Recent developments in technology have positioned artificial intelligence (AI) systems as a central force driving innovation across industries. As AI systems become increasingly integrated into organizations, it’s essential to align them with societal values and expectations while minimizing the risk of negative consequences. Responsible AI Principles aims to develop, deploy, and assess AI systems in ways that prioritize values such as; transparency, safety, non-discrimination, fairness, and data privacy. Therefore AI systems can enhance human capabilities while safeguarding against potential harm.
Seven Principles of AI
On 8 April 2019, the High-Level Expert Group on AI, established by the European Commision in 2018, released Ethics Guidelines for Trustworthy Artificial Intelligence. According to the Guideline, AI systems must meet seven fundamental requirements to be trustworthy.
1- Human Agency and Oversight
AI systems should support individualsto make informed decisions while supporting human agency and rights, without limiting autonomy. Moreover, human oversight is crucial to prevent AI systems, from undermining autonomy or causing harm. Therefore, robust oversight mechanisms must be implemented. These can be achieved through human-in-the-loop, human-on-the-loop, and human-in-command approaches
Read more on this principle in our previous article: The Importance of Human Autonomy in AI Design
2- Technical Robustness and Safety
Developers must design their AI systems to be secure, resilient, and safe. The principle of technical robustness and safety requires AI systems to be designed with a preventive approach to manage risks and minimise unintentional and unexpected harm. In order to achieve safety, AI systems should have fallback options and reproducibility in case of failure. Additionally, they should be able to demonstrate reliability, accuracy, and reproducibility.
3- Privacy and Data Governance
AI Systems must fully respect privacy and data protection throughout their entire lifecycle.To develop and deploy trustworthy AI, it’s essential to obtain data from reliable sources, process it lawfully, and use it responsibly. It is also important to keep data secure and be transparent about its usage. Organisations must implement adequate data governance mechanisms, considering the quality and integrity of the data.
4- Transparency
Transparency refers to understanding how AI system functions and makes decisions,which helps stakeholders to trust the system’s outputs. This principle consists of three key elements: traceability, explainability, and open communication about system limitations. In order to achieve Trustworthy AI, ensuring all of them is important.
AI systems should be able to provide clear, understandable explanations for their decisions. For example, in a recruitment process, if a candidate inquires why they were shortlisted for a particular role, a transparent AI system might explain that the selection was based on the candidate’s qualifications and relevant work experience, or that the recommendation was made due to their strong performance in skills assessments related to the role.
Additionally, humans should be aware when they are interacting with an AI system. Therefore they always must be informed before using the AI system, including including its capabilities and limitations.
5- Diversity, Non-discrimination and Fairness
AI systems should be designed and operated in a way that respects the dignity, rights, and freedoms of all individuals. ‘Fairness by design’ refers to the integration of this principle into AI systems, to promote inclusion and diversity throughout the entire lifecycle. The bias of the AI system’s output can arise from several sources, such as training data, the algorithms, or the interpretation of the outputs. Algorithmic bias can lead to discrimination against vulnerable groups based on gender, race, or disabilities.
Developers must ensure to remove any identifiable biases during the data collection stage and train their AI systems with diverse datasets to prevent unfair outcomes.
Additionally, strategies should be put in place to mitigate, identify, and correct biases in data, algorithms, and interpretation processes. In case of biased outcomes, reparation strategies should be implemented to rectify unfair results and prevent future biases.
6- Environmental and Societal Well-being
AI systems should benefit all human beings, including future generations. It must hence be ensured that they are sustainable and environmentally friendly. Moreover, they should take into account the environment, including other living beings, and their social and societal impact should be carefully considered.
7- Accountability
The accountability principle ensures actors of the AI systems are responsible for the proper functioning of AI systems and respecting all the trustworthy AI principles. This requires robust mechanisms to ensure traceability throughout the AI system’s lifecycle. It should also allow analysis of the AI system’s outputs and responses to inquiry, in a manner that is contextually appropriate and aligned with current best practice.
Organisations must implement effective risk management frameworks to identify and address risks transparently. Moreover, organisations should conduct regular audits of AI systems to identify and eliminate biases, ensure fair and nondiscriminatory outcomes, and foster transparency in AI.
How can Symmetry Compliance help?
At Symmetry Compliance we assist our clients to comply with the AI Act. We have solid experience in operationalising organisation’s compliance with data protection and governance legislation. For AI compliance, we set up a dedicated team ready to assist with a diversity of queries and assignments.
Contact us for more information and guidance.
Image: generated by Microsoft Bing
