The analysis of security and privacy risks in smart education environments

1. Introduction

The worldwide transformation to a post-industrial society inevitably has a strong impact on the education system in every country. One of the key trends in the 21th-century education is that it has become mass; the need for higher education arises in an increasing number of people [18]. Therefore, it is necessary to either increase the number of teachers or increase the efficiency of the educational system. In the concept of a traditional research university, where each teacher is also a scientist, it is impossible providing quantity of highly qualified teachers [41] enough to meet the mass need in higher education. Thus, the problem can have only one solution – to increase the efficiency of the educational process.

The diversity of various specialties and the multidisciplinary nature of modern professions has led to the need of education individualization. The computing technologies has ensured a significant increase in labor productivity; those tasks that previously took months of calculations and design are now can be done in several hours due to different computers’ programs. Thus, mass training of specialists with approximately the same competencies is not needed anymore. Nowadays, the challenge for the high education system is the training of specialists with a unique combination of competencies based on individual educational trajectories [7].

The research and development technology cycle is reduced to five years, and the cycles of specialist training in higher education are four to six years at bachelor and master degrees respectively. The traditional education program cannot handle such high rates of technology development. Thus, the principles of life-long/width learning were introduced to meet the requirements of continuous learning [3,11].

Online education tools can partially solve these problems [30]. The module structure of the online course allows upgrading it fast without damaging other parts. Furthermore, a traditional laboratory group contains approximately fifteen people, and around one hundred people can fit in the lecture room, but online classrooms can easily increase these numbers three and more times. However, online education, like ordinary education, is a process of learning, training, and socialization. That is why to substitute traditional education environment, the Smart Educational Environment (SEE) should ensure the effectiveness of each education process component, not just learning.

When developing the smart educational environment, the new education system actively uses different Information Technologies (IT) like cloud services, big data, Artificial Intelligence (AI), machine learning, and others. The development of any technology or system is always associated with the development of their safety and security. For example, modern civil aviation is simply unthinkable without strict protocols that ensure safety. It is also true for smart educational environments. Without a comprehensive analysis of the risks associated with online education, further development of SEE in this direction is impossible.

Taking into account the risks of online education, universities try to implement information technologies in a traditional education environment. Therefore, the different smart campus technologies like flipped classroom [6], blended learning (online labs, self-blended, rotation model, flex model, online drive) [36], and ambient intelligence classroom [2] were developed. In contrast to online education, smart campus technologies do not solve the challenges of mass education. Of course, several problems of traditional education can be mitigated with technologies of smart campus, but not the main problem of «mass higher education». It is still the same campus with the same restrictions in the localization, the capacity of classrooms and buildings, and the availability of social infrastructure for different groups of citizens.

The purpose of the article is to analyze the technologies that form the educational environment from safety, security, and privacy points of view. The analysis focuses on the online education environment and partially raises the questions of smart campuses.

2. Smart education environments

The development of online education systems was focusing on online learning processes at the first stage. As a result, various systems such as MOODLE, OpenEdx, and CANVAS for managing the learning process, i.e., Learning Management System (LMS), and content, i.e., Content Management System (CMS), appeared. A student interacts with the LMS that controls his/her access to educational content and schedule, monitors activity and training effectiveness. Teachers use CMS to structure educational content, set the pace of its passage, evaluation parameters, and other elements of the course.

The architecture of both systems is based on the modular principles that allow further improvement of online education environment functionality. The architecture of the educational environment of the first and second generation is quite simple and ensures the interaction between three main roles (student, teacher, and educational organization) during the education process as shown in Fig. 1.

OPEN IN VIEWER

Fig. 1.

The architecture of the first and second-generation online education platforms.

However, these platforms can host only massive open online courses (MOOCs) [1]. Testing the first and second-generation online courses identified the following main problems [12]:

The next important stage in the development of online courses was the introduction of the «connectivism»methodology [42]. This methodology places emphasis on the processes of generating knowledge that include the connection of different knowledge holders like people, books, etc. Students’ competencies suggest an ability to generate proof of statements with peer-to-peer (p2p) assessment.

When studying an online course with p2p elements, each student in the course becomes also a teacher that allows not only to reduce the teacher workload but also to master skills for assessing different solutions [27]. In addition, students create social groups similar to professional groups, and sometimes directly join the professional community at the appropriate venues – forums, conferences, etc. To provide these changes, the educational platform was revised as shown in Fig. 2.

OPEN IN VIEWER

Fig. 2.

The architecture of third-generation online education platforms.

