**5. Conclusions**

IoT systems and related services need to be truly supported by a pervasive, reliable, and effective intelligence to unleash their disruptive potential in our daily lives. Cloudification has so far helped, but the latest candidate to burst onto the scene is EI, whose rich, though fresh, recent literature reflects its broad appeal and usefulness. This survey provided both a quantitative and qualitative analysis of the large body of knowledge

related to EI and rapidly accumulated in the last decade by means of a systematic literature review of secondary study according to the well-known PRISMA guidelines.

As a final takeaway of this survey, we recognize that the ETSI MEC reference architecture provides a solid base for EI system engineering; however, the realization of AI functionalities is open, and intelligence has not ye<sup>t</sup> been considered as a built-in capability of the edge system [14]. As result, it is still not completely clear how and where the EI capabilities should be built into the edge systems to achieve its maximum yield, while further specifications (mainly for standardized APIs, software constructs, interoperability mechanisms, supporting infrastructures) need to be developed. In particular, the latest concept of the edge–cloud continuum, at its extreme, may lead to isomorphic EI architectures, allowing the identical service provision among edge devices, gateways, and servers [14,76]; from such a perspective, data and computation can be transferred dynamically and performed on any level of the cloud–edge architecture that provides the optimal QoS/QoE, thus ultimately diluting or even dissolving the boundaries between the cloud and edge.

To conclude, we attempted to disclose the wide research area of EI, and we hope that this survey can supply basic knowledge to enable new researchers to enter the area, current researchers to continue developments, and practitioners to apply the results, being confident that huge research efforts will be carried out to completely realize EI in the incoming years.

**Author Contributions:** All authors contributed equally to this paper. All authors have read and agreed to the published version of the manuscript.

**Funding:** The research leading to this work was carried out under the Italian MIUR, PRIN 2017 Project "Fluidware" (CUP H24I17000070001), and under the "MLSysOps Project" (Grant Agreement 101092912) funded by the European Community's Horizon Europe Programme.

**Data Availability Statement:** Data sharing not applicable.

**Conflicts of Interest:** The authors declare no conflict of interest.
