On 24th January 2024, the United National released the Global Principles For Information Integrity to better understand and limit how “new technologies have facilitated the spread of misinformation, disinformation and hate speech by many kinds of actors at historically unprecedented volume, velocity and virality”.
The broad proposals aim to ensure transparency, accountability, and fairness from all internet stakeholders, including technology and AI companies, advertisers and the private sector, news media, civil society organisations, and member states at large.
Actors should “refrain from using, supporting or amplifying disinformation and hate speech for any purpose”, with a particular focus on “crisis response” to “support information integrity around elections”. Key to this will be advertisers, who “should demand transparency in digital advertising processes from the tech sector to help ensure that ad budgets do not inadvertently fund disinformation”.
Likewise, media platforms bear responsibility for protecting their users, especially children and any other vulnerable or marginalised groups, from exposure to harmful AI-generated or mediated content.
Pavel Goldman-Kalaydin, Head of AI at Sumsub, commented: “These principles from the UN are a great first step towards tackling the grave threat of misinformation online – although the framework fails to mention deepfakes.
“Age estimation technology uses AI to determine if someone is above a certain age to detect minors. Although not perfect and still prone to race and gender biases, it helps social media platforms enhance age and content controls. This technology and labelling AI content are essential steps for protecting children from harm and people from misinformation, manipulation, and fraud.”
The number and quality of deepfakes are increasing and evolving daily, worldwide. Even with the most progressive technology, it’s getting much harder to differentiate between a deepfake and reality. Media platforms must stay vigilant – making sure they are not inadvertently contributing to the spread of misinformation.
A key tactic for this is watermarking of AI content, as TikTok is now doing. The UN report suggests to “clearly label AI-generated or mediated content, investing in and developing solutions at the organisational level, to ensure that users can easily identify such content… This includes information in the metadata that identifies such content”.
Natalia Fritzen, AI Policy & Compliance Specialist at Sumsub, added: “Doubts loom over the vagueness of these requirements, and how effectively they can tackle the most common misuses of AI, i.e. the production of synthetic content, especially deepfakes. Worrylingly, recent data reveals that deepfakes increased 245% year-on-year in Q1 this year.
“When it comes to the section about “AI Actors”, for example, more specific actions that can be taken by deployers with the intent to mitigate AI misuse includes: remove malicious content from the training datasets, forbid inputs that violate company’s terms of use, human rights, etc, and, the most popular remedy, watermarks.
“Watermarks are not a silver bullet however, and questions remain about their effectiveness – mainly due to issues with technical implementation, accuracy, and robustness. Standardisation requirements must be set for watermarks, as we saw from President Biden’s Executive Order 14110.
“One thing the UN principles do well, is to emphasise the importance of AI literacy. With AI technologies evolving quickly, the population needs to be more and more aware of potential pitfalls in the contents they see, fostering what some call “pre-debunking”. Another important feature is the call for effective regulation, with legislations to be crafted in close collaboration with private technology businesses, acknowledging their frontline role in combating AI-related illicit activities, to establish a robust regulatory framework.”
The convergence of tech and brands 