Artificial Intelligence

Like Toddler Training Milestones

 Era of AI Rush started and like Gold Rush people of different walks of life are rushing towards Artificial Intelligence. Tremendous volume of data laying across their fields and engineers of respective fields are there to mine the data, to interpret the data, to reach out the solution for existing problems and sometime to identify the future problems.

Being a Database technocrats for the past 16 years and dealing with huge volume of real time data my curiosity towards AI keep raising these days. When there are millions and trillions of lines are being coded to get the data feed, data extracted, data cleaning, data analysis and in reports publishing, what is so different in AI? Isn’t it just another set of programs? Before getting into the answer let us discuss a little more about human intelligence. 

One of my friends daughter who is around 1 year old started speaking. She has been taught to identify fruits like Apple, Orange, Banana, Mango etc. I visited them recently and the little one said “Juice” pointing me a raw mango. Probably she wanted to have Mango juice. My friend and others in the family told her that it’s raw mango and with ripen only one can make juice. She seems to be confused and was keep gazing at the raw mango. I picked a ripen mango and made her to touch it and feel the difference between a raw and ripen mango and the colour differences. Soon she will learn to identify the raw and ripen mango and such differences among the fruits. When the kids are of age around 18 months they get the ability to recognize different colours, the similarities and differences in shape, size, texture, etc. 

So are Machines and Computer Systems. Machines are trained with adequate data to identify and classify.

When a child is growing, everyone around teaches the child and trains her/him repeatedly on various tasks. After certain age the child without anyone giving training starts acquiring knowledge by what has been taught, through observation, reasoning out, performing the task, analysing the output and correcting itself. Learning by self the child starts performing task like watering a plant, kneading dough, identifying which are edible and which are not, climbing the stairs with proper body balance, eating, talking, reading, writing etc. For every task the child performs, it understands that it is the best action from the appreciation it receives.

Likewise the Machines or Computer systems can be trained in learning on it’s own, to reason out or to draw inferences appropriate, respond to the situation and in problem solving. Machine learning and training enable them to perform familiar task in familiar environment without any pre coded program or human intervention. Apart from the training understanding of Languages, Speech Recognition, Gesture Recognition, Perception are the other cognitive processes that aids Machines in Learning. Machine are trained on taking the best action considering the current state in the given environment or situation. The reward structures are designed in such way that it will help the machine to identify the best action. The target is to Maximize the reward. 

In Machine Learning which is a subset of Artificial Intelligence a suitable algorithm is selected and with the available dataset, mathematical model are built, trained and deployed to predict on new dataset. For example the MNIST dataset has the collection of 60,000 images of hand written single digit numbers. Using K-Means algorithm model training can be done to identify any handwritten digit.  

In Nov 2018 Stanford researchers have developed a powerful deep learning algorithm called CheXNet that evaluates chest X-rays for signs of disease. In just over a month of development, their algorithm outperformed expert radiologists at diagnosing pneumonia.

Such development of Machine learning or Deep Learning algorithms and creating models based on these algorithms for specific purposes are very small sect of AI. With continued improvements the model performs human task with higher level of accuracy in a shorter span of time.

Similar to toddler training; AI based Computer Systems are trained. Toddler will grow into an adult and starts performing, out performing in the unexplored field with the help of inborn intelligence. So are the Computer Systems with complex and powerful AI algorithm will acquire such intelligence in near future.