JERUSALEM, Nov. 4 (Xinhua) -- Researchers have shown that an artificial intelligence (AI) model can learn and improve its predictive ability using only data it generates itself, without exposure to real-world information, Israel's Tel Aviv University (TAU) said in a statement Wednesday.
The study, published on the scientific website arXiv, presented a new approach to AI training that could help machines develop learning skills without relying on large collections of human-created material, according to the statement.
Researchers from TAU and Stanford University trained two computer models together, with one creating symbol sequences and the other learning to predict the next symbol.
The first model gradually improved at creating learning exercises that helped the second model become better at making predictions.
The team tested the trained model on several types of information, including language, images, sounds, music, computer code, and DNA sequences.
The results showed that its ability to make predictions improved as more computing power was used during training.
The researchers stressed that the method can't teach AI facts about the real world without exposure to real-world information. However, it could help future AI systems learn more efficiently when they encounter real-world data. ■
