This post is as follows:
עברעבר (Hebrew))
New research finds that artificial intelligence systems learn in ways that closely resemble human learning processes, shed light on how adaptive behaviors emerge, and provide potential guidance for building more intuitive AI tools in the future.
Research published in Proceedings of the National Academy of Scienceswe investigated how AI systems can quickly combine fast, context-based inference with accumulated learning. In human terminology, this is roughly equivalent to how short-term and long-term memories interact.
Researchers trained AI systems using Meta-Learning. This is a technique that teaches you how to train a model into a model rather than how to perform a specific task. Over time, AI began to show more flexible problem solving. According to TechXPlore.
For example, in one task, the system was asked to interpret familiar combinations of elements that were not seen before, such as matching colours with animals to identify things like “green giraffes.” This model was trained on thousands of comparable tasks and then managed to make the correct generalization based on previous examples. This type of recombination is considered an important feature of human reasoning.
The study also revealed several trade-offs. AI has better retained information when tasks require more effort or generate errors, just as they are likely to remember experiences that involve problem solving and fixing. In contrast, tasks that were easily completed based on immediate context tended to increase flexibility, but did not contribute much to long-term retention.
Although this study is based on computational models, the findings may help bridge the gap between neuroscience and machine learning. It suggests that certain AI systems may not only mimic human output, but may also reflect similar internal learning dynamics.
AI tools continue to be integrated into decision-making environments and understanding how machine learning systems acquire and apply knowledge, from healthcare to defense, can play a key role in improving transparency, reliability and collaboration with human users.
