AI experts demand more AI regulation around deepfakes in open letter

Experts demand AI regulation: A coalition of artificial intelligence experts and industry leaders, including renowned figure Yoshua Bengio, has issued an open letter advocating for increased regulation surrounding the creation and dissemination of deepfakes, citing potential societal risks.

In the letter, spearheaded by Andrew Critch, an AI researcher at UC Berkeley, the group highlights the current prevalence of deepfakes, which often involve sexual imagery, fraud, or political disinformation. With rapid advancements in AI technology, deepfakes have become increasingly difficult to distinguish from authentic content, necessitating the implementation of safeguards.

Titled "Disrupting the Deepfake Supply Chain," the letter proposes several regulatory measures. These include the full criminalisation of deepfake child pornography, imposing criminal penalties on individuals involved in the creation or dissemination of harmful deepfakes, and mandating AI companies to implement measures preventing the production of harmful deepfakes by their products.

Over 400 individuals, spanning various sectors such as academia, entertainment, and politics, have already endorsed the letter. Notable signatories include Harvard psychology professor Steven Pinker, Joy Buolamwini, founder of the Algorithmic Justice League, and former Estonian presidents, along with researchers from prominent organisations like Google and OpenAI.

The call for regulation underscores the importance of ensuring that AI systems do not pose harm to society. This concern has been amplified since the introduction of ChatGPT by Microsoft-backed OpenAI in late 2022, which showcased remarkable human-like conversational abilities and raised questions about the potential misuse of such technologies.

Prominent figures, including Elon Musk, have previously raised alarms about the risks associated with AI development. Musk, in a letter signed last year, advocated for a temporary pause in the advancement of systems surpassing OpenAI's GPT-4 model to mitigate potential dangers.

(With Reuters inputs)