| 4. |
- Flammini, Francesco, Senior Lecturer, 1978-
(author)
-
Artificial Intelligence (AI) applicata agli Autonomous Systems
- 2018
-
Book (other academic/artistic)abstract
- The study of artificial intelligence applied to autonomous systems has in recent years aroused growing interest at the international level, and it is expected that this interest will continue to grow in the coming years [34]. It is a fairly well known fact that in the past many technologies now used in the civil field have seen the light, more or less secretly, in the military sector. Consider, for example, the so-called ARPANET, developed by the US defense department, which anticipated the modern Internet, but also algorithms for data encryption, thermal cameras, and many other commonly used technologies. Today the scenario has partly changed, shifting the leadership of innovation towards other domains, since there is a considerable boost to the technological development in the civil field with the advance of connected society paradigms like Smart-City and Industry 4.0. One example is related to the self-driving vehicles, born in the military sector, which are developing more rapidly in the civil sphere with the attractive self-driving cars. It is therefore important to transfer enabling technologies from one domain to another (cross-fertilization) and to draw appropriately from the outside (open innovation). This is achieved through studies and researches such as the one addressed by this monograph. The objective of this study is to analyze the principles, the basic methodologies and the operational tools of artificial intelligence applied to autonomous systems, at the modeling and technology level, in order to replace human-controlled vehicles with autonomous or semi-autonomous vehicles (e.g. drones) in high-risk operating environments, as well as to reduce human errors and to speed up response times, for example in operations command and control centers. The study presents an overview of the information fusion approaches to enable artificial cognition, mentioning several relevant applications in the military field, already at an advanced phase of development or even at an embryonic level. These approaches can be used to strengthen weapon systems and defense means, with greater ability to adapt to the operational context for the dynamic management of uncertainties and unforeseen events, as well as for experiential evolution and learning. Future applications include not only self-driving vehicles and smart weapons, but also the strengthening of soldiers through prosthetics and exoskeletons. Many of the future projections have been formalized by the working group on Symbiotic Autonomous Systems – which the writer is a member of – of the Institute of Electrical and Electronics Engineers (IEEE), enclosed in a special White Paper [34]. The present study addresses the impact of the Artificial Intelligence (AI) on the use of the military instrument when this technology will be applied to military assets and weapon systems, taking into account the different declinations of AI, including: • deterministic (semi)autonomous systems implemented through Boolean logical operators (eg Event Trees); • (semi)Autonomous systems based on probabilistic / stochastic models for the representation of knowledge and inference (eg Bayesian Networks); • (semi)Autonomous systems based on trained artificial neuronal models (ANN, Artificial Neural Networks). These approaches are based on different models of machine learning, which can be supervised or not. They apply to classification and clustering approaches in modern data analysis approaches, particularly in the presence of large amounts of information (big data analytics). This study distinguishes between semi-autonomous AI models, which require the confirmation of decisions by human operators (DSS, Decision Support Systems), and complete autonomy, which presents predictability problems impacting the verification and validation process and therefore system safety. These are the cases in which the aforementioned ethical, procedural, normative and legal implications are more relevant [1]. The introduction of autonomous systems equipped with artificial intelligence involves transformations also at the level of military logistics, which can be interpreted in two directions. On the one hand, it is necessary to plan the procurement of enabling technologies, the so-called deployable systems based on secure wireless networks, and the updating of systems to support complete digitalisation, which is an essential pre-requisite for the adoption of the instrument. The other side of the coin is the use of a higher level of automation in military logistics, supported by the AI. Here we can mention the automatic multi-objective optimization algorithms for decision support (eg genetic and evolutionary programming), the computation of the most efficient paths (in terms of time, energy, etc.), the dynamic definition of optimization priorities, as well as aspects of resilience through automatic re-planning of the route in the event of interruptions on the predefined trajectory. For all that has been said so far, it is clear that the development of the AI will have consequences on the future organization of the armed forces, both for the conduct of the operations and for the structure and numbers of the defense sector. As in other areas subject to automation through the use of new digital technologies, even in the military one the human role of decision supervision, feedback and control of high-level operations will remain decisive for many years. At the same time, however, the need for training and specialization in line with the complete computerization will arise, with significant impacts in terms of information security (or cybersecurity), which will require increasingly specific skills. The fact that complete autonomy would be possible in the event of unavailability of personnel in control centers implies not only a higher level of security, but also the possibility of reducing organizational redundancies by dedicating resources to different and more specialized tasks. As already underlined, there are significant ethical and legal implications related to future decision-making processes for the choice of using force through a weapon system governed by an artificial intelligence, potentially endowed with a high level of autonomy. It is therefore essential to define clear and shared limitations and conditions of autonomy for the verifiability and traceability of the decision-making process. In particular, in order to govern decision-making and prevent ambiguities, it is essential to apply the well-known RACI (Responsible Accountable Consulted Informed) paradigm, which defines for each action who is responsible for its implementation, who is associated with its administrative / legal responsibility, who will have to be consulted for further information and possible approval, and finally who will have to be simply, but obligatorily, informed. All aspects related to international safety certifications that regulate design, development and verification of systems whose malfunctions can impact on the safety of people are also essential. Many of the current reference standards are no longer adequate if we consider the current and anticipated evolution of AI, and therefore they will have to be adjusted accordingly.
|
|
