Committee calls for bill to address AI threat to human rights

A cross-party group of MPs and peers, comprising members from the Labour, Conservative, and Liberal Democrat parties, have identified human rights risks posed by artificial intelligence (AI) which existing laws are insufficient to address. The Joint Committee on Human Rights (JCHR) has published a report urging the creation of a new AI bill to tackle the scale and severity of these threats.

Alex Sobel, the committee's chair, stated that no current legislative or regulatory approach to AI is adequate, either globally or within the UK.

Dario Amodei, the head of Anthropic, an AI company, proposed a plan that includes global regulation, industry-wide regulation, and independent monitoring of AI models as they are developed. He emphasized the need to slow down the development pace and echoed similar sentiments from other prominent figures, including Sam Altman of OpenAI and Elon Musk. The latter two have also expressed agreement with Amodei's proposal.

A major concern regarding AI is its potential to exhibit biases against certain groups if trained on large datasets, often scraped from the internet, which may include racist, sexist, and other undesirable material. Additionally, AI can be employed to generate and disseminate misinformation. One human rights concern raised in recent years is the use of AI to create and distribute harmful content, such as sexualized images of women and girls, and facial scanning without consent.

In a summary of its 100-page report, the JCHR points out numerous instances where AI has led to human rights abuses, like the creation of sexualized images of women and girls and unauthorized facial scans. The committee advocates for the establishment of a single, independent AI oversight body on a statutory basis. They argue that the current legal framework governing AI is fragmented and challenging to navigate, leaving gaps in protection.

The JCHR recommends that the AI Bill establish a regulatory regime that categorizes different risk levels and imposes more stringent obligations on higher-risk AI systems and models. Obligations should apply at all stages of the AI lifecycle, ensuring those responsible for designing, modifying, and utilizing AI systems are aware of their responsibilities and these are effectively addressed.

Some AI uses should be outright prohibited as they conflict with human rights, such as subliminal techniques and inappropriate profiling or biometric data applications.