Realizing trustworthy AI through data

The use of artificial intelligence (AI) is rapidly increasing and its impact on our everyday lives can’t be ignored. However, with great power comes great responsibility – making it crucial to develop and use AI in a responsible, trustworthy manner.

• Ensure AI training data is representative and diverse: Include a wide range of data sources that accurately represent the diverse user groups that will interact with your AI system to avoid biased or discriminatory outcomes.
• Implement data privacy and security measures: Protecting data, especially sensitive data, is paramount. Implement robust security measures to safeguard personal information and ensure compliance with data protection regulations.
• Provide data transparency: Be transparent about your data collection process to establish trust and credibility in your AI. Users should clearly understand how their data is collected, used and stored.
• Implement data governance: Develop and put into practice data governance policies and procedures that provide a framework for data management. They should cover data quality, access, retention and sharing, to maintain data consistency, accuracy and integrity.
• Establish mechanisms to address data incidents: Implement a clear and well-defined process to effectively handle data breaches and incidents, including incident response plans, communication protocols and procedures for notifying affected individuals or authorities.
• Educate employees: Provide adequate employee training on data privacy, ethics and responsible data handling practices to ensure everybody understands their role and responsibilities in protecting data.
• Engage with stakeholders: Regularly gather feedback from individuals whose data is being processed – including customers and relevant communities – to address their concerns, incorporate diverse perspectives into data practices and promote transparency, accountability and responsiveness in data handling.
• Continuously monitor and evaluate data practices: Conduct regular data audits and assessments to identify emerging data risks, address evolving data challenges and ensure ongoing compliance with ethical and regulatory standards, including staying up to date on emerging technologies, best practices and regulations.
• Incorporate ethical principles during development: Integrate ethical considerations such as fairness, non-discrimination and human oversight into your development process to avoid perpetuating biases or discriminatory outcomes and ensure the responsible and ethical use of data.

• Web scraping: While the automated scraping of data from the internet can be useful for quickly gathering large quantities of data, it often bypasses consent or permissions – directly conflicting with data privacy and copyright regulations. 
• Social media datasets: The use of data from social media platforms to train AI poses significant privacy and consent issues. Those platforms often house personal and sensitive information, therefore using social media datasets without having secured explicit, informed consent from social media users can violate privacy laws and damage trust.
• Off-the-shelf datasets: AI developers must exercise due diligence, conduct thorough audits and understand how off-the-shelf datasets were collected and processed to ensure compliance with AI regulations.
• Explainability: Under the EU AI Act, users have a right to understand how a decision was made by an AI system, posing a challenge for existing complex, black box machine learning models. Ensuring interpretability and transparency during AI development, data preparation, training and deployment is no longer optional.