## The Evolution of Standardization in Artificial Intelligence: A Holistic Approach

In recent years, the landscape of standardization for artificial intelligence (AI) has undergone significant transformations, increasingly leaning towards a holistic approach that encompasses the entire lifecycle of AI systems. This shift is characterized by a profound integration with risk management principles, which are essential for ensuring the safety and reliability of emerging technologies.

Initially, early standards focused on rather specific aspects, such as the terminology used in the field of AI or on particular application use cases. However, as technologies evolved and the complexity of systems increased, the work of standardization committees began to expand significantly. Today, there is an effort to establish a continuum of standards that accompany AI systems from conception to decommissioning, thereby ensuring a more effective and integrated risk management approach.

This cyclical approach includes several crucial phases. We start with the conception and design phase, which lays the groundwork for the creation of the system. This is followed by data collection and preparation, a fundamental step that must be executed with care to avoid bias and misunderstandings. The training and testing of the model require careful supervision: it is in this phase that the system begins to “learn.” Once implemented, the model must be constantly monitored to ensure it operates correctly; any anomalies or inaccuracies must be identified and corrected promptly. Finally, we reach the decommissioning stage, where the system may be turned off or replaced, and the data handled appropriately.

A fundamental concept that runs through all these phases is integrated risk management. Recent regulations, such as those suggested by specific ISO/IEC standards, highlight the importance of considering risk assessment and management not as a one-off activity, but as a continuous and iterative process. It is essential that risks—be they technical, ethical, social, or legal—are identified and mitigated at each stage of the AI system’s lifecycle.

For instance, biases in AI models must be addressed from the data collection phase, as unfair or unrepresentative data gathering can lead to distorted and unreliable outcomes. Even during the model training phase, it is crucial to maintain vigilance and ongoing correction. Lastly, monitoring the model’s performance once active is vital to ensure compliance with the highest standards of quality and reliability.

Adopting a lifecycle-based approach and risk analysis is not just good practice; it is imperative for building AI systems that can be considered “trustworthy.” This trustworthiness should not be an afterthought but should be integrated from the outset into the very fabric of the system. Only then can technologies be developed that are robust, resilient, and aligned with human values.

In conclusion, the standardization of artificial intelligence is rapidly evolving, and the direction we are heading is increasingly focused on ensuring the safety and reliability of systems. This is a journey that requires ongoing commitment from all stakeholders involved, from design to decommissioning. To stay updated on this topic and other developments related to technological innovation, we invite you to follow us on our social media profiles. Your participation is important in building a more aware and responsible future for technology use.