Artificial Intelligence Agents can now improve their own

The way of learning artificial intelligence agents is essential to change. A new research published in arXiv offers a system called PILOT and this system is able to improve themselves instantly when working with artificial intelligence models that perform long-term tasks.

Available artificial intelligence systems often work as follows:

• After the task is completed, the model is updated
• Any intervention or direction change may not be performed during active studies
• Learned lessons can only be applied in future work
• feedback is delayed because experiences are processed after collection

PILOT system reverses this approach. The researchers stand out that self-improvement should be live. In this way, the experience of agents can be used for two purposes: instant improvement of performance in the active task and updating the permanent system architecture.

This method can revolutionize complex, multi-stage tasks. An agent can learn from errors while the task continues, change the strategy and improve its success. At the same time these real-time improvements contribute to better performance in future similar tasks.

This research represents an important progress in the adaptiveity and learning capacity of AI systems. In the coming period, such live learning methods seem to make artificial intelligence agents become smarter, more efficient and more reliable.