The model, they claimed, is currently trained to detect caste discrimination within all forms of text data, including status updates, messages, tweets, comments, etc.
"We work with customers to iterate on the baseline model on a regular cadence (typically monthly) to ensure that we are flagging the content our customers need. This process ensures that our results are customised for each customer so they can trust the results," a statement said.
The model can help recognise and respond to toxicity immediately before it evolves into an even bigger problem, the statement said.
It added that the model offers the power to understand the community while maintaining the data privacy requirements.
The statement noted that the AI model surfaces results that can be plugged into the existing moderation efforts of its clients.
This includes webhooks into internal systems for customers to manage users (warn, suspend, ban, etc), content (remove a post, limit who can see it, etc), send alerts, send for moderator review, and even pipe into analytics platforms to see trends over time, it added.
"Caste discrimination is one of the oldest forms of evils still existing in Indian society. Sadly, it is also reflected in the cyber spaces, we together with Spectrum has created an AI tool that will help social media platforms..." Amitabh Kumar, founder of Social Media Matters, said.
Initially, the model is trained to work with several languages English, Hindi, and Hindi-English mix and it will be upgraded further, he added.
Justin Davis, CEO at Spectrum Labs, explained that the hardest part of building an AI model that can effectively detect caste discrimination online is defining and understanding not only what caste discrimination is, but also what it is not.
"...their (Social Media Matters) insights and expertise helped us navigate the nuances, history, and politics of caste discrimination, to build a tool that can combat it effectively and inclusively," he added. PTI SR BAL
Disclaimer :- This story has not been edited by Outlook staff and is auto-generated from news agency feeds. Source: PTI