Road safety applications of Artificial Intelligence can be broadly grouped into two main types/phases of processing, requiring different types of data, methods and expertise. On one hand, AI can be used to handle sensor and machine data and translate them into reliable and meaningful data sets for road safety purposes; requiring more technological and industrial expertise. On the other hand, AI can be used to dive deep into data in order to identify hidden properties, correlations and combinations; requiring contextual and statistical analysis expertise. In both types, engineering skills combining data and context analysis skills are necessary.
Road safety is one of the most complex domains requiring multi-disciplinary analysis, becoming even more complex when big data are involved. The probability of erroneous analysis is quite high and becomes even higher with the advent of too massive data available and the multitude of analysis techniques. Very often we look where the data are and not where the problems and the solutions are, resulting into important investments not bringing the desired results.
The appropriate application of Artificial Intelligence techniques and optimum use of big data into road safety has to follow the fundamental principles of data science: need for appropriate data, followed by optimum use of analysis techniques, which require sufficient effort and expertise and efficient management of the processes.
Firstly, appropriate data are needed in order to meet properly established clear objectives for the specific application or problems to solve. Data should be matched with road safety goals, ensuring being directly relevant and contributing meaningfully to the overarching objectives. Careful and meticulous collection of road safety data becomes imperative, emphasizing the importance of process accuracy. Subsequently, a rigorous data cleaning process is needed in order to enhance the quality and reliability of the information. Moreover, recognizing the significance of meta data is crucial, as it provides context and structure to the collected data, facilitating their wide and shared use.
The optimum use of analysis techniques requires a strategic and dedicated approach. To begin with, by adopting the concept of “think big but act small”, it is imperative to proceed step by step without however losing the big picture. Exploiting existing small- or large-scale experiences becomes a valuable resource, allowing for the extraction of insights and lessons that can support the analytical process. Furthermore, learning to find and exploit the appropriate methods for a given context is essential, emphasizing the importance of adaptability and an understanding of various analytical approaches. It is crucial to bear in mind that correlation does not necessarily imply causation, necessitating a cautious and discerning interpretation of relationships within the data. Fostering an interdisciplinary mindset is the key which recognizes that insights from diverse fields can enrich the analytical process and lead to more comprehensive conclusions.
Exploiting AI and big data in road safety requires sufficient effort. Starting with small scale road safety examples and applications to be tested, allows for the evaluation of strategies before progressing to larger-scale implementation. In addition, simple statistical methods can establish a solid foundation which encourages the development of building blocks that can be progressively integrated into a more comprehensive analysis framework. Particular attention should be given to the selection and exploitation of the right road safety examples, ensuring that the insights gained are directly applicable to the specific context. And of course good results depend directly to properly judging, planning and devoting the necessary time, effort and expertise throughout the process; in fact what we put is what we get.
Efficient management of AI use in road safety involves an harmonious match of ambitions with the respective budget, effort and expertise. Recognizing the importance of building capacity, both in terms of road safety data and expert knowledge, is fundamental for sustained progress. Initial external assistance provides valuable insights and support, allowing for the gradual establishment of internal capacity. Then, synergies are progressively created, both methodological or contextual. Understanding that efficient management is a continuous journey rather than a short trip underscores the need for adaptability, continuous improvement, and a long-term perspective in achieving sustainable advancements in road safety.
The fact that human factors are the key crash risk factor, makes that for the full exploitation of AI potential there is high need of data from the road users. However, the big challenge is to convince the users that we can achieve reliability and efficiency by preserving in parallel personal data privacy. This is easy and very much possible to achieve in theory but in practice it is quite a challenge. As long as humans are in the process no process can be fully safe and humans will always be part of the overall process. Public and private system operators who have access to the source and/or processed data and hackers and malicious people will always attempt to have access to the data.
There is already great progress and a lot has been done so far, with both technological and institutional solutions. Firstly, several security techniques exist (blockchain, multi-step verifications, etc.) to prevent unauthorized human access in several steps of the processes. Secondly, in more and more countries worldwide, comprehensive rules and control procedures are foreseen, with several levels of independent controls to prevent that no one has access to the complete sets of data. In our world and in our systems, there have always being new problems and new solutions then new problems and new solutions, so AI systems respecting data privacy is an eternal quest and Authorities, Experts and the society should continuously strive to contribute and cooperate as needed in this quest.
Bridging the theory and practice of Artificial Intelligence for Road Safety is a major challenge that we have to work systematically and with persistence. Identifying the problems and the key solutions is already a fundamental first step, then expertise and technology should be put together towards reliable and useful AI applications for the improvement of road safety globally and ultimately reach the vision zero fatalities by 2050.
Intervention at the 6th International Traffic Safety Forum at Dammam, in December 2023
