The current state of the art in AI indicates its applicability to natural language processing be it text understanding, or text generation, or both. Among other applications, AI-based technologies already placed on the market demonstrate their capacity for generating relevant test assignments following text provided as input. Those applications are still in their research and development phase and the outcome is still unknown. What’s important, though, is would it be allowed from the regulatory point of view when such AI-aided knowledge assessment courses on the whole and for energy sector staff in particular are proved to be a working solution, ready to go in production and placed on the market?
The issue therefore is what exactly is the regulatory framework in Ukraine for that and what it might be like in the foreseeable future. While Transformer-based technologies being marketed as Generative AI are trending for now, the dimension in focus allows the approach used to be technology-agnostic. And the key elements of the regulatory framework to consider are as follows: the national legislation in general, the sector-specific regulations, the supranational layer, the international layer.
The AI-related provisions at the level of the national legislation are in their infancy for now. Which means that the Government did approve the Concept of the Development of Artificial Intelligence in Ukraine together with the implementation plan while those are not of regulatory nature per se and add nothing to the regulatory landscape beyond what the public authorities should do. Besides that, the concept does not cover energy sector at all, except for protection of critical infrastructure within the cybersecurity direction, education and employment sectors are not covered in terms of putting AI technologies into practice. No other act of legislation provides for requirements for application of AI technologies, except for those being fragmentary or incidental. The civil law branch of legislation allows for AI applications by analogy (some are of the utmost gravity, yet legally fragile) but that is not relevant to the subject under discussion here.
At the level of the sector-specific regulations application of AI technologies is not directly governed for energy, education, employment sectors and intersections thereof. Except for where ICT, computers, software, etc. could be interpreted so widely that those include AI. But mostly it could not because being enacted long before AI technologies become reality. Mostly, the sector-specific regulations are just ignorant of AI technologies and applicability thereof.
That said, the regulatory significance of the national legislation and the sector-specific regulations in terms of AI amounts to non-existent at the moment. There also exists an unexpected regulatory side effect: under the laws and regulations that restrict everything except for what is expressly allowed, AI applications are restricted by default; under the laws and regulations that allow anything except for what is expressly restricted, AI application are allowed by default.
The supranational layer as such is currently not relevant for Ukraine. The closest to that in essence could be the legal regime of the Commonwealth of Independent States and what the EU Association Agreement brings in its provisions. But the former is not worth considering because Ukraine makes what’s possible for that to come to an end, the latter is mostly about what Ukraine is to implement at the level of the national legislation, and both are substantially instances of the international layer to be considered further. Yet in 2022 Ukraine applied for the EU membership and was granted the EU candidate status. So, the supranational layer is not relevant just yet but might well be relevant in the foreseeable future. Following this trajectory, the Ministry of Digital Transformation of Ukraine unveiled the Roadmap for the Regulation of Artificial Intelligence in Ukraine that reads that in terms of AI Ukraine takes the regulatory track that is currently being developed in the EU.
Which brings us to the EU AI Act. In 2021 the European Commission communicated its proposal for a regulation laying down harmonised rules on AI abbreviated as AI Act, in 2022 the European Council provided its position on the AI Act, in 2023 the European Parliament adopted amendments on the AI Act. Following the EU legislative procedure that means that the regulation is adopted in the first reading and the last phase of the negotiations before the regulation is passed into law is under way. The outcome and final wording remain unknown but the documents unveiled to the general public allow to make suggestion what that could possibly be.
The general approach is that all depends on the level of risk for an AI system: unacceptable risk, high risk, minimal risk, no risk. The question then is how in practice AI-aided knowledge assessment courses would possibly be classified? More specifically, whether that would qualify as high-risk or not?
It would be crucial here if AI-aided knowledge assessment courses fall under the critical areas and use cases considered to be high-risk. From the point of view of energy sector, it is decisive whether or not an AI system intended to be used as a safety component in the management and operation of the supply of electricity, and it is definitely not. From the point of view of education and employment sectors, it is critical whether or not an AI system intended to be used for the purposes of assessing in terms of access, initiation, promotion, termination, task allocation, and it is rather not. Yes, AI-aided knowledge assessment courses appear to fall under the critical areas and use cases but where an AI component used just for generating test assignments, that is not a safety component, is not directly involved in the supply of electricity, is not used for the purposes of assessing, and therefore should not qualify as high-risk.
One way or another, that kind of the supranational layer incarnation is hypothetical because it is still unknown whether and when Ukraine would be granted the EU member status, what would be the final wording of the AI Act and its implementation requirements, what would implementation the Roadmap for the Regulation of Artificial Intelligence look like.
The international layer is also of hypothetical relevance for Ukraine for now. There is nothing at this level of regulatory significance. But there are at least two important factors here to remember. First, AI is global in essence, is not physical borders or jurisdiction dependent, and it requires global solutions which are international treaties or equivalents. Second, Ukraine is not like other jurisdictions in terms of how international treaties work. It is much more common that legal consequences of an international treaty in force for a nation state are just its obligations to implement that at the national level in the national legislation. It is however not like that for Ukraine, in Ukraine an international treaty in force is directly applicable, legally binding without any action on the part of the state in terms of implementation into the national legislation, and takes priority over provisions of the national legislation. That is why the international layer would matter a lot. The question is what and when would it be, and how adequate would be solutions at the international level. Look at the International Atomic Energy Agency. What could be more important than nuclear field and what could be more powerful than this organisation! And still, following its publication on AI accelerating nuclear applications, science and technology it is evident that Generative AI technologies went unnoticed just before those exploded, and those are beyond agenda of the technical meeting on the safety implications of the use of AI in nuclear power plants that takes place these days. That appears to be anything but adequate.
Considering the regulatory framework in its key elements brings us to conclusion that it is silent about AI technologies. Which means that AI-aided knowledge assessment courses are not restricted, at the very least where AI-based technologies are used for generation of test assignments, even for energy sector staff. Exactly the way it is not restricted that a human as an expert uses a computer and word processing software for test assignment generation, even where the latter provides spelling and grammar autocorrection or predictive text input technologies. A human still remains in the loop.
