Aindriya Barua was 11 years old when they started writing codes on a sheet of paper; they didn’t have a computer. When they went back to the school, their computer teacher was surprised. The codes worked.
It has been 16 years since then and now the 27-year-old computer science engineering graduate from Tripura has made fighting hate a mission in their life. And they are doing it using artificial intelligence.
At 24, Barua built a bot that tech giants should have - it can detect hate comments, hate codes, hate with context and remove it from platforms, not just for the users.
But it all started with art, as they tell Decode in an interview.
“I am a political person, so I started illustrating my experiences through art on Instagram,” they said. But soon, the comments on Barua’s posts started increasing along with the views. Much of those comments were filled with hate, queer-phobic and casteist remarks.
“You can’t solve gendered hate without looking at the intersections with caste and other narratives,” Barua said. So, Barua decided to solve the problem.
Barua is queer, indigenous, and neurodivergent. Barua identifies as non-binary and their preferred pronouns are they/them.
Sometime in 2020, Barua posted an illustration on Instagram against AFSPA (Armed Forces Special Powers Act) that existed in Kashmir and several regions of North East India. The comments came flooding. “I was called Umar Khalid ki rakhail, was asked if I eat snakes,” Barua recalled. Then there were hundreds of comments with ‘Jai Shri Ram’ written on it. Innocuous as it may seem, it wasn’t for Barua. “They are not praying in my comments, those are essentially hate comments,” the AI engineer said.
It got worse.
Barua didn’t censor themself on Instagram and continued creating art and that started irking a lot of nameless anonymous accounts. What started as trolling soon moved to doxxing. Their account which had about 10K followers got 60k engagements for each post. Many of these accounts then started tagging randomsena.
In a detailed report earlier, Decode reported about the man behind randomsena- Abhishek Singh from Bihar- who mobilises online trolls to target individuals who he deems have offended Hindu gods or Hindus and badgering police to arrest such individuals. Randomsena makes videos of his ‘raids’ after his ‘targets’ delete, apologise or are arrested by the police.
This is exactly what happened with Barua.
“When the hate comments started coming in and they started threatening me with FIR, I laughed. It was happening online, it wasn’t real,” Barua said.
But soon after these trolls contacted Barua’s classmates from college. Though Barua wasn’t a part of the college WhatsApp group, their friends in it called and told them and everyone had started talking about Barua’s Instagram post.
The troll accounts found out details of Barua’s life and started posting them on their Instagram handles. They shared Barua’s sister’s name, their phone number, and one even went to the extent of mentioning about her sister’s child. The trolls wanted Barua to apologise for her post. Every few minutes, there would be over a hundred comments on their post.
“And that’s when I got scared,” Barua said. “My teeth started chattering and it wasn’t even winter,” they added.
Barua got in touch with some lawyer friends and wrote somewhat of an apology note saying that it was a misunderstanding. But this is when Barua decided they had been through enough. They had to solve the problem.
They started looking for datasets on the Internet that could be categorised as hate in Hinglish (a combination of Hindi and English words), but there wasn’t any. So then 24-year-old started collecting their own dataset.
“I had built a strong community on social media by then, so I asked my followers to send link whenever they spot hate speech,” Barua said. It worked.
Barua built a document and categorised them into hate speeches that fall under political, communal, racist, casteism, queerphobia, sexism, ableism and others. “I wrote a scraper and it slowly started growing. I had thousands of links and comments on it,” they said.
When Barua signed up for the UNFPA hackathon, they did another call for volunteers to help them manually tag hate comments. Forty five people signed up. At the end of it, Barua, with the help of those volunteers, was able to create a bot for Reddit.
The bot, once active on a subreddit, constantly monitors the comments being posted. If hate speech is detected from a user, the bot comments with a warning. The bot gives three such warnings; the fourth time the same user uses hate speech, it permanently bans the user ID from posting on the subreddit. It then sends a message to the user the reason for the ban.
“People who contributed to the dataset had their own lived experiences,” Barua said.
Barua’s bot Shhor AI is unique because it can detect code mixed typing where a regular hate speech tracker doesn't work. For instance, when someone uses asterix and exclamations instead of writing the complete slang word. It even detects context based hate speech. “Calling someone ricebag may not seem like hate to tech platforms, but it is essentially hate because it implies that someone got a bag of rice to convert into a religion,” Barua explained.
While Barua’s AI bot was built to track hate, they never stopped receiving hate. It got exhausting for them. Barua took up another job as an AI engineer in a fintech company. “But I couldn’t do it. I wanted to fix the problem that no one else was,” Barua said.
Barua has now built an API that can be enabled in different platforms. On being asked if they have reached out to Meta because that’s where they received most of the hate, Barua said, “Meta probably doesn’t want to solve the problem. In fact, if you scroll through some post’s comments, they show the hate comments right on top.”
Every once in a while, Barua still gets threats. A few months ago, a vehicle stood in front of their house in the outskirts of Agartala and played a song, “Har ghar bhagwa chayega…”
“The lyrics of the song was scary. The vehicle parked in front of my house and kept playing the song in loop,” Barua said.
Barua’s parents who are farmers have asked them not to do the work that they have been doing. “They used to get calls from relatives asking me to stop, they were scared. But now they have accepted,” Barua said.
The 27-year-old now aims that their AI bot be used in platforms such as dating or gaming where people of marginalised communities face a lot of hate.
One of Barua’s first AI projects was in fact to help their farmer parents. “I made a system to automate irrigation that would sense the soil moisture and turn on water as required,” they recalled. It was one of the first times Barua had realised that they can solve real world problems with code.
But can the problem of hate on the Internet be solved? Barua laughed when asked. “I sometimes feel I am stupid. I am someone who comes from intersections of marginalisation and I am trying to solve a problem that tech companies with so much resources don’t even care about.”