some of the key takeaways from the workshop while also highlighting the opportunities and risks that need to be carefully navigated.

The launch of generative Al tools such as ChatGPT in late 2022 laid the groundwork for 2023 to become the year Al went mainstream. These easy-to-use tools, which produce human-like output in seconds, have captured our imagination and given us a taste of what an Al-enabled future may hold.

With the current buzz around Al, it's easy to think Al is new-but that's not the case. Al has existed in one form or another for years, while the term itself has been used since the 1950s. In fact, Al has been used within the aviation industry for decades. What is new is the speed at which we are now moving - both in terms of technological advancements and adoption.

As with all technologies, there are risks and compliance issues relating to Al that must be understood and managed. This article introduces some of those key risks and issues, together with the steps aviation businesses should take to reap the benefits of Al whilst successfully navigating the challenges.

Artificial intelligence

Al is the simulation of human intelligence in machines to conduct tasks that normally require human intellect. Al systems mimic human-like cognitive functions-such as learning, reasoning, problem-solving, translation and decision making. They analyse data, make predictions, adapt to changing circumstances, and perform specific tasks autonomously.

The significant developments in Al that we've seen in recent years have been made possible by breakthroughs in machine learning, the availability of large datasets and increases in computing power.

Opportunities for aviation

The aviation industry's complex operations - involving vast amounts of data and a track record of successfully leveraging cutting-edge technology - make it an ideal candidate for Al adoption, both on the ground and in the air.

On the ground

• Predictive maintenance: Al systems analyse historical data. and real-time information from aircraft to predict when components may fail, enabling proactive maintenance- reducing downtime and preventing unexpected failures.
• Crew scheduling: Al systems optimise crew scheduling, taking account of factors such as crew availability, legal requirements and operational efficiency-helping airlines manage their workforce efficiently.
• Customer touch points: Al-powered virtual assistants enhance customer service by providing real-time information about flights and baggage. Passenger personalisation better serves customers, improving experiences - from creating unique offers to autonomously resolving queries, complaints and claims. Supply chain management: Al systems optimise logistics- reducing delays and enhancing efficiency.
• Security: Al systems underpin advanced screening/threat detection - enhancing the effectiveness of airport security protocols.

In the air

• Flight planning: Al systems optimise flight routes based on real-time weather conditions, air traffic, and fuel efficiency- minimising fuel costs and emissions.
• Air traffic management: Al systems assist air traffic controllers to manage airspace - predicting conflicts, optimising traffic flow, and dynamically adjusting routes to minimise congestion. Training simulations: Al-driven simulations provide realistic scenarios for pilots and crew to practice emergency procedures and improve decision making.
• Safety: Al systems identify safety-related trends, allowing preventive measures to be taken - improving safety standards.

Navigating the challenges

The legal risks and compliance issues associated with Al systems and their outputs are constantly evolving as Al technologies develop and adoption increases. These risks and issues may be divided into input issues relating to the creation, licensing and training of Al systems, and output issues relating to the materials generated by Al systems. There are also overarching risks and issues - from compliance with law to commercial issues (e.g. will the system work and deliver acceptable returns on investment?) and ethical considerations (c.g.job displacement, an issue at the heart of recent US Actors' Union strikes).

Input risks

• Use of data: Al systems consume and produce data - and to get the outputs an organisation wants from an Al system, data will need to be input into the system. However, the nature of data and what you can and cannot do with it is, from a legal perspective is commonly misunderstood. Organisations often mistakenly think they 'own' data and can do what they like with it just because they have access to it. However, there are no property rights in data per se: nobody 'owns' data. Instead, there are rights, obligations and restrictions that apply concerning data usage. Examples include intellectual property (IP) rights, confidentiality obligations, contractual restrictions on usage, and legal restrictions imposed by the type of data - such as data protection laws relating to personal information.
  • These rights, obligations and restrictions may have an impact on whether data may be used in an Al system (or any other given purpose for that matter). So, if you want to use data as part of an Al system, you must make sure you have the right to use data in that way. Unfortunately, this is not easy given that most organisations will have received the data they hold over time from multiple sources in differing circumstances - each with its own understanding relating to use.
  • Indeed, 2023 saw the filing of several lawsuits alleging that Al system creators had misused data as part of the creation/training of their systems - such as the action recently launched by the New York Times (NYT) against OpenAl (the company behind ChatGPT) and Microsoft alleging that their chatbots were trained using NYT's proprietary content without permission.
• Data quality: to get reliable results from an Al system, input data must be accurate, consistent, current, complete and relevant.

