For the purposes of this Regulation the notion of publicly accessible space should be understood as referring to any physical place that is accessible to the public, irrespective of whether the place in question is privately or publicly owned and regardless of the potential capacity restrictions. (Recital 9)
The AI Act regulations may apply largely to AI activities conducted in physical spaces that are accessible to the public. However, the act does not explicitly state its jurisdiction over AI activities that are carried out within private environments like offices and homes. As such, AI models developed solely for research purposes within a controlled private environment may potentially fall outside the scope of this regulation.
Online spaces are not covered either, as they are not physical spaces. (Recital 9)
This implies that online spaces may not be included within the purview of the AI Act, which primarily concerns AI activities within physically accessible public spaces. Consequently, when models are made available for public usage on digital platforms like HuggingFace, the AI Act may not be directly applicable. Thus, the personal responsibility of researchers in such a context may not be specifically defined under this regulation.
Please note, this conclusion is derived mainly from the interpretations of Recital 9. Other parts of the AI Act covering research, development activities, prohibited AI practices, transparency obligations, as well as definitions of high-risk AI systems, might provide additional context or present alternative perspectives that could impact the understanding of AI researcher responsibilities under different scenarios for AI model deployment.
Remember that the act has complex legal implications and is subject to evolution and further refinement in its interpretation. Therefore, for a full and legally definitive understanding of specific obligations and responsibilities, it would be advisable to seek guidance in consultation with a legal expert.
This Regulation should help in supporting research and innovation and should not undermine research and development activity and respect freedom of scientific research. It is therefore necessary to exclude from its scope AI systems specifically developed for the sole purpose of scientific research and development. (Recital 2f)
This part of the AI Act explicitly excludes AI systems intended solely for scientific research and development from its scope. This suggests that models developed purely for research purposes are indeed regulated differently than commercial ones. Herein, the regulation recognizes the importance of not hampering research and innovation in the AI space.
and to ensure that the Regulation does not otherwise affect scientific research and development activity on AI systems. Under all circumstances, any research and development activity should be carried out in accordance with the Charter, Union law as well as the national law; (Recital 2f)
Moreover, the Act expresses the intent to ensure that the research and development activities on AI systems are not adversely affected by the regulation. However, it clarifies that any such R&D activities must comply with the Charter, Union law, and national law. While it does not explicitly specify the responsibilities if a researcher’s model is used publicly (like on a model hosting platform), it emphasizes the need for all activities to adhere to legal norms.
For the purposes of this Regulation the notion of publicly accessible space should be understood as referring to any physical place that is accessible to the public, irrespective of whether the place in question is privately or publicly owned and regardless of the potential capacity restrictions. Online spaces are not covered either, as they are not physical spaces… (Recital 9)
While Recital 9 does not directly provide implications on the issue of a research model being used publicly on a service like HuggingFace, it lays out the definition of “publicly accessible space” within the context of the regulation. It expressly states that online spaces are not considered as such because they are not physical spaces. This could potentially impact how the regulation is applied when AI models are used in digital environments.
“This Regulation shall not apply to research, testing and development activities regarding an AI system prior to this system being placed on the market or put into service, provided that these activities are conducted respecting fundamental rights and the applicable Union law. The testing in real world conditions shall not be covered by this exemption.” (Article 2, Paragraph 9)
Based on this quote, it appears that models developed purely for research purposes are generally not subject to the rules of the AI Act. This exemption applies provided that the research activities respect fundamental rights and applicable Union law. Note that the exemption does not cover “real-world” testing conditions, suggesting that if a research model is utilized outside of controlled testing, the regulation might apply.
”Any distributor, importer, deployer or other third-party shall be considered a provider of a high-risk AI system for the purposes of this Regulation and shall be subject to the obligations of the provider under Article 16, in any of the following circumstances: (a) they put their name or trademark on a high-risk AI system already placed on the market or put into service; (b) they make a substantial modification to a high-risk AI system that has already been placed on the market or has already been put into service and in a way that it remains a high-risk AI system.” (Article 28, Paragraph 1)
In the situation where a research model is taken and deployed publicly on a service like HuggingFace, evidently, the entity or individual deploying the model will be seen as a “provider” under the AI Act and will be subject to the obligations set out in the regulation. This applies whether their use of the model involves commercial motives or not.
”The provider of a high-risk AI system and the third party that supplies tools, services, components or processes that are used or integrated in the high-risk AI system shall, by written agreement specify the information, capabilities, technical access, and or other assistance, based on the generally acknowledged state of the art, that the third party is required to provide in order to enable the provider of the high risk AI system to fully comply with the obligations under this Regulation.” (Article 28, Paragraph 2a)
The responsibility of researchers in the event that their research model is deployed by another party seems to be influenced by the agreements made between the researcher and the deploying party. The researcher might need to provide information, capabilities, or assistance to allow the deploying party to comply with the regulatory obligations. Therefore, researchers should ensure clear contractual terms defining the obligations of both parties.
”The developers of free and open-source AI components should not be mandated under this Regulation to comply with requirements targeting the AI value chain and, in particular, not towards the provider that has used that free and open-source AI component.” (Recital 12c)
This indicates that AI components developed for research purposes (which often employ free and open-source licenses) are not subject to the same restrictions as commercial AI components under this Regulation. This offers some relief to researchers regarding their direct responsibilities towards end users, including providers that might use their research models.
”Developers of free and open-source AI components should however be encouraged to implement widely adopted documentation practices, such as model and data cards, as a way to accelerate information sharing along the AI value chain, allowing the promotion of trustworthy AI systems in the Union.” (Recital 12c)
While not legally obligated, researchers are encouraged to follow best practices for documenting their AI models to promote their appropriate and efficient use.
The AI systems specifically referred to under points 1 to 8a stand for critical use cases and are each considered to be high-risk AI systems pursuant to Article 6(2), provided that they fulfil the criteria set out in that Article. (Annex III)
Based on the information from Annex III, if a model falls under one of the listed high-risk categories and meets the associated criteria, it may be subject to additional regulation. There’s no explicit mention of differentiating these categories based on research or commercial use.
All interpretations stated above should be cross-verified with a legal professional to understand the full scope and implications of the EU AI Act.
This Regulation should help in supporting research and innovation and should not undermine research and development activity and respect freedom of scientific research. It is therefore necessary to exclude from its scope AI systems specifically developed for the sole purpose of scientific research and development and to ensure that the Regulation does not otherwise affect scientific research and development activity on AI systems. (Recital 2f)
The words “specifically developed for the sole purpose of scientific research and development” denote that AI systems that are exclusively built for research purposes aren’t within the regulation’s scope. It means that these systems are not regulated in the same way as commercial models.
However, the caveat is the term “sole purpose”. It suggests that if a model is created for research but later applied in a public service like HuggingFace, its categorization might change and come under scrutiny of the regulation.
Under all circumstances, any research and development activity should be carried out in accordance with the Charter, Union law as well as the national law; (Recital 2f)
This statement establishes that while research models may be excluded from the regulation, irrespective of their final application or deployment, they are still expected to adhere to the Charter, Union law, and national laws. While the Act doesn’t appear to directly make researchers responsible for subsequent uses of their models, this statement suggests that they must ensure their models are compliant with these laws. Still, they would likely need to track the use and abuse of their models in other settings to safeguard compliance.