Will we soon run out of data to train artificial intelligence?
Artificial intelligence is on everyone’s lips – especially the AI software ChatGPT. Because the language model impresses with amazing results. This is also due to the intensive training of the AI. But according to forecasts, the data to train artificial intelligence could soon run out.
In recent years, researchers around the world have been developing ever more powerful AI algorithms. Some of them already help us in everyday life. Other approaches, on the other hand, quickly disappear due to their susceptibility to errors. In any case, one thing is of particular importance for artificial intelligence: data.
Because every algorithm needs data in order to react to possible inputs in a meaningful and correct manner in the long term. Developers have therefore relied on high-quality content such as research results, studies or other scientific papers for years. But already soon the data could run out to train artificial intelligence.
Training artificial intelligence: will the data soon run out?
Initial forecasts show that AI algorithms could have consumed all high-quality data as early as 2016. Then there would only be two options for training artificial intelligence: either you feed it again with already known data or you switch to “inferior” content.
The boundary between good and bad sources is considered to be blurred. In general, however, the origin can be determined. While high-quality data comes from professionals, bad data is the product of online discussions and forums. So everyone can imagine the dangers of training artificial intelligence with data from the Internet.
AI training with duplicate data?
Nevertheless, research harbors hopes. Because experts are currently investigating to what extent training with already known data can improve an AI. The first results make it clear that artificial intelligence can also get better this way. Similar to humans, AI should also understand something different on the second or even third run.
So it remains to be seen which path AI developers will use in the future. If the forecasts are correct, then there are only three to four years left to find a new way. Until then, it remains to be seen to what extent artificial intelligence can map society.
Also interesting:
