With enormous potential in the mobility sector, artificial intelligence has numerous applications in our industry. Depending on the criticality of the operations on which it is implemented, we can apply different types of models, known as “black box”, “gray box”, or “white box”. Each color is related to the knowledge of the origin of the data and the possibility of understanding how the results are achieved -explainability-.
Artificial intelligence -hereinafter, AI- has experienced exponential growth, especially in consumer goods companies. Its ability to segment audiences and personalize commercial messages has revolutionized marketing. Process automation has helped reduce administrative costs, and prediction systems allow for more accurate demand and supply adjustments. Even with a restrictive legislative framework, and considering the logical limitations for the management of personal data, AI offers an infinite range of possibilities.
But there are sectors, such as mobility, where AI applications must adapt not only to a regulatory context but also – and above all – to a critical environment where safety always comes first. In this sector, all companies face the same challenge when addressing AI projects: the so-called “black box” challenge.
As the name suggests, the black box refers to the inability to “see” the internal functioning of AI. We talk about a “black box” when algorithms provide results -suggest actions, make predictions, detect anomalies, etc.- but do not inform about the underlying reasoning behind the result. This is especially evident in models based on Deep Learning, such as Generative AI.
Consumer goods, leisure, and other non-critical areas do not face this dilemma. They do not have to explain how or why their AI models reach certain conclusions. Neither Meta nor Amazon, for example, need to detail why their software recommends certain content or products; their algorithms are mainly based on finding similarities between products and consumers. This does not pose legal or security issues; it is enough to comply with regulations and demonstrate that it is commercially reliable and viable.
The situation is different for companies that develop and manage critical infrastructures such as public transportation. At Alstom, we cannot rely on any solution that makes critical decisions in an opaque manner without explaining the exact reasoning that has been followed. This does not mean that we do not integrate AI into our systems. On the contrary, AI is becoming increasingly relevant to improving our services and the reliability of our technologies. It means that, like with all our systems, we use AI prudently and responsibly, applying the strictest safety standards and the most rigorous testing in every situation.
AI Models in Mobility
At Alstom, we use three different “types” of AI, depending on the project and its requirements. We use the so-called “black box AI” in projects that do not pose security issues. For example, to control passenger density in the network and help optimize passenger flow on trains or adapt schedules and operation services.
The so-called “gray box AI” is used to support engineers in their developments, identifying different situations, improving calculation accuracy, or assigning parameters to formulas to enhance simulation models. The simulators that recreate railway systems are also considered “gray box.” These simulators reproduce subsystems -and their behavior- to offer a view of an integrated environment and facilitate decision-making. Using historical data, it helps calculate the maximum number of possible scenarios, classify them, and identify response options.
Finally, we find “white box AI,” with interpretable and deterministic models that can explain their behavior, detail how they make predictions, and which variables influence their results. The relatively recent development of this type of white box AI allows us to expand AI applications in the field of mobility. When the origin of the data can be demonstrated, it ensures that they are explainable and reliable, and consequently, they can be integrated into critical processes from a security standpoint. This is where the different big data technologies and algorithms we already use to improve maintenance, monitor the fleet in real-time, or predict possible incidents in equipment and infrastructure are framed.
Developing the AI of the Future
Advances in AI have been rapid in the last decade, with various applications in numerous sectors beyond technology and consumer goods companies. In the railway sector, we increasingly use AI to improve train scheduling, speed, predict passenger demand, enhance asset management, or reduce energy consumption, among others. At Alstom, we continuously work to refine these applications and find new innovative ways to use AI safely and effectively.
Rail transport is the most sustainable mode of transport, and thanks to digitalization, we improve its efficiency and profitability. To take advantage of all the benefits of digitalization, the sector must continue to invest, expanding the use of AI intelligently and thoughtfully, adapting this technology to the most suitable applications. And always bearing in mind that data does not always wear white.
