Friday, October 14, 2005

Machine Teaching instead of Machine Learning

Being a disciple of the Machine Learning paradigm, I am not-so-proud to state that what is called 'Machine Learning' these days is actually more like Machine Teaching. Being a student who has decided to dedicate most of his time to his academic endeavours, I can honestly that it is I who is 'Learning.' Each time I learn something new I program it into the little box next to me. Believe me, this box is not learning anything. Until I feed it training data it doesn't really have any motivation to do anything on its own.

The problem is that these classifers and clustering units are very dependent on humans giving them data. It take a lot of intelligence (on the part of the human) to make a decently smart (on the part of the machine) algorithm, but it is far from intelligent.


  1. Yes, humans help humans learn. But independent of other humans, a single human will still learn from the interactions with the environment around them. They will make decisions such as 'stay indoors and use google to learn about airplanes' or 'go to the park and look at trees' or 'go dancing with my significant other.' Depending on what they decide to do, humans will learn different things (but they choose what they want to do).

    However, modern machine learning algorithms do nothing of this sort. They require a human operator to choose what data to feed into them. Algorithms aren't saying to themselves, 'I think that I'll collect word distributions from newsgroups' or 'I will learn how to classify people based on images of their face.'

    It is still the operator (human) that decides what type of data the algorithm will 'see'. A human who goes out dancing and has lots of fun will say to themselves, 'That was fun, let do it again next weekend.'

    Can we really expect algorithms that have no say about the data they will 'see' to be intelligent?

  2. Hi,
    I am a PhD in Adaptive control from Indian Institute of Science, Bangalore India.
    At present I am working as a consultant in Supply chain management.
    I have a son, 13 years old, a high functional autistic boy.
    This topic, 'Teaching to learn' came to my mind when I was thinking about 'How to train him to learn on his own?'. SInce autistic children behave almost like computers, I though of connection bet. the so called 'generalized learning' and training autistic children.
    Pattern recognition algorithms, I feel, is slightly generalized algorithms, as we parammetrize the 'feature vector'.
    I am not expert in learning theory, but I have got to do some research on my own as it 'd benefit my son.
    Please give your feedback to