Get Started: AI vs ML vs Data Science

Despite of all the buzz around AI, ML or Data Science, people do find it difficult to understand what exactly these terms mean. When, where and how we apply it. How are they correlated ? This article targets to clear those doubts of noobs and helps you to get started with the journey of AI.


Research in the area of Artificial Intelligence began in the mid 1900s. Back then scientists had come up with different techniques to develop expert systems and tried infusing knowledge base into the machines. It used rule base containing number of if-else rules. Early days involved the practice of LISP Programming, FOPL, Prologue, etc. We all are aware of the famous Turing test which was designed in 1950 to test the intelligence of computers. Even the Genetic Algorithms and Neural Network architecture were formed way back but unfortunately they hardly had any efficient system to train Neural Networks. Now let’s have a look at some modern AI definitions.

Here are few definitions:-
AI (Artificial Intelligence) - Making the machine to act intelligently for a wide variety of tasks.

ML (Machine Learning) - Provides statistical tools which enables us to understand, derive conclusions from labelled or unlabeled data. It can be supervised, unsupervised or reinforced(reward based learning). It enables computer to take decisions without any human intervention.
(i) Supervised - We have past labelled data which helps in training the model. The model is further tested on unseen test data.
(ii) Unsupervised - we don’t have labelled data. We try and find patterns in the dataset like clusters( KNN, etc.)
(iii) Reinforced - fairly new class of ML algorithms. Involves taking random decision to begin with, an expert gives reward/penalty based on correct/incorrect decisions. Aim is to create a model to maximize the awards.

DL (Deep Learning) : Making machine to learn in a manner as the human brain does. It is a subset of Machine Learning. Deep Learning algorithms involves training of multilayered and deep neural networks. This is computationally heavy and only been possible to implement with the onset of General Purpose parallel computing by modern GPUs.
(i) ANN (artificial)
(ii) CNN (convolutional)
(iii) RNN (recurrent)

Data Science is applied in all the above class of algorithms as the pre-runner to get clean/relevant input data for the models.
Before we leap on the technical aspects of the modern AI technologies let’s look at a small story:



There was a six year old kid who was being taught to greet people at his home. It’s his family tradition that everyday regards should be given to elders and causal hello to juniors or peers.
So the way the olden days AI scientists created rule base, his parents started to teach him like —
If you see your Grandfather then give regards.
If you see your Grandmother then give regards.
If you see your father then give regards.
If you see your mother then give regards.
If you see Uncle Amit then give regards.
If you see Aunt Riya then give regards.
If you see Sam then say Hello.
If you see Ana then say Hello.

These rules were fed into the child’s knowledge base and he greeted every family member in similar way everyday. One day Uncle Tarun visited his house. Now this kid got stuck because he wasn’t taught to greet him. Uncle Tarun’s name was not present in his knowledge base.
The conclusion is that the creation of knowledge base will be limited and shall not cover all the cases (unless we hypothetically feed all the data in the world).
Now realizing that his parents followed another approach to teach him. This time like modern AI scientists, they taught him one single rule that whenever you see anybody elder then give regards and if anybody is younger or of same age then say hello. But now the question arises — How to determine whether the person is elder or younger ?

This question’s gonna be answered by the Data Science.
For determining the age of a person his mind needs to perform data analysis on the human data which it has collected in the past six years.
After doing data analysis his mind came to the an equation (simple linear model) : —
Equation :-
value_calc = x * (hair_loss) + y * (face_maturity) + z *(voice_modulation) + p*(height) + c


(Here, x, y, z, c are modal parameters (constants) whose values can be determined as per the regression techniques)
Above charts help in visualizing what the training data set may look like. This is how Data Analysis is done.

Prediction can be made by following the rule : —
if ( value_calc > 15 )
Person = elder
Greetings = regards

If (value_calc <= 15)
Person = younger
Greetings = Hello

#15 is an assumed threshold value

The above if-else is similar to what we call a Decision Function in ML domain, e.g. signum , logistic function etc. We’ll discuss more on this in the part II .
Once we have found all the parameters x,y,z and have our algorithm ready then each time a different person is is witnessed, his features (hair loss, height, maturity and voice) will be perceived and given as the input to the brain, brain will apply the above algorithm to predict if the person is elder or younger and then respond accordingly.
That’s all !

In the above story, the kid is like any computer of this world which has to exhibit artificial intelligence, his parents are like ML engineers and his mind served as a Data Scientist.
Therefore whenever a machine has to be made capable of taking decisions by itself without any human intervention, machine learning comes into play. We can design certain algorithms for the system to evaluate the case based on the given conditions/inputs and come up with a decision or value. Evaluation is done by analyzing the data which requires following different techniques like data cleaning, transformation, mining, etc. ( Also make a point that data analysis and data mining is a subset of data science. Even in unsupervised learning where clustering techniques are used comes under data science)

We try to find out trends and variations in our data set and then finally come up with a model(equation/algorithm) which best suits the trend in the data. Once that model is fed into our system then it will allow the system to apply the same to any given input and then come up with a final decision/result. So the conclusion is that we need to understand data science to come up with most accurate ML algorithm to create the best suited model and what we finally get is an AI application.
More on details of Machine Learning and Deep Learning in Part 2.

Coming soon…

-by Karenite Kell

Comments