Facebook whistleblower docs: How AI is hurting India – and Indians
Written by Tulika Avni Sinha
Indian Express, 19 . 12. 23
In 2022, India’s internet usage reached 52 per cent, with over 467 million social media users. The number of social network users in the country is projected to cross one billion by 2025. It is, therefore, no surprise that platforms such as Meta (formerly Facebook), Instagram, WhatsApp, YouTube, and Snapchat find the country their largest audience globally. LinkedIn and X (formerly Twitter) too have India amongst their top markets, close behind the US. These platforms and algorithms have massive influence in spheres of political, religious, social, and cultural aspects of the lives of millions of Indians. It is unsettling to see that this impact continues to stir violence, community unrest, and hatred amongst people without accountability or effort to improve public awareness.
In 2021, a whistleblower leaked internal documents and images from Facebook and its subsidiaries (primarily Instagram). In November 2021, the Public Interest Tech Lab at the Harvard Kennedy School received copies of most, if not all, of these documents from an anonymous sender. The Lab has since digitised and published the received documents to a website accessible online. These documents date back to 2016 and after and expose many problematic revelations related to the platform’s direct and indirect role during elections in the US, India, and other countries, among other issues. As would be expected, many of these documents disclose the internal discussions and analytics for the platforms’ (Facebook/Meta and Instagram) performance, findings, and plans of action for India in detail.
Other than the more prominent reveals around the subjectivity and unexplained discretion in the application of policies and moderation guidelines, as evidence to the warnings being issued about all hailed Artificial Intelligence (AI), is how engineers and product teams behind the platform always played catch-up with their algorithmic functioning. The developers and designers of these algorithms at no point clearly knew how their algorithms were functioning or what they might have learned in the process to achieve the objective assigned to them. For example, images of chats throw light on the confounded and shocked reactions of team members upon analysis that new users are shown polarising content, hate speech and misinformation as “recommended” to generate engagement and interest; their news feed is filled with fake news items and graphic content divisive in nature; the algorithm incentivised bad or angry behaviour and anti-democratic norms from users to achieve higher participation; people with limited literacy skills were exposed to more misinformation and less civic content. Several documents also display the organisation’s awareness about the platform’s impact on the increased religious and political violence and the spread of misinformation. There was evidence that online memes, posts and shares led to offline violence and killings.
“Playing catch up with algorithms” is not the industry exception. It is the norm. Machines learn from the training dataset provided to them and are constantly improvising to achieve their assigned objective with the highest achievable efficiency as per their model.
Users need to understand that since these platforms and others are developed outside of India, their preparedness to understand and manage nuances of language, culture, and regions fall short of that required. In the case of Facebook and Instagram, while “India” was constantly marked as a “high market” and “high risk” country, both policies and technology lacked local contextualisation to accurately flag and demote inflammatory content. For example, the documents show that a lack of sufficient classifiers in local languages (such as Hindi and Bengali) allowed harmful content to slip through. Advertisements disseminated misinformation and hateful content, evading fact-checks and moderation guidelines.
These are just some examples of exaggerated algorithmic dependence. There are numerous others in the Facebook archives and in other studies globally that should serve as caution for the Indian audience. For example, In 2021 Twitter admitted that “Right-leaning news outlets, …., see greater algorithmic amplification on Twitter compared to left-leaning news outlets”. A research study by Joy Buolamwini found that “leading tech companies’ commercial AI systems significantly mis-gender women and darker-skinned individuals.”
The hype around algorithmic efficiency can lead to an amplified technology dependence, especially amongst a population that does not understand its limitations and shortfalls. Machines primarily operate to achieve their objective. The idea of machines achieving a human-like general intelligence is far-fetched, especially with the multiplicity of objectives, environment, morals, and emotional parameters humans operate around. Algorithms do not question if a practice is harmful and exacerbates inequity. While they can help us significantly improve our everyday efficiency, the judgement and authority need to be in human hands, coupled with an aware and perceptive mind that can question the why and how.
AI applications deployed in a country like India need to be developed and/or trained with inclusive, representative, and contextualised data from the country’s population to adequately reflect its diversity of people, languages, cultures, and context. If we fail to think about the ethics of AI now, we shall propagate and, in all likelihood, promote the world’s popular biases, and shortcomings to the future without adequate attention to rectify its ignorance and unawareness; meanwhile worsening lives of minorities and underprivileged groups who are not sufficiently represented in the training datasets. For the applications trained on datasets from the developed world, these shortcomings and biases shall not even be our own but of a population we neither belong nor relate to.
It is time the general population is made aware of how algorithms work, to use them, and to use one’s own judgement and perception. The human capabilities to reflect, debate, question, and deny cannot be undermined and need to be developed further, especially if the country is to reap benefits from this technological revolution that will change the way we work, communicate, learn, and live.
The writer works at Harvard University on research related to data privacy and ethics
No comments:
Post a Comment