LLMs, Including ChatGPT, Employ Manipulative Practices in Web Design, Study Reveals

Computer scientists from University of Glasgow, Technical University of Darmstadt and Humbold University of Berlin found that LLMs can include manipulative design practices if they are asked to build a web page. Many studies have been done on this topic and have found that a lot of human web developers use dark patterns or practices to lure visitors on a web page to do things or to stop them from doing some things.

Large Language Models have become capable of designing web pages but the researchers wanted to know if LLMs also use dark patterns in their designs. 20 participants asked ChatGPT to design a web page for an e-commerce website for the sake of experiment. Each participant offered different suggestions for the web page to ChatGPT and explained what they were looking for in the web page.

The results of the study showed that every web design created by ChatGPT used manipulative design practices. LLMs even generated fake reviews and fake documents which was a concerning thing. The experiment was done on other LLMs too and each result was the same. This suggests that LLMs have learned this skill from humans and now they can be leveraged to manipulate visitors on the site.

Read next: Bluesky Strengthens Impersonation Policies Amid Rapid Growth and Rising User Concerns
by Arooj Ahmed via Digital Information World