Edit

Share via

Responsible AI considerations for intelligent application workloads

  • Article

Intelligent application workloads must adhere to responsible AI principles to ensure fairness, accountability, transparency, and ethical behavior.

Design the AI system to treat all users fairly, hold developers and users accountable for its performance, ensure transparency in AI operations, and adhere to ethical standards.

At Microsoft, we're committed to the advancement of AI driven by principles that put people first. Generative models have significant potential benefits, but without careful design and thoughtful mitigations, such models have the potential to generate incorrect or even harmful content. Microsoft has made significant investments to help guard against abuse and unintended harm, which includes incorporating Microsoft's principles for responsible AI use, adopting a Code of Conduct, building content filters to support customers, and providing responsible AI information and guidance that customers should consider when using generative AI.

Power Platform copilots and generative AI features follow a set of core security and privacy practices and the Microsoft Responsible AI Standard. Power Platform data is protected by comprehensive, industry-leading compliance, security, and privacy controls.

Learn more:

Core principles of responsible AI

Responsible AI core principles include fairness, accountability, transparency, and ethics. Ensuring that an intelligent application workload built with Power Platform adheres to these core principles involves several key practices:

  • Fairness: Use diverse and representative training data to minimize biases. Regularly update training data and enlist auditors to validate fairness and equity.
  • Accountability: Define clear roles and responsibilities for team members involved in the AI project. Establish and adhere to ethical standards that prioritize fairness and accountability.
  • Transparency: Ensure that users know they're using a workload that uses generative AI capabilities. Clearly communicate why an AI solution was chosen, how it was designed, and how it's monitored and updated.
  • Ethics: Foster an inclusive workforce and seek input from diverse communities early in the development process. Regularly assess and test models for ethical concerns and disparities in performance. Establish a governance framework that includes regular audits.

Incorporate these practices into your development and deployment processes to create an intelligent application workload that adheres to the core principles of responsible AI.

Data privacy and security

Ensuring data privacy is critical, especially as the intelligent application workload may handle sensitive data. When planning an intelligent application workload with Power Platform, it's essential to address several key risks and implement effective mitigation strategies:

  • Platform features: Understand native controls and platform features that safeguard your data. Microsoft Copilot is built on the Microsoft Azure OpenAI Service and runs completely within the Azure cloud. Copilot uses OpenAI models with all the security capabilities of Microsoft Azure. Copilot is integrated in Microsoft services like Dynamics 365 and Power Platform and inherits their security, privacy, and compliance policies and processes, such as multifactor authentication and compliance boundaries.
  • Data encryption: Service-side technologies encrypt organizational content at rest and in transit for robust security. Connections are safeguarded with Transport Layer Security (TLS), and data transfers between Dynamics 365, Power Platform, and Azure OpenAI take place over the Microsoft backbone network, ensuring both reliability and safety. Learn more about encryption in the Microsoft cloud.
  • Access controls: Data is provided to Copilot based on the access level of the current user. Implement role-based access control (RBAC) using Microsoft Entra ID to ensure that only authorized users can access data. Apply the principle of least privilege to limit access to only what is necessary.
  • Monitoring and auditing: Detect and respond to potential security incidents by regularly monitoring access and usage of the AI system. Maintain detailed audit logs to track data access and modifications.
  • Compliance and governance: Ensure compliance with relevant data privacy regulations such as GDPR (General Data Protection Regulation), HIPAA (Health Insurance Portability and Accountability Act), and CCPA (California Consumer Privacy Act). Implement ethical AI practices to avoid biases and ensure fairness in AI outputs.
  • User education and raining: Train users on security best practices and the importance of data privacy. Keep users informed about updates and changes to security policies and procedures.

Learn more: FAQ for Copilot data security and privacy for Dynamics 365 and Power Platform

Bias awareness and mitigation

Recognize the importance of addressing biases in the system and ensure fairness to avoid biases in AI responses.

  • Diverse and representative data: Ensure that the training data is diverse and representative of different demographics to minimize inherent biases. Regularly audit the data for biases and imbalances, and take corrective actions as needed.
  • Bias detection and mitigation tools: Use tools and techniques to detect biases in the AI models, such as statistical analysis and fairness metrics. Implement debiasing techniques, including resampling, reweighting, or adversarial debiasing, to reduce biases in the models.
  • Human-in-the-loop: Incorporate human review and feedback loops to identify and correct biases that the AI might introduce. Establish an ethics committee or governance board to oversee AI development and deployment, ensuring ethical standards are met.
  • Transparency and trust: Ensure users know they're using a workload that uses generative AI capabilities. Clearly communicate why an AI solution was chosen and provide information on how it was designed and how it's monitored and updated.
  • Continuous monitoring and improvement: Continuously monitor the AI system for biases and performance issues, and update the models as needed. Ensure models remain fair and unbiased by regularly retraining models with updated and more diverse data.

Ongoing monitoring and evaluation

Continue improving your intelligent application workload. Establish a framework for continuous monitoring and evaluation, and incorporate user feedback and evolving ethical standards into updates.

  • Feedback loops: Establish feedback mechanisms where users can report inaccuracies, which can then be used to refine and improve the models.
  • Monitoring and auditing: Detect and respond to potential security incidents by regularly monitoring access and usage of the AI system. Maintain detailed audit logs to track data access and modifications.