Algorithm Driven Vs Artificial Intelligent System

During a short tea break at the office yesterday I recognized that many of our colleagues in the tech industry still struggle to differentiate between AI (artificial intelligence) and linear algorithm. I told them that I will write something for them and so I am sharing this with everyone today.

An example of algorithm driven system is like solving a linear equation to find a point in a straight line, say y=mx+c. If slope x, and y intercept (c) are known then the value of y can be calculated. This is possible because characteristics of the line are captured in the equation.

What happens if property of a system cannot be defined by an equation? Or a number of complicated algorithms ? Say for example, a system has to differentiate between: ‘flower’ and ‘leaf’. It's extremely difficult to define characteristics of a flower, as there are some flowers which are similar to leaves and vice versa. Until now only human intelligence and experience can understand these differences.  AI comes to solve these and similar problems.

AI also has a closed loop feedback system where positive and negative results are fed back to the system for further learning. But this feedback system is not similar to closed loop control system. In closed loop control system output is fed back to input for control purpose, but the decision making algorithm is not changed based on input/output.

If a system is required to differentiate flower and leaf, input: ‘large numbers of flowers marked as positive’. Also input: ‘large numbers of non flower items marked as negative.’  From these two input the system creates Neural Networks which contains all the characteristics of a flower

Neural Network is the output of the above mentioned learning exercise (using various proven methods, may be I will write about methods in the next post). The neural network defines characteristics of flowers (like the equation to represent a straight line). Next time, when the system sees an object it will use the generated neural network to identity whether it’s a flower or not. The newly identified flower is annotated and go back as an input to learning process.

MTCNN is an example open source Neural Network for human face recognition that work with Tensorflow. Tensorflow is set of open source mathematical library helps to create Neural Networks