An update on our dedication to secure, ethical generative AI advancement | AWS Machine Learning Blog

Responsible AI at Amazon

Responsible AI is a longstanding commitment at Amazon, with a focus on safety, fairness, robustness, security, and privacy. The company has embedded these principles into its development processes and educational initiatives for employees.

Tools and Offerings for Responsible AI

Amazon has developed over 70 internal and external tools to support responsible AI practices. The company has published or funded over 500 research papers on responsible AI and delivered tens of thousands of hours of training to its employees. Amazon also offers free responsible AI training courses for people of all ages and backgrounds.

Guardrails and Model Evaluation

In April 2024, Amazon introduced Guardrails for Amazon Bedrock and Model Evaluation in Amazon Bedrock to enhance safety, security, and transparency in AI services. These tools make it easier for customers to implement safeguards and evaluate models effectively.

AI Service Cards and Transparent Generative AI

In May, Amazon published AI Service Cards for Amazon Titan Text Premier to provide customers with information on use cases, limitations, and design choices for AI services. These cards aim to promote transparency and responsible AI practices in generative AI applications.

Testing and Security Measures

Amazon employs red-teaming and other testing techniques to evaluate the safety, security, and fairness of AI systems like Titan Text. The company uses proprietary evaluation datasets to assess performance across various use cases and prompts.

Combatting Disinformation with Watermarking

To address concerns about AI-generated content and disinformation, Amazon’s Titan Image Generator includes invisible watermarks by default. The company introduced an API in Amazon Bedrock to verify if an image was generated by Titan Image Generator, enhancing transparency and combating disinformation.

Collaboration for Trustworthy AI

Amazon collaborates with organizations like the U.S. Artificial Intelligence Safety Institute Consortium to establish measurement standards for trustworthy AI. The company also works with global partners to combat deceptive AI use, protect elections, and address issues like child exploitation.

Promoting Responsible AI Use

Amazon is committed to promoting safe and responsible AI development through education and training initiatives. The company offers free courses on responsible AI practices to drive innovation while mitigating risks.

Examples of Generative AI in Action

Amazon highlights examples of generative AI applications across industries, such as Brainbox AI’s virtual building assistant for energy optimization and Gilead’s use of AI in drug development. These examples showcase the potential of AI technology to drive positive change.

Continued Commitment to Responsible Innovation

As a pioneer in AI technology, Amazon remains dedicated to fostering safe, responsible, and trustworthy AI development. The company emphasizes collaboration with stakeholders to deliver innovative AI solutions with trust at the forefront.


Comments

Leave a Reply

Your email address will not be published. Required fields are marked *