Amazon has long been committed to developing responsible AI, embedding safety, fairness, security, and privacy into its processes. This commitment involves not just the creation of more than 70 tools and mechanisms supporting responsible AI, but also educating employees and the public through thousands of training hours and free courses. In the multi-faceted approach adopted by Amazon to transform responsible AI from mere theory into tangible practice, new tools, testing methods, and partnerships play an essential role.
Recently, Amazon has made progress in its responsible AI efforts by launching new tools and capabilities for building and scaling generative AI safely. In April 2024, the company introduced Guardrails for Amazon Bedrock, a unique solution offered by a major cloud provider, enabling businesses to build and personalize safety and privacy protections for generative AI applications.
Another move was the publication of the AI Service Card for Amazon Titan Text Premier, which provides a unified place for customers to find information on intended use cases, responsible AI design choices, as well as other vital details for AI services and models.
To ensure the safety, security, and other crucial aspects of its AI systems, Amazon employs different techniques like manual red-teaming, where human testers probe the AI system for flaws, and automated benchmarking against publicly available and proprietary datasets. By adopting such strategies, Amazon constantly strives to demonstrate that its AI systems are as secure, fair, and transparent as possible.
In response to the necessity for users to differentiate AI-generated visual content from human-made ones, Amazon introduced default invisible watermarks for all images created by Amazon Titan Image Generator, thus providing a measure to counter the spread of disinformation. Besides, Amazon launched an API that checks for the existence of this watermark, confirming if an image was generated by Titan Image Generator.
Further, to keep up with safety risks, Amazon collaborates with governments, other companies, and the broader AI community. It contributes to advancements in AI safety measurements and methodologies while also joining in efforts to combat deceptive AI use in elections. Two initiatives Amazon partners with include Thorn and All Tech is Human, both committed to designing AI services to mitigate exploitation risks.
Amazon’s commitment to responsible AI also entails using it to address climate change, enhance healthcare, and support various initiatives fostering education. An instance of such applications is Gilead, an American biopharmaceutical company accelerating life-saving drug development with AWS generative AI.
Moreover, the company proffers courses on potential risks surrounding generative AI systems and other critical topics. Amazon’s “AI Ready” initiative aims to provide free AI skills training to 2 million people globally by 2025.
Maintaining its dedication to driving AI innovation while ensuring necessary safeguards, Amazon is committed to promoting the safe, responsible, and trustworthy development of AI technology. The major focus remains to foster groundbreaking generative AI innovation, keeping trust at the center.
