AI Ethics & Policy Insights

Week of October 16 , 2023. Issue 12.

Paige

Oct 16, 2023

Week in review:

EU & JAPAN: Of approaches to AI standards between the EU and Japan, the European Commission Vice-President for Values and Transparency Vera Jourova told Reuters, “I see a lot of convergence in how we look at AI and generative AI.” While nothing has happened to indicate that the two industrial powers are actively working together, this signals a potential alliance in the G7 between France, Germany, and Italy - all EU states - and Japan as the group works to identify standards and frameworks for generative AI. Jourova also indicated that Japan has a deep knowledge of the technology that surpasses that of China.
TIKTOK & AI: The age of AI-enabled deception is upon us, and companies and news organizations are taking note. An AI-generated (fake) video of Barack Obama addressing the death of his former chef, in a voice that passes as his own, was discovered by NewsGuard, a company that monitors misinformation on the web. These AI-powered videos are popping up more and more, and while companies have been putting policies in place requiring that they content be noted as fake or AI generated, this isn’t stopping creators of many of these videos. The account that posted the Obama video, among others, has received millions of views across its content.

Watermarking for AI

The Context:

A major concern of the generative AI era is that people will not be able to distinguish between human-created content and AI-generated content. This has led companies and individuals to try and provide a durable solution and minimize risk in AI uses - watermarking.

The Challenge:

We’ve already begun to see scenarios in which a person creates content with AI and uses it to fool people. One person submitted AI-generated art to a contest - and won. Others are using AI to generate content that will fool others, such as synthetic images of disasters, and deep-fake videos that can look and sound like the person they’re meant to imitate.

Why it Matters:

Transparency is a critical principle of responsible AI creation, use and deployment. The idea behind this principle is that people would know:
- if and when they are interacting with AI
- if and when AI is being used to make decisions that will influence their life (resume selection, credit line approval, etc.)
- if and when content has been generated by AI
AI watermarking helps maintain transparency in several of these critical cases by embedding a mark into the model, chatbot or output, which can be detected and read by a computer.
AI watermarks can be added to machine learning models to detect whether it’s been used to make a prediction, and it can also be added to the output of the models (text, images, etc.) to indicate whether it was created by AI.

Where we go from here:

This technology is new and not standardized. This means that not all ML models are using this method, and those that are using it are not using it in the same, standardized way.
In an idea world there would be a taxonomy for watermarking so that each model would have its own unique identifier that would be easy for computers to not only find, but understand which model or company the mark belonged to.
Another challenge includes what happens when the text or image in question isn’t digital anymore. If someone prints it or writes it word for word into a doc, how could you trace the digital signature?
This is a rapidly developing area as the world moves to try and combat deepfakes and general tomfoolery with AI. I’m keeping an eye on DeepMind, as they’re making major strides in this area.

AI Policy Updates

US & China: AI Military Advantage

The Biden administration is planning to take steps to broaden the restrictions for exporting AI chips and other equipment to China, according to officials.

This is an extension of the restrictions the Biden Administration rolled out a year ago, and will close loopholes left in the last rollout, but will also expand the restrictions to less advanced, but still powerful, chips.
While the goal of these actions is to keep China from using American-made equipment to gain an AI military advantage over the US, the administration is taking steps to set limits while not harming China’s economy, which could lead to escalated tensions between the world’s super powers.