Key Takeaway
With increasing levels of spam and cyber bullying present on social media platforms, Instagram has adopted an AI engine called DeepText, a program earlier discovered by Facebook, in an effort to make the platform safer for its users.
Attempting to root out rubbish from the Internet is understandably a herculean task. As more social media platforms continue to gain momentum, incidence of cyberbullying has also seen a significant rise. A survey conducted by Ditch the Label, an international anti-bullying charity, suggests that 42% of more than 10,000 UK youth between the ages 12 and 25 reported Instagram as the platform where they were most bullied.
With a goal of eradicating spam and making a better product for users, Facebook (which acquired Instagram in 2012 for $1 billion), built DeepText. DeepText is “a learning-based text understanding engine that can comprehend, with near-human accuracy, the textual content of several thousand posts per second.” It covers more than 20 languages and is based on recent advances in AI, as well as a machine learning concept called Word Embedding, which is designed to mimic the way humans interpret language.
Whenever it comes across a new word, the system tries to interpret contextual meaning from the other words around it. For instance, the word “white” will have different interpretations when placed next to the words snow, sox, house or power. In such a situation, DeepText would study the surrounding information before zeroing in on the most appropriate interpretation of “white”. It gets more efficient and accurate over time.
DeepText was built, primarily, as an in-house tool that would allow engineers at Facebook to categorise significant masses of data and create classification rules. They could build products that would allow users to have a more comfortable experience on their platform. For instance, if someone is on Facebook talking about the Chicago White Sox, the system should be able to interpret it as a baseball game, and categorise it as a sports team.
Creating a safer internet using context clues
“Powered by machine learning, today’s filters are our latest tools to keep Instagram a safe place. Our team has been training our systems for some time to recognize certain types of offensive and spammy comments so you never have to see them.”Instagram CEO, Kevin Systrom
The team at Instagram believes the technology behind DeepText can help them to better combat cyberbullying on the platform. Although its original intention was to highlight photography, users who leave Instagram often do so as a result of negative text and comments. Often these are created by bots (sometimes humans, too) who market cheap products or simply harass users.
With the introduction of DeepText, spam and negative comments that violate the platform’s community guidelines will vanish immediately once detected. However, the person who typed the message would still be able to see it on their account, making the system more complicated to game. The technology will automatically be incorporated into a user’s personal feed, but it can also be easily disabled via account settings.
Can AI determine tone and intention?
Like any machine learning platform, DeepText will continue to improve its accuracy over time. In the beginning, it’s inevitable that some abusive content will escape as the system analyzes more and more data. The concern, however, is false positives: constructive criticism, or helpful comments that the system may consider abusive or inappropriate and decides to delete. This raises the question: is it possible for an AI to determine tone and intention?
In one instance, DeepText comically caught on to a disparaging lyric in a Kanye West song called “Famous”. Almost every line in this sequence got barred when it was put through the AI system.
Naturally, lyrics to this song have consistently been banned by Instagram’s AI text system. Although the AI is considered “near human”, we’ve not yet reached the point where it can decipher nuances in meaning, tone and context across cultures or scenarios.
Therefore, the risk facing Instagram after incorporation of DeepText is that the filter may change the essence of the platform and make it overly sanitized. Making their AI system more proficient to work 100% like the human brain and be able to distinguish between abusive language and constructive criticism across a wide cultural and linguistic landscape will be a significant challenge.
Going forward
Systrom’s ambition isn’t just cleaning up the platform that he runs, but in light of an ongoing international conversation against spam and cyber bullying, he also aims to validate the fact that toxicity online isn’t ineluctable. He says, “Maybe trying sends a signal to other companies that this is a priority, and starts a national and international conversation that we should all be having about creating safe and inclusive online communities, not only for our kids but for our friends and our families.”
Social media should aim to provide a safe virtual environment for its users and an effective tool for community engagement; it should not facilitate abusive trolling and bullying. The incorporation of AI into platform algorithms has definitely helped control malicious content on social media platforms, but AI algorithms need more work. Leading companies such as Instagram and Google are investing more time and financial resources towards making these platforms clean and free of cyber bullying with the help of AI.