The concept of 3rd generation online courses is the possibility of having a dialogue between both teacher and students, and students themselves. However, the 3rd generation platforms do not provide sufficient opportunities such as educational games, virtual or augmented reality for different types of interaction.

The development of cloud computing, augmented and virtual reality technologies has transformed many professions. A specialist has essentially become a remote operator of a technical system like a robot, aircraft, and machine. Therefore, online education is now possible even in areas such as medicine, wherein full-time education was traditionally required to introduce a person to real working conditions. One more example is the joint projects or collaborative development in IT industry where the product is the result of collaborative work of different specialists – analytics, developers, system administrators, and so on.

The smart educational system should ensure the adaptation of future specialists to such working conditions. Therefore, online education is no longer an opportunity, but a new requirement, especially in the field of IT specialists training. The smart education environment should implement the concept of collective creativity and project work, but these opportunities for joint development are not available within the framework of the 3rd generation platforms. All these factors led to the development of fourth-generation educational platforms with special online courses [40].

The key features of these platforms are the expanded use of design approach in joint projects, synchronous sessions in groups, and usage of cyber-physical/virtual laboratories. The video conferencing system (VCS) solve the problem of synchronous session’s organization only at the first approximation. However, VCS is not able to solve this problem completely, since the effectiveness of training for a student largely depends on the degree of his immersion in the process. It is also true for teachers; it is quite difficult to immerse yourself in the lecture process staying in-home environment because existing different noise sources or other bad conditions. The online lecturing rooms such as Oxford HIVE, Harvard hub, and others [46] are partially mitigating this challenge. However, common universities and schools, where the need for these systems is often high, cannot buy them due to high costs. In this regard, the development of virtual reality technologies and virtual lecture rooms received impetus [35,45].

The support of virtual and online laboratories also changes the educational environment. Such laboratories can be both simple CAD as cloud service and/or a complex laboratory that allows a student to remote control the robotic complex using special programs (Fig. 3(b)) that students are developing during the learning process as shown in Fig. 3(a).

OPEN IN VIEWER

Fig. 3.

(a) The cyber-physical laboratory “autonomous service robots” in PNRPU. (b) The example of program developed by student to control robots in laboratory.

The example of the Autonomous service robotics online laboratory at the master degree program in Perm National Research Polytechnic University (PNRPU) demonstrates that the concept of 4th generation educational platform allows educating specialists in technical sciences and engineering. In this laboratory, students can develop special programs to control, navigate, and solve other tasks on real robots, including joint development of robots’ missions for different business cases. For example, students developed missions of robot waiter and robot animator. Thus, now the educational environment is a complex distributed system in which a plurality of users interacts with a set of services as shown in Fig. 4.

OPEN IN VIEWER

Fig. 4.

The architecture of fourth-generation online education platforms.

It should be noted that not only the architecture of the system as a whole is complicated but also each component individually. As a result, one teacher is no longer able to provide development and support for all elements of the educational course. Therefore, the development of online courses is usually teamwork nowadays.

The proctoring module in LMS improves the evaluation system. The LMS proctoring module based on AI should have access to the microphone, video camera, and other software on the student’s computer to ensure the honesty of exam or test passing. In addition, SEE proctoring and authentication system stores biometric personal data to recognize voice or face that carries significant information security risks [21,48]. Moreover, AI-based course assistants supplement students learning process in LMS. An example of such modules is chatbots, which helps students answering their questions and reducing the teacher workload [15,38].

The existing problem of cheating and copying answers for practical tasks can be solved through cyber exercises (online training ground or cyber ranges in information security). Training ground is a model of an information system in which user should solve certain tasks with a scenario. The training ground is constructed according to a given template, but each time with a new random answer to the task. Thus, the student should perform an algorithm to solve the task; he/she can copy this algorithm, but not the answer to the task itself.

The technology of online training grounds creates an effective combination with the trend towards the gamification of education [24,31,43,44]. It is quite simple to create competition from the passage of such training grounds and reward the best students. Herewith, the uniqueness of the training grounds provides transparency of the competition results. At the same time, the framework of online training grounds allows all students to repeat tasks until completion. Thus, students who failed exercise can study more effectively since they gain experience faster if nothing prevents them from repeating the attempts to solve exercise until successful result. The students who passed an exercise can further repeat it to improve their skills.

