​The rapid advancement of generative Artificial Intelligence (AI) presents immense opportunities but also significant ethical challenges. In this article we explore the ethical issues surrounding generative AI and introduce SAP’s approach to ensuring safety and responsibility in a trustworthy AI environment.

Ethical Challenges of Generative AI

Generative AI systems, such as Large Language Models (LLMs), have the potential to impact individuals and the environment both positively and negatively. Key ethical challenges include:

• Damage Prevention: It is essential to prevent harm to individuals and the environment. This requires a strong focus on AI safety and ethical principles.
• Human Control: Humans must retain control over AI systems, not the other way around. Often, the data and neural networks used are difficult to control, which can undermine trust in the results.
• Fairness and Non-Discrimination: Poor dataset management can lead to biases and discrimination, especially when certain groups are systematically disadvantaged. SAP conducts extensive bias testing to ensure that training data, test datasets, and models are free from prejudice and discrimination.
• Transparency and Explainability: To effectively manage AI systems, their functioning must be understandable. This requires not only technical knowledge but also an in-depth understanding of the training data and the internal mechanisms of the models, which is often hindered by untransparent data.
• Data Privacy and Management: Training data may contain sensitive information that models could memorize and potentially misuse. Another risk involves the input data, as the model should not learn from this data. SAP implements strict measures for data anonymization and pseudonymization to ensure that sensitive information remains protected and no security risks arise.

Approaches to Addressing Ethical Challenges

To tackle these challenges of using Generative AI, SAP adopts a comprehensive approach encompassing multiple levels:

• Ethical standards and engineering excellence: Adhering to strict ethical standards and excellent engineering practices is fundamental to developing trustworthy AI systems. SAP aligns with international guidelines and best practices.
• EU AI Act and responsible AI regulation: The EU AI Act provides an important framework for regulating responsible AI. SAP integrates these regulatory requirements into its development processes to ensure that AI solutions comply with legal standards.
• External consulting and ethics advisory board: By involving external experts and ethics advisory boards, SAP ensures an objective and well-founded evaluation of the ethical aspects of its AI systems.
• UNESCO recommendations and guiding principles: UNESCO’s recommendations serve as additional guidance to integrate global ethical standards into AI development.

SAP’s Ethics Assessment Process

SAP has established a structured ethics assessment process, detailed in the AI Ethics Handbook. This process includes:

• Redline Cases: Identifying scenarios that could potentially cause harm to the environment or personal freedom.
• Requirements for responsible and trustworthy AI: Defining criteria to ensure that AI systems meet ethical standards.

Developing Safe and Responsible Generative AI

SAP pursues a holistic approach to developing safe and responsible generative AI. This includes implementing orchestration services that provide additional security and responsibility features.

Orchestration Service in the Generative AI Hub

The orchestration service, part of the Generative AI Hub, enables companies to configure generative AI quickly and securely. This service includes:

• Grounding: Adding additional information to the query process to reduce hallucinations and increase response accuracy. In this context, hallucinations refer to the AI's creation of “facts.” If the system does not know an appropriate answer, it is supposed to transparently indicate this and avoid unfounded assumptions or speculations.
• Data Masking: Masking identifiable information such as names, dates, or sensitive user information to prevent biases. For example, traditional AI systems may favor male users, which can be disadvantageous in recruitment processes.
• Filtering: Automatically filtering out harmful or unethical content (both in input prompts and outputs).

Conclusion: The Future of Safe and Responsible AI with SAP

The development of generative AI offers tremendous opportunities but also significant ethical challenges. With a comprehensive approach that combines ethical standards, regulatory requirements, and innovative technical solutions, SAP sets benchmarks for safe and responsible AI. The orchestration service in the Generative AI Hub and strict ethics guidelines ensure that SAP’s AI solutions are trustworthy, fair, and transparent.