[ Chandramouly Pandey ]
In the last few days, social media has been buzzing with the new '1980 trend' with vintage filters, grainy pictures, aviator glasses and people generating images with great enthusiasm, trying to bring back and live in the nostalgia. Amid all this fun, its utmost important to understand the growing privacy concerns over these AI images, which are overlooked by the common people.
Trends like this are not a new phenomenon. Earlier, Ghiblify from ChatGPT used to turn selfies and images into young, short-ish versions similar to characters from Studio Ghibli in Japan. Soon, this trend was followed by Nano Banana AI images, which famously produced images in a cinematic, Bollywood-style 90s retro look with sarees and kurtas. This transformation takes merely minutes to produce images, but the trend is more than just an image transformation. When someone uploads an image from their gallery, the system processes it through facial recognition technology to map facial expressions and mood, which it can then use to generate more human-like AI images.
Apart from this, any selfie or picture is more than an image; it's the data that you're feeding to LLMs to train the model. A photo carries metadata, including the time and place of the picture, the device used to take it, and the precise location. Algorithms then process all this data to develop a profile of an individual and decipher mood, choices, preferences and habits. These are then used for personalisation and creating filter bubbles around users through information overload.
According to a report published by NYU Abu Dhabi, AI systems collect data in multiple layers. This includes what you type, what you upload, and most importantly, what the system is secretly driving out along with these data, including IP address and device model.
Another report by Wired raised concerns over the mindless uploading of images in AI chatbots. Concerns over the precision of AI processing data made the headlines when a girl uploaded a picture of her face during the Nano banana trend, but AI generated her full picture with accurate positioning of a mole on her hand which was not present in the uploaded picture. This incident unveiled serious privacy concerns about AI's data-processing capabilities.
Another important aspect is to understand the data storage and processing policy of these AI bots. Images uploaded for fun on social media may fade in quick time, but the images on AI chatbots remain for a longer period. ChatGPT keeps chats till the user deletes them or upto 30 days or even longer if de-identified from the account. Gemini retains chats for more than 18 months, which is a substantial time to process data and retrain models through machine learning. Claude stores the data for 30 days even after the user deletes the chats from their side, and Meta AI has yet to specify any time period for data deletion.
What is more important to understand is that, despite the data having been removed, once it is processed to train the model, it is almost impossible to remove it from the AI's memory.
The concept of manufacturing consent propounded by Noam Chomsky in the late '80s is working well in the era of artificial intelligence. Tech giants are manufacturing consent for our personal data and metadata through these social media trends and we are, from time to time happily handing them our personal data in the name of social media trends and fun which can turn dangerous if data gets leaked or servers get hacked. It's important to become AI literate before using AI. Skipping trends, refraining from uploading others pictures, especially children's, without their consent and turning off the model training option in the settings and deleting older chats that are not required could be some possible solutions. (The contributor is a research scholar in the mass communication department of Rajiv Gandhi University, Rono Hills.)
