Microsoft recently released its first-ever transparency report, outlining its efforts to advance responsible AI platforms in 2023. The report, titled “Responsible AI Transparency Report,” highlights the company’s commitment to safely deploying AI products and its progress in meeting the voluntary agreement signed with the White House in July of that year.
Throughout the report, Microsoft emphasizes its achievements in developing 30 responsible AI tools over the past year and expanding its responsible AI team. Notably, the company required teams working on generative AI applications to assess and mitigate risks throughout the development process. Additionally, Microsoft introduced Content Credentials to its image generation platforms, which watermark photos to identify them as AI-generated.
Furthermore, Microsoft extended access to Azure AI customers for tools that identify problematic content such as hate speech, sexual content, and self-harm. These tools also assist in evaluating security risks, including the detection of jailbreak attempts and indirect prompt injections in AI models’ data ingestion.
To bolster its AI safety measures, Microsoft has expanded its red-teaming efforts. This includes in-house red teams tasked with identifying and bypassing safety features in AI models, as well as third-party testing of new models before release.
Despite these efforts, Microsoft’s AI rollouts have faced challenges and controversies. For instance, when Bing AI was introduced in February 2023, users discovered instances of the chatbot disseminating incorrect information and inappropriate content. Additionally, loopholes in the Bing image generator allowed users to create insensitive images, prompting Microsoft to take action to prevent further misuse.
Natasha Crampton, Microsoft’s chief responsible AI officer, acknowledges that responsible AI is an ongoing endeavor with no definitive endpoint. She underscores Microsoft’s commitment to continuous improvement and building on its progress in the coming years.
In summary, Microsoft’s inaugural transparency report sheds light on its strides in promoting responsible AI practices while acknowledging the challenges and areas for improvement in AI deployment.
