AI Agents are software-based agents that learn to perform specific tasks in an unsupervised manner. These agents leverage key AI techniques to master their tasks, such as:
Environment understanding via ‘Visual Cortex’-like deep convolutional networks – making sense of visual input (images, video, etc.) in order to understand the current state of the environment they operate in
Understanding motion and events via recurrent neural networks – essentially, learning to understand how other elements interact with the environment, how they move around, etc.
Policy Modeling via deep neural networks - creating a model for decision making given a state and the expected rewards. This policy, which is learned from experience, serves as the 'brain' of an AI Agent
The learning process of an AI agent is similar in a way to how humans learn – they are exposed to experiences in which they get to take actions and live with their consequences. The results are then used to build an understanding of what works well and what doesn’t in specific situations. Over time, an AI Agent learns to generalize from the experiences it had, and is able to cope with new experiences better. It has been shown, the given enough experiences, an AI Agent get reach and even surpass human performance in specific tasks. However, it is important to understand that while humans master a wide variety of tasks, AI Agents are generally focused at a single task (e.g. AI Agents can master chess but can't also ride a bicycle, play basketball, speak and dance).
What sets AI Agents apart from more traditional software agents (or programs) is their ability to demonstrate intuition-like behavior – coping well with situations that differ from what they’ve seen before. For a traditional software program, anything different from what the programmer had in mind while coding it, could become a serious problem leading to an unexpected results (e.g. bugs, errors, etc.). This difference makes AI agents more reliable and robust and reduces the cost of maintenance – it makes them better fitted to deal with problems that require intuitive thinking.
Comments