Standards for AI: A Recap of the AI Workshop for the SMPTE Standards Community

Many people still consider Artificial Intelligence (AI) to be a lawless territory, a technology that could destroy us if we aren’t careful with it. However, AI is simply a tool. A rapidly evolving tool, but a tool, nonetheless. As a Standards Development Organization (SDO), SMPTE is obligated to take on the interoperability challenges presented by AI in the media and entertainment industry and work together with other industry organizations to ensure a trustworthy, viable eco-system exists so AI can flourish in a safe way.

During the June Technology Committee meetings, a special workshop was held regarding AI Standards in media. SMPTE worked with the leaders in AI from ISO/IEC and C2PA to discuss opportunities for standardization regarding AI in Media. A true meeting of industry experts, this workshop shined a bright light on the future of an innovative technology and how it relates to media and entertainment.

SMPTE Taskforce on AI in Media

The workshop began with an update from the SMPTE Taskforce on AI in Media. Earlier this year, the Taskforce released an engineering report detailing the need for Standardization in AI for media applications. Done in conjuncture with the Entertainment Technology Center (ETC) and with input from the European Broadcasting Union (EBU), this report detailed what AI is, the issues facing it, and how Standardization would help regulate the entire industry.

Taskforce Chair, Fred Walls, lead this section of the workshop, and explained how the Taskforce Report came together and what opportunities for standards in AI exist. Whether people like it or not, AI will be a big factor in the media industry. He also outlined the challenges of AI specifically the complex, expensive, rapidly evolving, and presents moral, ethical, and legal hazards.

In response to these issues, the SMPTE Taskforce surveyed the SMPTE membership and the media industry at large. It investigated the status-quo of AI innovations and looked at what interoperability challenges exist. They also reached out to many SDOs to see which ones were working on AI standards. While many technologies were being worked on, Standards were a different story. Only a handful of organizations have made significant progress finding interoperable solutions to AI issues.

The Taskforce brought together stakeholders from many companies. These companies held discussions on AI ethics, platform performance, generative AI, and so much more. These discussions led to a consensus on areas where standardization in AI would be beneficial. These areas include metadata, benchmarking, recommended practices, and cloud computing.

SMPTE is in a unique position, as we serve both users and producers of new technology. This is why it’s imperative that we be at the forefront of AI standardization. The Taskforce has already proposed several projects, so progress is being made.

ISO/IEC AI Effort

The following presentation came from an organization that is ahead of the curve when it comes to AI Standardization. With 31 published Standards, 32 active projects and with 65 national bodies involved, ISO/IEC JTC1 Sub-Committee (SC) 42 is all about making AI trustworthy while not hindering the technology’s growth. They do this by focusing on AI Standards at a foundational level.

As US Vice Chair of the Insights Committee for ISO/IEC SC 42, Lenora Zimmerman highlighted the thought processes behind each of SC 42’s new projects. All Standards from this subcommittee must address foundational structures in AI starting from development. They also take diversity and customer expectations into account to avoid the ever too prevalent problem of program biases.

Context of use is also an important factor when trying to create a Standard for AI. How do we want this technology to be used? How do we regulate it in a way that’s safe, but also flexible? The secret is creating a reliable industry ecosystem with checks and balances. ISO/IEC SC 42 has a number of working groups (WG) that create such an ecosystem including WG 2, which deals with Data and analytics, WG 3, which deals with bias and ethics, and WG 4, which deals with use cases and applications of AI.

C2PA and DMS-II

While Standards are being created to ensure AI is trustworthy, technology has already been created that prevents it being used maliciously. Andrew Janks is the Executive Chair of the Coalition for Content Provenance and Authenticity (C2PA), which developed a specification that binds provenance information in a manifest to a piece of media like images, videos, audio, documents, and even fonts. This can be done through watermarks, metadata or references to databases. He described how C2PA binds data to content in a way that’s cryptographically secure, allowing you to track the entire lifespan of a piece of content, as well as the programs used to create said content.

This means that if a piece of content was created with generative AI, C2PA will be able to detect it. This is a game changer. Embedding provenance information with C2PA will establish a greater level of trust, ensuring audiences will have access to original content, and protect said content from AI training programs that have less than clear legal status.

C2PA already has some real-world use cases, and that is why a new project was proposed to develop a standard which enables the carriage of a C2PA manifest in SMPTE’s Material Exchange Format (MXF), a container file format for professional digital video and audio applications. Once the project is approved and completed, MXF will be able to carry C2PA credentials through a metadata scheme called DMS-II. MXF files with C2PA credentials can be easily used by common content creation and productions tools, meaning creators and audiences can trust their content.

Conclusion

The workshop ended with an open discussion on AI in media. The crowd had some valuable insights to share and discussed the pros and cons of potential AI Standards. The consensus, however, was that Standards are pivotal to regulating this bold new technology. AI has opened the floodgates, and many people are afraid we’ll get swept away in its wake. However, through proper implementation, cooperation, and innovation, AI can be a productive and trustworthy technology that vastly improves the media industry.