Have you heard of the dead Internet theory? It suggests that the Internet is largely fake and dominated by bot activity focused on producing algorithm-driven content rather than genuine human creation. It seems that this theory may hold some truth, in light of a new study claiming that 57.1 percent of online content is generated by artificial intelligence.
The study, “A Shocking Amount of the Web is Machine Translated: Insights from Multi-Way Parallelism,” claims low-cost machine translations are driving the increasing number of translated web pages. The researchers found that there’s a selection bias in what gets translated into multiple languages. This translated content is often shorter, easier to predict, and covers different topics compared to content that is only translated into one language.
As to how the researchers arrived at the figure of 57.1%, they analyzed a vast collection of web content and identified translation pairs. They found that 3.63 billion sentences in their dataset are part of groups translating to three or more languages.
- 90s AI Yearbook: How to generate your own retro-themed images
- How to create your own Korean AI profile picture
The high occurrence of AI-generated content has experts concerned about “model collapse,” the gradual deterioration of machine learning models as they train on uncurated synthetic data from other models, including earlier versions of themselves. Research scientist Dr. Ilia Shumailov from the University of Oxford shared his thoughts on the matter with Windows Central. He highlighted how model collapse can initially impact poorly represented data and ultimately lead to diminished output diversity.
You can read the study here.
This article, Study reveals 57% of online content is AI-generated, was originally published at NoypiGeeks | Philippines Technology News, Reviews and How to's.
Comments
Post a Comment