Facebook looks to restrict harmful content with new investments in AI

Facebook at the present is facing numerous investigations globally with regards to privacy violations as well as the spreading of biased and harmful content

Facebook has reportedly confirmed it would make long-term investment towards developing proactive Artificial Intelligence (AI) for detecting content that violate its policies.

Apparently, the social media giant has developed a new approach to object recognition, termed as Panoptic FPN, for helping AI-based systems to understand the context of content through backgrounds of photos.

Manohar Paluri, from the AI team at Facebook, stated recently that the company is working on its AI systems to help identify more of these problematic online content by understanding them with minimum supervision needed.

Paluri said advances made in natural language processing have enabled the company to develop a common digital language for translation to recognize harmful content across multiple languages. He claims that training models which have a combination of audio and video signals improve results further.

Citing reliable sources, Facebook at the present is facing numerous investigations globally with regards to privacy violations as well as the spreading of biased and harmful content through its social media platforms, including messaging platform WhatsApp.

The company is supposedly coming up with best practices in terms of fairness, in every step during product development, for ensuring AI is able to protect people and no discrimination is done against them.

Sources mentioned that there are risks associated with AI models when trained by humans on datasets consisting of people. In case any limitations are there in the datasets, along with flaws or other problems, the AI models could perform differently for different subjects.

For managing this risk, Facebook says it has built a new process for inclusive AI. It offers guidelines for helping programmers and researchers design datasets, test new systems and measure performance of product through the lens of inclusivity, sources added. Dimensions for vision include age, gender and skin tone where as for voice, it includes age and gender as well as dialect.

 

Source Credit: https://www.cnbctv18.com/technology/facebook-investing-in-inclusive-artificial-intelligence-to-curb-harmful-content-3163681.htm