Output risks

• Intellectual property issues: Al system outputs typically lack 'human creators', which creates uncertainty as to whether the outputs may be protected using existing intellectual property rights such as copyright.
• Transparency, explainability and control: understanding how Al systems arrive at their decisions is essential - particularly when it comes to validating decisions made or informed by Al and ensuring they have been reached in a compliant way. Without this understanding, Al system users are effectively relinquishing a degree of control over their businesses.
• Legal liability: organisations using Al systems will, generally, be liable for their outputs under English law- although they may, in turn, have recourse against any third parties that developed or provided such systems. While the nature of any liability will depend entirely on the relevant circumstances, it's possible for potential liabilities relating to Al system use to arise under contract, for IP infringement, defamation and for the breach of data protection, consumer, competition, employment and product liability laws, among others.
• Other issues: examples of other issues range from bias and discrimination to hallucinations (where Al systems produce credible looking outputs that are flawed or incorrect) and offensive or inaccurate output-issues often caused by incomplete, biased or inaccurate data.

Regulatory landscape - EU and UK perspectives

Just as the development of Al technologies is moving apace, so too is the regulatory landscape governing its use as governments search for the right balance between protecting their citizens from potential harms and creating a regulatory environment that doesn't stifle innovation.

This is particularly the case in the EU, where lengthy negotiations between the European Parliament and Member States in December 2023 saw the provisional agreement of the new EU AI Act. Aimed at ensuring ethical and responsible Al use- and with a focus on protecting fundamental rights and consumer safety - the Act defines mandatory requirements applicable to the design/development of certain Al systems and adopts a risk-based approach, establishing different obligations for providers and users depending on risk levels.

The EU's approach differs from that of the UK. Currently, there are no UK laws specifically governing the use of Al. Instead, Al is regulated by various existing laws that apply to the use and outputs of Al systems, rather than Al systems themselves - e.g. data protection laws. This approach is unlikely to fundamentally change based on the UK Government's March 2023 paper 'A Pro-Innovation Approach to Al Regulation' where- rather than imposing sector-specific rules or regulating Al technology itself-existing UK regulators will be tasked with applying a framework based on the following five principles as they determine appropriate: safety, security and robustness; transparency and explainability; fairness; accountability and governance; and contestability and redress.

It's an approach that is intentionally agile and iterative given the fluidity created by ongoing advancements in Al technologies and which is intended to make responsible innovation easier, strengthen the UK's position as a global Al leader, drive growth and increase public trust.

Get the right advice early on

As we've seen, the use of Al systems can give rise to complex legal issues - so getting the right advice early is key to ensuring the success of your Al project and avoiding potentially costly issues down the road.

Clyde & Co has a global Al working group that brings together sector specialists across our key geographies to advise clients on the safe adoption of Al technologies. From advising on risks and compliance issues relating to the development, training, licensing and adoption of Al systems through to advice on Al regulation, policies, audits, disputes and more, we offer a one-stop shop for clients looking to reap the benefits of Al by navigating the challenges.

If you're considering an Al project, please keep the following key takeaways in mind:

• Deal with data issues upfront: know the source of data and any rights, obligations and restrictions governing its use. Only use data for the relevant permitted purposes.
• Address issues relating to the development/use of Al systems early: from the development or licensing of the Al system to the use of data entered into the system and allocation of responsibility for system outputs, make sure you know where you stand.
• Protect against bias and discrimination: outputs containing bias or discrimination could lead to costly claims.
• Protect outputs: understand the IP position regarding outputs and protect them with contractual restrictions, if necessary. Manage cyber risks: put in place robust protections.

*Published in Regional International: The official magazine of ERA (January / February 2024)