The AI sector has barely been born when it is about to experience its first major crisis. Quality natural data is starting to be lacking.
A year after the release of ChatGPT, it is now undeniable that AI is working real wonders, both from the point of view of productivity and creativity, with certain language models even being capable of creating original works ex nihilo… apparently. Because for the designers of this technology, there is no chance; if AI has reached this stage of development, it is thanks to the formidable quantity of data she ingested to train.
According to Professor Rita Matulionyte of Macquarie University in Australia, the machines could arrive quickly exhausted all available data. In other words, AI will soon have nothing left to train itself on, and in the very near future, chatbots could start rehashing the same ideas. The most alarmist forecasts evoke this major shortage by 2026.
AIs have consumed almost all the data available for their training
The situation is serious. If “an algorithm is trained on an insufficient amount of data, it will produce inaccurate or poor quality results “. AI designers are well aware of this, but unfortunately there is nothing they can do to compensate for this shortcoming. Some people try to train their machines with sets of data created by chatbotsbut this technique is already showing its limits, with research indicating that this training leads to “confusing and worrying” results, like this trailer for a children’s film, with nightmarish results.
Natural data is therefore a finite resource. Somewhat in desperation, OpenAI, the designer of the famous ChatGPT, launched an appeal to communities and other organizations holding large data sets. The latter are increasingly rare, therefore more and more valuable, and with the great shortage that is coming, the entire AI industry risks experiencing a slowdown as sudden as its start was explosive .
Source: The Conversation