AI-enabled IoT Security: Does If-then-else equal Artificial Intelligence?
The Internet of Things (IoT) brings upon a new infrastructural paradigm, where network objects are individually connected to the Internet, and are accessible via unique identifiers. Undoubtedly, a key enabler that rises to meet this technological challenge is the underlying wireless communications technology. Accordingly, the growing popularity and ubiquitous presence of WiFi, ZigBee, and Z-Wave constitute an indisputable proof on the sustained progress towards achieving this technological revolution.
While the IoT’s core concept, architectural requirements, and enabling technologies have been intensively analyzed, the security-aware design of an IoT gateway has received less attention from the scientific community.
Indeed, when one mentions the security within the context of IoT, most of the time it finds solutions containing fashionable buzz-words, which frequently involve “Artificial Intelligence” (AI), and “Machine Learning algorithms” (notice the quotation marks). To this end, when we search for the state of the art solutions in the field of IoT security, in many cases we stumble upon advanced algorithms that promise to learn and to solve all of the security problems found in IoT. Obviously, with the technological advancement, there are more and more tools that are available as open source and that provide the implementation of very advanced AI algorithms. Furthermore, starting early from the student years, people are encouraged to use readily-available tools for solving different problems. Let us take, for instance, the case of the artificial neural network. This very basic structure is nowadays presented to bachelor students, who are capable in only a few weeks to build their own networks and to approximate some phenomenon (e.g., a mathematical function). Even better, nowadays, (in many cases) students are not even expected to know how to implement a neural network, since we have all these interesting tools that are readily available (e.g., Tensorflow, PyTorch).
However, once we come to focus on the actual implementation of these intelligent and advanced algorithms, suddenly, the palette of solutions becomes rather scarce, and the only way out seems to be the implementation of a “lightweight” version of the “intelligent” algorithm. However, if we come to the point of being able to reduce an AI-based algorithm to a sequence of “if-then-else” statements (very simply put), then is this still AI? Even better, when discussing about embedding IoT security into gateways, why are we not discussing more about the appropriate coding style, instead of running after and just trying to use the most “fashionable” solution?
These are all issues and questions that the GHOST project is facing. It is not easy, and we do not claim to have found the perfect solution. However, GHOST does bring some practical solutions that could fuel the next generation of IoT gateways. The most valuable lessons that have been learned are certainly behind all the deliverables and behind the software modules. These represent the many (and, unfortunately, undocumented) discussions about coding style, about refactoring, about debugging, and ultimately, about producing a practical solution for a secure and intelligent home IoT gateway.
Kalos Information Systems