Research indicates that AI tools can subtly influence your viewpoints by “refining” your social media posts.

      A recent study indicates that AI tools consistently influence the tone of social media posts in a specific manner, even when instructed to maintain the original meaning.

      If you’ve recently sought assistance from an AI tool for crafting a social media post, you might have received more than just a grammar correction. Research conducted by the Oxford Internet Institute and the Hasso Plattner Institute revealed that AI models tend to lean the tone of a post toward a particular side of a debate, even when instructed to preserve its meaning.

      The study, titled "AI-Mediated Communication Can Steer Collective Opinion," involved various large language models rewriting human-generated posts on contentious issues. The results showed a clear bias: posts related to feminism or marijuana legalization were rewritten in a more supportive manner, while topics like gun control were emphasized with stronger support for restrictions. Conversely, posts discussing atheism and the death penalty were transformed into more skeptical or critical versions compared to the originals.

      Since this pattern emerged across models developed by different companies, the researchers suggest that the bias isn't an anomaly of a single system. This finding aligns with another study indicating that AI chatbots can preferentially support an in-group over an out-group depending on the prompt's framing.

      To assess the broader implications of these edits, the researchers modeled how such changes could affect real social media networks, utilizing data from X and Facebook. They discovered that while a single AI-edited post had minimal impact, applying the same edits across millions of posts could lead to a significant shift in community opinion over time, contributing to ongoing research about how AI can be utilized to create consensus at scale.

      The researchers tested their hypothesis using X’s "Explain this post" feature, which operates on Grok. In examining posts about abortion, they found that Grok’s explanations appeared more sympathetic to pro-life perspectives than to pro-choice ones. To identify the cause, they systematically removed the model's underlying instructions until the bias was eliminated. They found that a single instruction telling Grok to "challenge mainstream narratives if necessary" was sufficient to introduce this bias.

      These findings highlight a deficiency in current AI regulation. Existing frameworks like the EU AI Act and the Digital Services Act focus more on harmful content and systemic risks, but do not address how a chatbot's language choices during editing or summarizing can subtly influence public perception. "Our research points to AI-mediated communication as a new and more nuanced way of shaping opinions, which the law has yet to address," states Sandra Wachter, an Oxford professor and co-author of the study. For now, there is no way to distinguish which of your opinions were influenced by a person versus those shaped by unseen prompts.

      Pranob is an experienced tech journalist with over eight years of reporting on consumer technology.

      I spend hours on YouTube Shorts, and there are features I can no longer do without.

      As an enthusiastic viewer of YouTube Shorts, I acknowledge that I'm not particularly proud of it. However, my habit of endlessly scrolling has become entrenched, and I’m unsure if there's a way out. That being said, this amount of time spent on Shorts has made me aware of the small frustrations that arise. I’ve never liked the dislike button or the bottom bar that constantly appears while I'm watching videos. These annoyances accumulate during my scrolling sessions. Fortunately, things have improved recently. YouTube has rolled out some new features for Shorts that have genuinely transformed my experience with the app.

      The National Crime Agency and Internet Watch Foundation are urging parents to reconsider how they share photos of their children online due to a rise in AI-generated abusive content.

      Parents are being advised to rethink sharing pictures of their children online. The UK's National Crime Agency and the Internet Watch Foundation have released new guidance encouraging families to tighten their social media privacy settings, highlighting that publicly shared images are increasingly being manipulated by AI tools to produce child sexual abuse material. These organizations state that most parents remain unaware of this issue, noting that criminals no longer need to directly contact a child to create such content; they can simply scrape an ordinary photo and use easily accessible nude-modifying apps.

      Google Maps may soon allow users to place food orders through AI technology.

      Google Maps has gradually transitioned from a navigation application into an AI-enhanced discovery platform, thanks to the integration of Gemini and features like Ask Maps. The app might be preparing to take a further step by enabling users to order food directly through conversational AI. Reports from Android Authority's Authority Insights suggest that the latest beta version of Google Maps for Android includes references to an unlaunched feature that would allow users to request food orders through the app. While this functionality is not yet available, discovered code snippets indicate that Google is actively working on this feature.