Expanding ethical considerations in foundational models

The rise of generative AI and the foundational models that fuel the growth of other AI use cases offers exciting possibilities, but it also raises new questions and concerns about ethical design, development, deployment, and use.

The IBM AI Ethics Committee publication Foundation Model: Opportunities, Risks, and Mitigation addresses these challenges and explores the benefits, risks, guardrails, and mitigations of the technology.

This white paper describes the potential risks associated with the underlying model through the lens of three categories: ethical, legal, and regulatory.

1. traditional. Known risks from older or older forms of AI systems.
2. Amplification. The now known risks are further enhanced by the intrinsic properties of the underlying model, especially its unique generative function.
3. new. New risks inherent in the underlying model and its unique generative capabilities.

These risks are structured according to whether they are associated with the content provided to the underlying model (input) or the content generated (output), or with additional challenges. These risks are presented in a table that highlights why they are a concern and why it is important to consider them during the development, release and use of the underlying model.

The white paper also highlights some of the available mitigation strategies and tools, including the watsonx enterprise data and AI platform and trusted open source AI tools. These strategies focus on balancing safety and innovation and enabling users to experience the power of AI and underlying models.

Foundational Models: Opportunities, Risks, and Mitigation takes you on a journey to realize the potential of foundational models, understand the significance of the risks they may pose, and learn about strategies to mitigate potential impacts.

Read the foundational model opportunities, risks and mitigations Explore the AI ​​Risk Atlas and other watsonx product documents Learn more about AI ethics at IBM