Issue #6: Reinforcing stereotypes
Since the rise of Generative AI in the public domain, many concerned individuals have worried that it would fuel misinformation. While it has indeed done so in some instances, these cases have generally been easy to debunk.
Even the notion that Generative AI would significantly accelerate the process of creation of misinformation has not demonstrated success on a large scale either. However, this doesn’t mean that Generative AI hasn’t found effective use in the field of information manipulation and propaganda. One of its most impactful applications has been in reinforcing stereotypes. What do I mean by this? You might assume I’m referring to scenarios where the user gives a vague prompt, and the AI generates a white figure for certain professions or automatically assigns brown faces to terrorists, or assumes certain languages like Arabic are linked to terrorism. But no, that’s not what I’m talking about.
What I mean is that Generative AI has become a tool for propagandists to represent certain stereotypes without needing access to existing materials that already promote these ideas. They are creating pseudo-original art that helps them depict specific characteristics. These works are “pseudo-original” because the AI generating these depictions has likely been trained on millions of images and artworks of real people, which is then tailored according to the preferences or prompts provided by a human.
Recently, India Today, one of India’s oldest English-language magazines, shared a cover art on their social media. The cover features AI-generated art depicting a bearded Muslim man and a veiled woman, essentially, visibly Muslim characters. The title reads, “The Muslim factor: Fragmented in the past, the community’s vote seems to be consolidating in this general election, impacting results in 86 constituencies.”
A person unfamiliar with India’s political landscape might perceive this as some unique insight into the Indian election. However, anyone with even basic knowledge of Indian politics would know that the Muslim vote has consistently been against the Sangh for years, even if it has been fragmented. The artwork alone is merely a skeleton; it is the accompanying text that reveals the full intent. This combination of image and narrative makes it clear that the portrayal is not as novel or surprising as it might initially appear. It is designed to reinforce the stereotypical image of a Muslim that the average bhakt has in mind.
Creating AI artwork that depicts certain stereotypes is not a crime. Even accompanying the artwork with text that contains hateful dog whistles and coded ideas is not illegal. Is this the fault of the AI model? Absolutely not. The AI generated exactly what it was asked to, and in fact, certain features were likely enhanced through subsequent human edits.
So, what does this mean for an Indian Muslim like me who has to witness it? Honestly, I’m not sure. It’s certainly better than receiving death threats or being lynched. Someone will probably read this and get angry at the AI company providing these propaganda publishers with their technologies. However, even if Generative AI were to disappear, it wouldn’t change my situation. I would still be a target for exploitation and demonization. The infamous “Jihad chart” displayed on prime-time television a few years ago didn’t require AI. It only required an environment of hate.
As I conclude this blog, I want to remind everyone that, according to a report by India Hate Lab, a Washington D.C.-based organization that documents hate speech against religious minorities in India, there were 668 documented cases of hate speech against Muslims in India in 2023. It is pertinent to note that the report comprises of hate-speeches delivered by prominent individuals. There are no fringe elements involved here. Today, the fringe is the mainstream.
— — — — — — — — — — — — — — — — — — — — —
Stuff I have been reading + watching:
There are no recommendations with this one. I am honestly too exhausted :)
