Introduction
Modern day Artificial Intelligence depends hugely upon Neural Networks or some versions of it. It is is a learning algorithm that is inspired by the structure and functional aspects of biological neural networks. Although the concept behind is nice, but the way it is being implemented might need a change of mindset.
How Humans see things
As humans, we intuitively know that pictures have a hierarchy or conceptual structure. For example,
The ground is covered in grass and concrete
There is a child
The child is sitting on a bouncy horse
The bouncy horse is on top of the grass
We recognize the idea of a child no matter what surface the child is on. We don’t have to re-learn the idea of child for every possible surface it could appear on. With AI systems, this is very difficult to achieve
How AI does it
For an AI to identify a child in any surface, it has to be fed with a lot of child images and I mean a lot. This is called the Training Data. There are opensource projects which request people all around the globe to send images of birds, elephants, horses, children and what not! These images would be used as training data for the AI to identify a bird in a cage or an elephant in a zoo among other animals, tiger in a jungle. For me, coming up with a huge amount of training data or preparing the training dataset manually is exactly opposite to the concept of ingenuity and brilliance that our previous generations has striven for.
How AI should do it
I believe, the learning should be done by combining AI techniques with the state of the art robotics technology. We need to have a contraption that looks like a pet dog or a cat following us around. It will take images of the environment and process that info. Let’s call it our pet AI. Some logic need to be built in to the pet AI so that it knows how to respond to actions from the environment or how to react to environmental inputs so it doesn't get hindered during navigation at every step. If it gets blocked by a wall, then it needs to find a path on its own or by following its owner, maybe from home to outside and even on to the car. When this AI pet finds a shape which seems interesting and nothing that matches its Database, it will ask the owner what it is (maybe show the question in it’s forehead screen or something similar), which the owner can respond to by voice or through a chatting device. Conversely, the owner may ask the AI pet about things in the environment, which the pet responds to. If it guesses correctly then fine, otherwise it will be corrected, maybe through voice or chat input. The best part about it is, we can gather the learning of the AI pets all over a town, and combine the results to build even more powerful AI systems and it will improve gradually.
Summary
We will have a swarm of pet AIs collecting info and updating a common database that will be constantly updated and upgraded.
Top comments (0)