"You may shoot me with your words,
You may cut me with your eyes,
You may kill me with your hatefulness,
But still, like air, I'll rise."
-Maya Angelou
Code-mixing is common in developing countries where we use English keyboard to type in our native languages creating a complex language with variable spellings and pronunciation, unknown to existing tools. Banning abusive words gets tricky as abusers can use varied spellings.
That's where AI comes in, AI understands the context of sentences, identifying hateful essence beyond fixed rules. Most abuse in Indian social media is in Hinglish- code-mixed Hindi-English.
AI requires a significant amount of training data to excel at its job.
Tagged data for Indian languages is already scarce. No good data was directly available.
Aindriya had to take up on the huge task of manually building a vast repository of Hinglish comments, correctly tagged as hate or not, to train the AI.
"Jab zulm-o-sitam ke koh-e-garan,
Rooi ki tarha ur jaenge
Hum mehkoomon ke paaon tale,
Ye dharti dhar dhar dharkegi
Hum Dekhenge"
"When the mountains of oppression & cruelty,
Will float away like carded wool.
Underneath our feet- we the governed,
The ground will echo like a thumping heartbeat,
We will see"
(Urlish)
-Faiz Ahmad Faiz
(Aindriya Barua's) hate speach detection algorithm for social media platforms can scan, detect & classify hate speech from text, images & videos- a solution that can detect offensive content in Hinglish & has the potential to be applied to regional & vernacular languages too. It's a well conceptualized & executed tool that can work towards addressing technology-facilitated gender-based violence. Congratulations Aindriya!
Andrea M WojnarA monthly report released by Meta on May 31 (2022) showed that both Facebook & Instagram had seen more than an 80% rise in hate speech & violent inciting content... where Islamophobia is rampant, followed by castesim, sexism & queerphobia, especially directed at trans people... Back in 2021 machine learning engineer & political artist Aindriya Barua decided to do something about this problem after getting death & rape threats when making socio-political art...
Learn MoreAindriya Barua, a 24-year-old Software Engineer from Tripura, has built a bot that many tech giants couldn't...
Learn More"Chitto jetha bhoy shunnyo, uncho jetha sheer,
Gyan jetha mukto, ...
Bharotere shei swarge koro jagorito"
"Where the mind is without fear & the head is held high,
Where knowledge is free ...
Into that heaven of freedom my Father, let my country awake."
(Benglish)
-Tagore
Existing AI systems trained only in major languages can't recognize code-mixed language, like Hinglish, Tamlish, etc. Shhor's AI is specifically trained to tackle this issue.
Today, it's not just comments & captions, but there's hate in memes, i.e images, & the ever-booming content-pool of reels/short-form videos. Don't worry Shhor has it all covered!
Marginalizations are intersectional & create compounded social stratas, intensifying the abuse. For instance, a lesbian faces hate, but a Dalit lesbian experiences even more severe hate. Shhor's novel research reveals these unnamed social hierarchies & their influence on discrimination through real world data. By shedding light on these complex dynamics, Shhor AI contributes to a more comprehensive understanding of discrimination and facilitates targeted interventions for social justice.
Miscreants often use symbols, emojis & different spellings to by-pass the moderation tools in social media, which is why often, even big social media companies cannot detect these aggressions hidden in plain sight. Fear not. Shhor's powerful AI sees it all!
Shhor strives to make the internet a safe space for every marginalized community out there. Hence, Shhor identifies eight kinds of hate- Queerphobic, Gendered, Communal, Political, Casteist, Ablelist, Racist & General hate.
Shhor is a one-stop-shop for legal guidance, technical help, or mental health resources. Shhor provides an AI tool that can be used to identify the IPC laws certain abuse breaks & simplifies legal recourse.
"Tum Kala Kamal Likho, Hum Lal Gulab Likhenge,
Tum Zameen Pe Zulm Likh Do,
Asman Pe Inquilab Likha Jayega"
You write black lotus, We will write red rose
As your pen scribbles injustice on the earth,
Revolution will be written on the skies.
(Hinglish)
-Aamir Aziz
Being a one-person team, Aindriya handles various responsibilities, including data collection, research, AI development, backend/frontend development, UI-UX design, dev-ops, marketing, PR, & community outreach. But it's not sustainable, & hiring a talented team is essential for future growth.
Currently, Aindriya has demonstrated the effectiveness of the AI model through integration with Reddit. Aindriya has an extensive plan layed out aiming to create a versatile bot that can be utilized across all social media platforms. This ambitious goal requires a strong team of skilled developers.
There's a need expand research from Hinglish to other languages & their code-mixes. The challenge is in dataset building. With enough resources, Aindriya plans to turn this process into an employment opportunity for minority communities, providing them with job opportunities & empowering the oppressed.
Soon, Aindriya wants to launch this full-fledged tool for free to everyone on the internet, with a bunch of exciting & meaningful features.
Research on image processing can be built on top of the current text-based model to further improve image & video analysis.
With expandd datasets & access to high computation power, the accuracy of the model can be further improved using advanced algorithms.