Researchers claim that AI chatbots possess an uncanny ability to decipher an individual's personal details merely from their online typing patterns. And that online users are risking their privacy with every keystroke.

Computer scientists in the Swiss-based science school ETH Zurich described how their innovative research, which hasn’t been peer-reviewed yet, may bring up some big internet privacy issues.

It turns out that AI can accurately deduce a user's gender, age, location, and more based on their textual interactions.

The study's findings bring new fears.  The idea that AI's capacity to "infer personal data at a previously unattainable scale" is scary. Because the feature could be exploited by malicious actors, including hackers. 

"This is very, very problematic," expressed one of the study's authors to online tech magazine Wired. Indeed, the implications of this research are vast, as the AI's prowess in personal data extraction could have significant repercussions for user privacy and security.

DuckDuckGo Founder Testified in Antitrust Trial that Google Stifles Competition - USA Herald

Ring Attention Revolutionizes AI Models: A Breakthrough to Unlock Vast Memory - USA Herald

AI Chatbots Deduct with Uncanny Accuracy

The research, which was published in October, highlights that AI chatbots have the capability to accurately guess various sensitive attributes, including gender, location, age, place of birth, and even occupation.

These are characteristics that are typically safeguarded under privacy regulations. The authors of the study further emphasize that AI's skill in deducing personal data could be put to use by hackers who may employ seemingly innocuous queries to extract sensitive information from unsuspecting users.

To substantiate their findings, the study examined how large language models, the very engines powering chatbots like ChatGPT, could be prompted to deduce personal details from a dataset of 520 real Reddit user profiles and their posts spanning from 2012 to 2016.

Through manual analysis of these profiles and a comparison with AI-generated predictions, the researchers unearthed some alarming insights.

Of the four AI models subjected to testing, GPT-4 emerged as the most accurate in inferring personal details, boasting an astonishing 84.6% accuracy, according to the study's authors. Other models assessed included Meta's Llama2, Google's PalM, and Anthropic's Claude.

Interestingly, the study also noted that Google's PalM exhibited a tendency to decline around 10% of the privacy-invasive prompts aimed at deducing personal information, while the other models were even less discerning.

The implications of this research are profound, as it underlines the potential risk AI chatbots pose to user privacy. For instance, the study revealed an instance where the AI deduced a Reddit user's location as Melbourne simply based on a reference to a "hook turn."

"A 'hook turn' is a traffic maneuver particularly used in Melbourne," explained GPT-4 after being prompted to identify details about the user. Such seemingly innocuous clues could lead to the unwitting exposure of personal information.

This study is not the first to raise concerns about the privacy implications of AI technology. In a separate research effort published in August, it was determined that AI could decipher text, such as passwords, based on the sound of a user's typing, recorded over Zoom, with an astonishing accuracy rate of up to 93%.