In summary, the key challenges of 4th generation smart education environment:

Smart education environments are inheriting both security threats from traditional systems and facing completely new threats in the information space. The first step is the identification of the information security and privacy risks and threats. Risks are associated with key challenges of the IT implementation into the educational environment. These key challenges are:

The mix of real and virtual space (or deep penetration of digital technologies into reality) causes the impossibility to process risks through avoidance or share. The complexity of IT systems in education does not allow singling out the only entity that will be responsible for risk processing. For an attacker, the entry point into the SEE can be located in a variety of places like students’ or teacher Personal Computers (PC), mobile devices, educational environment servers, course components, web service, and so on.

Following the concept of lifelong learning, the educational system must accompany a person almost throughout his life, especially intensively in the first 25 years [29]. Digital profiles of the student collect all achievements and personal data for this task. The source of the threat to digital profiles can be Cyber Crime (CC) groups, insiders with access to the database, and competitors.

In SEE case, social networking is a platform to support longtime communications between professional groups. Therefore, malefactors can collect users’ data on the platform to apply Open-Source Intelligence (OSINT) methods and Advanced Persistent Threats (APT). Possible threat sources are cybercrime groups, fraudsters, cyber intelligence services, competitors.

The third-party authentication and authorization protocols are common solutions in a distributed online education environment. Perhaps it is the only possible solution. That is why communication channels should have high-level security to avoid Man-in-the-Middle (MitM) attacks or other threats related with information modification. It is also true for remote access protocols, especially in the case of laboratory equipment. Threat sources are insiders, script-kiddy, and CK-groups.

The wide dissemination of AI as assistants on the course leads to the transfer of responsibility for the educational trajectory from a human (teacher) to AI assistant. Malefactors potentially can manipulate human choices through AI. Threat sources are insiders with access to the database, CC-groups, and competitors.

3.3. The key technologies to mitigate security and privacy risks in smart environments

Biometrics as an authentication factor has serious risks associated with the physical security of the person and the consequences of disclosure. It is easy to fix disclosure consequences for factors such as knowledge (password) or possession (phone). However, what to do if fingerprint or other biometric data are compromised? In this regard, the development of authentication technologies based on the zero-knowledge protocol is important to reduce risks related with an authentication system in SEE [16]. Nowadays, blockchain and cryptocurrency technologies are actively using principles of zero-knowledge proofs [20]. These methods are important not only for SEE but also in smart city services, especially when using biometric data [13,25].

The next question is when and why should robots substitute humans in education? In several cases, AI-based assessments of personal educational programs or skills are useful. In another case, studying a subject just because your case does not fit into the AI model can decrease the effectiveness of the whole education system, not only individuals’ learning process.

The vulnerabilities of machine learning algorithms to various kinds of attacks were described in articles [4,5,10,19,22,37]. People can do mistakes, but a robot will always mistake if its program was corrupted. If the training data set was poisoned or an improper AI algorithm used, an objective machine can deceive the people due to they do not know how this concrete AI operates. The AI technologies’ key threats are focusing around manipulating training data set, poisoning the machine-learning model through error correction mechanisms, and replicating training sample data through using the I/O interface as shown in Fig. 5.

OPEN IN VIEWER

Fig. 5.

The attack vector on AI-model.

The threats associated with the usage of AI forces countries to implement legislative regulation of the conditions and rules for its use in various services. In the recently published document [9], European Union plans to limit the usage of AI depending on the area of its application. The document assesses various risks that can be applied to SEE. The four main categories of risk according to this EU document:

Unacceptable risk includes AI systems or applications that manipulate human behavior to circumvent users’ free will.

High-risk includes educational or vocational training that may determine the access to education and professional course of someone’s life (e.g. scoring of exams).

Limited risk, when using AI systems such as chatbots, users should be aware that they are interacting with a machine, so they can take an informed decision to continue or step back.

Minimal risk allows the free use of applications such as AI-enabled video games or spam filters.

In conclusion, let us to outline and shortly discuss the key technologies in the field of security and privacy, which development is necessary for the mitigation of the risks identified above and the sustainable operation of any SEE:

Technologies for storing and processing biometric personal data – morphological biometrics and behavioral biometrics in proctoring and other services should be secured properly from spoofing and stealing. The integrity of data can be provided by a set of various methods as hash-digest or steganography. The major question is how to generate a unique biometric data sample for each use case? If the stolen data are useless in another service, it will reduce the profit as well as the attraction of an attack for hackers.

Technologies for transparent and controlled AI implementation in education – the AI-based tools in the crowd control systems such as the pattern recognition and the social scoring systems can help to detect potentially dangerous people. However, any mistake in a decision can ruin social trust in these systems. People should always know who gives them advice or assessment – human or robot. The securing AI technologies important not only as technical challenge but also as a response to ethical challenges. It will be helpful to have common criteria for AI-based IT systems because it includes not only functional security requirements but also security assurance. If AI can show how it receives a solution, it partially will solve the problem. However, it arises great privacy challenges in SEE because can cause the leakage of private data from the training data set.

