NELL, the Never-Ending Language Learner, is software developed by researchers at CMU that reads web pages constantly, forming models and refining existing models, based on newly formed models.
Contrast this with, how we learn a language. While we adopt statistical learning, similar to NELL, we have a huge advantage. Practice. We not only form models, but also test the models by applying it, whether by talking or communicating in another form.
Then, based on the feedback from the practice, from other people, we could decide to refine the model. This 'social learning', is the biggest handicap of NELL.
Also, none of us are purely 'language learners' as it is one of the many roles, human being, husband, son, dad, voter etc. Hence, we are not perfect in learning the language. Moreover, trying to build a perfect language learner machine is an elusive goal.
Also, the 'collective intelligence' that emerges out from community, tell us, at what level we are, relative to others. NELL has no cognizance of whether its knowledge of the language is 'basic' or advanced or expert-level. So, it makes some naive mistakes such as this and makes sophisticated inferences such as this.
So, if we could come to tolerate some degree of imperfection in the machine, then we could create a colony of thousands of knowledge bots, each having a particular emphasis thus pursuing specific sources, and each having a set of domain focus and having a particular set of Goals.
Over time, the entire internet could be infested with knowledge bots and we could come to rely on particular bots, based on our need, the specific task and the trust we place on the bot.
But, if we do not create a colony, we would end up in the same state, for statistical machine learning, as we did with Neural Networks, last decade, i.e., abandoned due to limitations discovered.
NELL needs socnet too, to keep it going, just like you.
No comments:
Post a Comment