Artificial Intelligence Introduction

Basically it refers to software algorithms that are able to learn – becoming better and better at carrying out one specific task as they are exposed to more data.

This article is based on [this video]


AI is defined as the simulation of human intelligence processes by machines espescially computers.

Many times intelligence is vague.

Sometimes binary,
sometimes continuance
sometimes Multidimensional

Some tasks complicated for humans is easy for AI and viceversa

To an extend a calculator or even a hello worlds program is AI .
But they operate within the limit of hard coded set of rules eg artichmatice rules for addition,multiplication.

What we call AI is something which is not limited within the set of rules.
eg: replying hi when some one suddenly turn ups. play chess based on oponents moves.
identify the genre of music the viewer likes (depending on gender and agegroup) and show next song based on that genre.

What we tody consider AI is a subset of what we considered AI earlier, ie machine learning.

Machine Learning

MAchine learning is the use of computer systems that are able to learn and adapt without following explicit instructions.
But by using algorithms and statistical models to analyze and draw inferences from patterns in data.

ie it learns more if more data is given, unlike preprogrammed programs
With machine learning two things are primarly done

  1. prediction based on historical, eg: if an xls of humidity and rainfall is given, it could predict humidity based on rainfall, by understanding its correlations.
  2. classification, identify the category to which a data encountered belongs. eg: feed a million photos of cats and dogs, later when a picture is shown, it will identify wetehre it is a dog or or cat or neither.

Types of Machine Learning

  1. Supervised Learning : We give labeled data, like humidity an drainfall given and cat
  2. Un supervided learning : some random (un labeled data) is given to train machine. AI program checks if anything new could be learned
  3. Semi supervided learning : Mix of above two. mostly labeled data is given, along with some unlabeled data
  4. Reinforcement Learning : Uses an agent and an environment to produce actions and rewards. eg: tell a robot to go by walking and kick a ball. but dont mention each mico steps of walking and kicking. so naturally robot ers a lot initially , go to wrong direction, wrong distance before or after ball, kick with wrong aim etc. when the robot performs exactly as we intend we give a positive reinforcement. tell it that it performed correctly. Thus it learns with reinforment given by master.

Does the computer actually learn?

For humans learning is often accompanied by thinking. But for robots, they dont think.
It is one thing to be intelligent and another thing to be conscious. It workd based on clear set of algorithms.
Ie no machines perform things by understanding things, but only based on algorithms.
Simply doing some complex tasks doesnt mean intelligence. Real intellingence is doing something by understanding what you do.
Machines only simulate intelligence using algorithms.

There are several algorithms for machine learning. Few of them are

  1. kevins clustering algorithm
  2. support machine vector
  3. regression.

Most important algoritm is deep learning

Deep Learning

Uses neural networkds which works similar to neurons in our body.
Big Data (lots of data is required.
The process of making predictions and classifications based on huge data is called deep learning.

Will AI overtake human intelligence?

Exponential Function
Moore's law : Number of components in a computer chip doubles roughly every 1-2 years

Types of AI

There are 2 types of AI

  1. Weak/Narrow AI : only one domain, eg: chess play, image recognitions, sound modulation etc.
  2. Strong AI : Multiple domains mutually connected. for eg: humans call play chess while brushing teeth, identify images etc

Image recognition is weak AI. Its implemented with supervised learning supra.
In social media sites, when you upload photos and videos, you give them more data.
it helps to identify your interests and your peers. show ads suitable for you.
Similar is the case of taging.
Men i kitchen tagged as women, black persons tagged as gorilla etc.

Explainable/Transperent AI

Natural Language Processing

What makes human intelligence differenf rom AI

Intutive learning
Intutive biology
Intutive psychology

Humans predict with Mental Models. AI doesnt have mental models.


  1. will ai remove jobs?

Jobs are usually replaced, not removed

  1. Can AI be creative?

If the requjired patterns are feeded with big data, new mozart music original was created

  1. how far from Human level AI?

Dont fear AI. fear misuse of AI, eg: deep fake videos(porn), hacking self driving cars, voice manipulations.

Moravec's Paradox.

Easy to train computers to do things human find hard and difficult to train which humans find easy.