The difficulty of defining AI
One of the most debated definitions in the current AI Act has been the definition of the AI systems. The policy makers agreed to disagree multiple times on the matter.
For example, the EU Commission opined that AI is a type of technology capable to display intelligent behaviour by analysing environment and taking actions, with some degree of autonomy, to achieve specific goals. The EU Parliament defined AI as the ability of a machine to display human-like capabilities such as reasoning, learning, planning and creativity. The Council of Europe advanced that an AI system is a machine-based system that for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that may influence physical or virtual environments. Different AI systems vary in their levels of autonomy and adaptiveness after deployment.
The above are just few of the attempts to define AI. For conciseness reasons we will not provide all of them.
AI Act definition and critique
The AI Act definition leverages the revised 2019 OECD Recommendation on AI (the “AI Principles”) and has been qualified by some researchers as being surprisingly broad and generic. The source is Article 3(1) of the AI Act which defines the AI system as a machine-based system
- with varying degrees of autonomy and that may show adaptiveness after deployment
- that infers
- from the input it receives
- how to generate outputs such as predictions, content, recommendations or decisions
- that can influence physical or virtual environments.
The definition is basically encompassing not only sophisticated systems capable of learning, adapting and making independent decisions but also simplified machine learning algorithms (e.g., regression models) executing specific tasks.
What the AI Act is leaving outside of the remit of the AI system definition is the more rudimentary rule-based systems (i.e., an issue flagging system for the user/administrator to correct) underpinned by automation.
As the European Parliamentary Research Service observes in its most recent study, while comprehensive and hence reasonably future-proof, this broadness of the AI system definition poses challenges, particularly in distinguishing between systems that should be considered AI and those that should not.
Automated versus autonomous software
The terms autonomy and automation are often confused. Nevertheless, these terms are not synonymous, but each has a distinctive meaning.
An automated system is instructed to perform repetitive tasks with predefined parameters. A typical example is the network intrusion detection software which monitor network traffic for suspicious activity, violations of security policies, and generate alerts if it detects malicious activity.
An autonomous system, on the other hand, learns from input data, adapts to dynamic environments, and evolves as the environment around changes. Typical example is a map navigating system such as Google Maps incorporating various AI technologies based on route optimization, predictive analytics, image recognition, personalization and natural language processing.
Inferring the output from the input
Unlike the automated systems based on rigid instructions, the autonomous systems are working through inferring outputs from the input received. AI inference is the second stage in a two-part machine learning process, where a trained machine learning model applies its knowledge to previously unseen data. This ability to apply the acquired knowledge from learned data to new scenarios without writing hard-code algorithms is what makes AI inference unique. Two approaches to inferring are mentioned by the AI Act, specifically machine-learning and knowledge-based approaches.
Systems such as chatbots, which process and generate natural language, diffusion-based image generators, or advanced facial recognition technologies that can identify individuals, are inferring from learned data.
AI Act Self-Assessment Questionnaire
When navigating the complexities of the AI Act definition of AI system we advise organisations to start with an AI Act Self-Assessment Questionnaire. You will know form the beginning if your IT systems fall under the definition of the AI system and if that is the case, what steps you need to take to comply with the AI Act.
We devised such a Questionnaire in 8 steps, and we aim to launch it soon on our website and in our Data Protection Management Tool. By completing all the sections of the Questionnaire, you will understand if the AI Act is indeed applicable to you or not.
Symmetry Compliance offers comprehensive services to support you at every stage of your AI development journey, ensuring your projects are both innovative and fully compliant with EU regulations. We develop compliance framework to guide your AI model development, focusing on non-discrimination and protecting fundamental rights to ensure your AI systems are ethical, fair, transparent, and aligned with the EU AI Act.
Ready to Make Your AI Projects are Compliant and Successful?
Contact Symmetry Compliance today to learn more about how we can help your business navigate the complexities of the AI Development Lifecycle under the EU AI Act. Let’s work together to build AI solutions that are safe, ethical, transparent, and compliant.
