Remember when we were kids,we were shown just a few pictures of cats and dogs and from that we could easily spot a new picture of cat? We learnt to classify things we perceiveAlso, we learnt to predict what would happen next in a sequence of events. For instance, in a peekaboo game when someone hid their face with their hands and took their hands away we expected to see their face.As we aged, we learnt not to just go by what we see or perceive, but reason. For instance even though we only saw a cat with three legs we would reason it perhaps lost a leg in an unfortunate accident or was born with a congenital deformity. We would not conclude it is a new species of catLastly we learnt very early to start planning to get what we want from our parents/guardians All these four capabilities - classifying what we perceive, predicting things in a sequence, reasoning and planning are some of the key aspect of intelligence. Machine learning attempts to achieve these capabilities artificially somewhat similar to how we learn (partly inspired/influenced by natural intelligence) - from data. However to date, machines need much more training data than we do. For example, a child can recognize a cat from just a few training samples. Current machines require lots and lots of training. Also current machines can do some sequence predictions well. But they are yet to do reasoning and planning anywhere close to what a child can.