Forecasts from experts highlight the impending risk of AI generating nonsensical data
As per the findings of University of Cambridge scientists, neural networks possess the capacity to yield entirely meaningless outcomes
The rise of artificial intelligence tools is an increasingly noticeable trend. Noteworthy figures like Elon Musk and Steve Wozniak, along with a consortium of 22,000 other experts, are going so far as to call for a pause in neural network training.
Despite the concerns raised by these experts, the progress of AI technologies continues at a rapid pace. Previously, systems like ChatGPT, Stable Diffusion, and Midjourney relied on human-generated content for training. However, there’s now a shift as neural networks gradually adopt AI-generated output as training data. This shift is driven by the abundance of AI-generated data flooding the internet, which serves as the primary information source for these networks.
Researchers from the University of Cambridge offer a concerning outlook: they predict that in a few generations, AI might churn out what can only be described as “garbage.” The root cause they identify is the issue of model collapse, resulting from AI’s misinterpretation of probabilities. Since neural networks lack the ability to assess the credibility of internet information, they will increasingly produce errors and meaningless outcomes over time.
