Artificial intelligence (“AI”) and autonomous vehicle technologies (“AVT”) have the potential to redefine how the aviation industry operates. While the operational changes that these technologies will bring are being widely explored, the legal issues raised by their rapid introduction into the industry are not. In this two part series, we will be looking at applications for AI in aviation and its effect on the legal liability and regulation of those who use it.
What is it and how is it used?
What is artificial intelligence?
No one agrees on the definition of AI. While the term dates to 1955, it continues to conjure thoughts of HAL from 2001, or Skynet from James Cameron’s Terminator movies. In reality, AI generally refers to algorithms and related statistical methods aimed at imitating (or exceeding) human reasoning, pattern recognition, problem solving and learning capabilities.
Modern “AI” algorithms are hungry for data. They rely on massive amounts of information (generally referred to as “big data”) to drive a mathematical “training” process called “machine learning.” Once these algorithms have been “trained,” they can quickly identify patterns, changes, and solve problems present in the processed data, provided that the underlying algorithm is designed to detect such patterns and changes, or to solve such problems.
Modern dictionaries define AI as a branch of computer science concerning the simulation or imitation of intelligent beings in computers, which includes the ability to learn and adapt to changing or unanticipated circumstances. (see Bernard Marr, The Key Definitions of Artificial Intelligence (AI) That Explain Its Importance). Amazon defines AI as “the field of computer science dedicated to solving cognitive problems commonly associated with human intelligence, such as learning, problem solving, and pattern recognition.” (see what is AI). AI is inextricably linked with the field of “big data,” because AI systems learn faster (and thus become “smarter”) with more data, and also to the advent of the “Internet of Things” (“IoT”), because IoT sensor technology “exponentially adds to the amount of data to be analyzed.” [Id.]
This machine-learning branch of AI has clear application to the aviation industry. A few among many are:
- Anomaly Detection: Observations or events that deviate substantially from expected patterns of behavior. Such insights could be used to identify risky passengers or potential safety issues.
- Fraud Detection: One of anomaly of great potential interest is fraud detection. AI algorithm can be used to prepare predictive models that flag certain transactions as potentially fraudulent, and alert human reviewers to take a closer look prior to approval.
- Air-Traffic and Fleet Management Optimization.
- Weather Analysis.
- Aircraft Diagnosis and Flight Planning.
- Image Recognition: The obvious application is for airport security; it could be expanded to include areas of an airport other than security checkpoints.
- Natural Language Understanding: For reservations, passenger complaints, etc.
- Recommendation Engines (also known as “chat bots”). To assist customers in navigating a wide-array of travel options when making reservations (mathematicians call this narrowing the selection set).
- Human Resources. This is another application of recommendation engines. AI can be useful in reviewing and promoting to the top of the pile the best candidates for various positions within an airline or company.
- Augmentation of perception, cognition and problem-solving abilities. While it may take decades before the travelling public will accept a pilotless aircraft, current AI technologies are already being used to “augment” pilots’ abilities, and that trend is likely to quicken as AI systems are incorporated further into avionics systems.
- Autonomous Operations: Baggage screening and handling; airport shuttles; passenger check-in, etc.
- Unmanned Aerial Vehicles (drones): This technology is upon us and will be rapidly expanded into unmanned cargo planes and other applications, which may or may not be controlled by humans on the ground. These craft will continuously communicate with, and learn from, each other as they operate.
 The term was first used in 1955 in a proposal for a summer academic workshop. “The Dartmouth Summer Research Project on Artificial Intelligence.” John McCarthy, Marvin Minsky, Nathaniel Rochester & Claude Shannon, A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence (Aug. 31. 1955), AI Magazine, Vol. 27, No. 4 (2006). The proposal provided that “[t]he study is to proceed on the basis of the conjecture that every aspect of learning or any other feature of intelligence can in principle be so precisely described that a machine can be made to simulate it.” Id.
Are autonomous vehicles artificially intelligent?
As suggested above, AI algorithms are capable of interacting with the physical world, such as by operating a vehicle, airplane, or vessel, if the algorithmic “brain” has been exposed to enough operational data. In other words, we are already in a place where human operators do not necessarily need to be in control. These “autonomous vehicles” can already safely navigate between points in “autopilot” mode without human assistance by leveraging both existing and rapidly emerging technologies such as big-data, cameras, sensors, active steering, cloud-sourced mapping, radar, GPS navigation, and lasers. In this way, it is appropriate to view autonomous vehicles as a quickly evolving application of machine learning and AI.
Now that we have an idea of what AI in aviation looks like, in the next part of the series, we will examine how regulators and the courts will deal with these issues.