Doomscrolling. A term coined to express a constant need to consume negative news back in 2020, but has evolved to describe endlessly watching short-form content. For You pages are the main culprit of this change in the definition, the algorithm catering right to what we want for engagement.
With such a rise in the usage of AI and the embedding of it in everything we do, I have some concerns on what is going into the databases that these AIs use. The data put into these models directly impacts their responses and quality. There needs to be a choice for users to be able to opt-out from AI training, enforcement of regulations and standards, alongside a general slowing to the push for AI.
Interactions and content on platforms like Instagram, TikTok, Youtube and X (formerly Twitter) are constantly being logged on your own device to build a picture-perfect feed, catering to what platforms believe you’ll enjoy.
These algorithms are slowly changing to be solely powered by AI to get a hyper-personalized feed. Content is being analyzed — not just for the algorithm, but also for training AI models.
Statements concerning whether your content will be used for AI training are tucked away in Privacy Policies, like the Meta Privacy Center statement that says, “We use information that is publicly available online and licensed information… This information could be things like posts, photos and their captions.” That information goes to databases to feed both generative and predictive AI.
Opting out of having your content used for AI training means privating your account in the case of Instagram and Facebook. There is no foolproof way of opting out in the U.S., as Meta holds the rights to object any manual requests of an opt-out.
X follows suit with this non-opt-out policy and I don’t see a reason why these processes to deny using content for AI training should be as convoluted as it is. Exercising your rights to data protection should be simple, as easy as clicking a button.
Examples of internet-based learning are in OpenAI’s models, ChatGPT and Sora AI, but also in Gemini — something we’ve all been forced to use because of AI Overview.
AI Overview is a prime example of how AI responses are hastily “cobbled together” and how under-developed standards are for these public models. With no opt-out button, we’re forced to see “answers” with varying amounts of authenticity. Responses from AI Overview link Reddit as a source 20-40% of the time (Yahoo) and are usually not reliable, with nonsensical examples like “adding glue to cheese to help it stick to pizza,”.
So, there isn’t much of a filter on what is and is not used for answers, seeing how often unreliable sources are referenced in AI Overview. This push for AI Overview also brings us quantity over quality. The process behind AI Overview can be explained simply as Gemini scouring the internet for buzzwords, with no further processing or thinking after it’s found and presents a slopped-together “human” response on a Dixie plate.
Even with all of its failures, AI has been relentlessly forced in front of us, snaked into simple searches and taking our data to attempt to improve itself. Artificial Intelligence shouldn’t be this “tool for everyday life” that we can’t turn off, especially when it showcases qualities that reveal issues in AI training and general logic.