Risk management technologies not based on expert assessment – in Table 2 the criteria for impact assessment were given, but is it a suitable for concrete Smart City? All current risk management frameworks involve experts’ decisions to receive assessment in impacts, probability, and other parameters. The risk management system is usually improved by n-of-m voting, but it protects only from m minus n mistakes. For example, the 2-of-3 system protects from only one mistake. In reliability theory, this problem is known as observation for the observer. Security specialists are facing the same problem in Public Key Infrastructures (PKI) with the chain of trust and root certification authorities. The development of risks assessment systems not based on an experts’ decisions can help us to look at the risk from the one or close points of view and mitigate this risk.

Authentication technologies in distributed and decentralized systems, including the development of authentication methods based on zero-knowledge proof – the best way not to lose a secret is never know it. Storing sensitive data on one server or moreover providing a single entry service for all SEE services are high risk. In an ideal zero-knowledge-proof system, the server does not store any sensitive data. The distributed registry like blockchain help to mitigate risks of trust chain and bottleneck. The problem of a Trusted Third Party (TTP) does not appear if data about pair {public key (authenticator), identity} stores in blockchain [47]. However, blockchain has several problems like well known 51% attack. In addition, it not only complicates spoofing attacks but also increases costs to revoke, delete or replace identities. Nevertheless, the threat source should compromise the majority of nodes in the blockchain system to spoof identity or provide another MitM-attack. In this case, an attack on the system can be too expensive for the threat sources with medium sizes and resources.

Secure architecture technologies based on zero-trust methodology – the distributed environments are weak for such attack techniques as social engineering or supply chain compromise. All these techniques are effective against zone-based security systems since entry points are in a trusted zone. The concept of zero-trust security architecture is based on the authentication and authorization independence from zone trust level [26]. The source should receive authorization for each action depending on trust level that can be increased with authentication and regardless from belonging to the zone. The article [32] proposes the access control system based on zero-trust for smart campus networks. The further development of the zero-trust methodology looks like a necessary solution to improve the architecture of all smart city services and the SEE as particular case.

The computing and communication based on quantum technologies – the Quantum Computers (QC) are potentially can decrease efficiency of authentication and encryption algorithms used. However, the current QC is far from solving this task. For example, QC needs ten thousands qubits to decrypt the 2048 bits RSA key during one year [17] or 20 million qubits to solve this task during 8 hours [14]. Whereas, the best QC simulations reaches only 256 qubits [8]. The communication system based on QC can provide alternative solution to the TTP problem. Therefore, building network based on QC technology for critical services of Smart City can significantly increase information security level. Although, it can be very expensive and should be considered as alternative solution in comparison to existing technologies with proved level of security.

The development of any technology involves the implementation of safety mechanisms, for example, modern civil aviation is not possible without strict flight safety protocols and policies. Smart city information technologies are affecting the interests of citizens, which means the developers should take into account the risks associated with privacy and personal data.

Modern educational platforms as part of smart city services use virtual and augmented reality technologies. Thus, the boundaries of privacy, the rights to anonymity and oblivion, and other privacy factors have to be determined in SEE. The smart educational environment stores confidential data that can be interesting for different threat sources. Moreover, denial of service attacks can destroy laboratory equipment, online classrooms, and other elements of the education environment since the system grants remote access to these objects. That is why the developers should take into account both the privacy and the information security issues. The analysis shows that the usage of biometric data and artificial intelligence in different education technologies causes the most dangerous risks.

In addition, there are issues of artificial intelligence ethics. For example, the applicability of artificial intelligence as personal assistants to teachers. On the one hand, big data analysis allows creating algorithms that are more aware of our preferences and desires; on the other hand, the algorithm does not have moral responsibility for its advice, unlike the teacher, it does not know what is kind or worth. Moreover, there are risks of manipulating the AI training samples to impose a students’ choice. At last, how to use personal data in the training of AI? GDPR gives a person the right to be forgotten, but how can it be implemented in an AI system that is already trained and operates? This question needs a systematic discussion [8,39].

The implementation of the zero-trust, the zero-knowledge proof, and the quantum computing technologies significantly improves security and privacy in SEE. Nevertheless, it is not enough to sustainability of SEE. Further development of technologies for risk management, storing and processing biometric and other sensitive data, and the managing AI security is need.