Pattern recognition owned by intelligent agent

Non-intelligent agent can only do jobs preprogrammed, and no or few judgements can be achived. However, intelligent agent is more powerful. For example, a news-clipping agent can be trained with prior knowledge of what kinds of news are serious (for example). Then, it can play in an adaptive way, selecting so-called serious articles and updating its knowledge by feedbacks from its users. AI pattern recognition technology can find its place here. From my point of view, theory used in adaptive testing is available too. Some trials on automatic article scoring have been done, and the related theory and technology can be employed to distinguish seriousness from non-seriousness. I am waiting for my first such kind of agent!