Where AI labelling norms are ineffective

While the proposed amendments to rules are well-intentioned, doubts remain if they can deter Al-enabled crimes

- NS Nappinai

The Union finance minister's images are manipulated through Artificial Intelligence (AI) deepfakes and misused to perpetuate financial fraud — this is, unfortunately, nota hypothetical case.

Indeed, many such cases have surfaced lately. Celebrities and non-celebrities alike have fallen victim to Al-enabled deepfakes, often involving sexual imagery and harming their privacy and dignity. Now, there are reports of attempts to manipulate voter choice in the upcoming elections using Al-generated fake images of actors, wherein their likeness endorses or criticises a party. Digital arrest scams are also increasingly using AI deepfake imagery and voice to perpetuate fraud.

The question before us, therefore, is: Can the proposed amendments to the Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Rules, 2021 — Intermediary Guidelines, hereafter —meet this challenge? Apart from introducing new vocabulary — such as “synthetically generated information” (SGI), an umbrella term that includes AI deepfakes—the guidelines also call for labelling online posts as SGI if any such content is used. The labelling should cover 10% of a visual or be featured in the first 10% of a voice-based post containing SGI. Such compliance is limited to intermediaries offering their computer resources, which enable, permit, or facilitate the creation, generation, modification, or alteration of information as SGI. Such labelling helps primarily in terms of transparency, informing users about the synthetic nature of the content. But, to the extent that the SGI post is itself a crime, or enables crimes, or is used to violate the privacy or dignity of individuals, mere labelling will not suffice.