The Internet is changing so much that some key people in the technology sector are starting to worry. Sam Altman, CEO of OpenAI, the company behind the famous ChatGPT, has recognized something that many previously saw as a conspiracy theory: The so-called dead internet theory could be going from being just a crazy idea to a reality.
This basically says that much of the content consumed on the Internet is no longer created by real people, but by bots and artificial intelligence systems that generate texts, images and videos automatically. In a simple way, There is already an immense number of articles, posts and even comments on social networks that are not made by humans, but by machines..
In this way, the line between what is true and what is false becomes so blurred that it becomes almost impossible to trust what you see or read.
Researchers even speak of a “debasement of the digital brain,” because this enormous amount of false or content created to be viral makes the artificial intelligence systems that are trained with them also worsen in capacity, precision and even common sense. It is the whiting that bites its tail and the AI begins to forget deep and original thinking because it learns from increasingly poor information.
The vicious cycle is obvious: the more garbage there is, the worse artificial intelligence becomes, and when AI gets worse, it produces more garbage content that further feeds an internet already considered dead. To make matters worse, this situation affects the creation of public opinion and can make it really easy for certain ideas or false news to spread uncontrollably.
Sam Altman and other experts believe that the solution involves drastic changesfrom limiting automatically generated content, creating tools that allow us to distinguish true information from synthetic information, and, above all, letting people know that this is really happening.
«Human-generated content is disappearing and we have no copy»
This is what some experts call model collapse or collapse of the model. Although it is also true that the phrase ‘Garbage In, Garbage Out’ has become better known (Garbage In/Garbage Out or GIGO, for its acronym in English).
That is why preserving original human content has become an absolute necessity. AI models learn from large amounts of data, and if that data does not reveal human reality, AI will not be able to understand us well or help us either.
For example, If you want a chatbot to explain a medical topic to you, what would be perfect is for it to have learned from texts written by real experts, not from answers generated by other machines that may have errors..
You may not realize it, but when you search for something on Google, read the news or visit a forum, you are interacting with a huge amount of information that mixes human and AI-generated texts. That can influence what you believe, how you inform yourself, and the decisions you make.
Según Sam Altman, More than 100 billion words are already generated a day with AI, many of them accessible on the internet.
A study of Bloomberg Research put this problem into perspective, analyzing how eleven major language models, including GPT-4o and Llama-3, used more than 5,000 harmful cues, which «produced bad results.»
To avoid this collapse, Some researchers are working to create archives and databases with original content, prior to the explosion of generative AI. For example, projects like LowBackgroundSteel.ai collect websites, texts and data created before 2022 so that they are not lost.
There are also initiatives like GitHub’s Arctic Code Vault, which stored a copy of millions of lines of open source code in a mine in Norway, as an archive for future generations. They are like digital time capsules that preserve the human essence online.
In addition, techniques are being created so that AI models can identify and differentiate content generated by humans from that generated by machines, and thus prevent them from ‘feeding themselves’ with synthetic texts that degrade their quality.
Get to know how we work in NoticiasVE.
Tags: Chrome, Artificial intelligence, Software