DEV Community

Cover image for Appreciating the "Learning Problem" - Why AI will never replace your job
Nukala Suraj
Nukala Suraj

Posted on • Updated on

Appreciating the "Learning Problem" - Why AI will never replace your job

so... let's play a game !!!

i'll show you a series of patterns, and each pattern maps to a value of either 0 or 1

a game of puzzles

now consider this pattern, which number would this pattern map to?

a question from the puzzles

if you said 0 you're probabily right

if you said 1 you're also probabily right

¯\_(ツ)_/¯

the point here is that whatever your guess was...

  1. you had no idea what each of the tiles represented
  2. you had no idea what the co-relation between the tiles was
  3. you had no idea what 0 or 1 represents

just like AI and this guy here

meme

he has 0 understanding of what the puzzle means and what his guess means, but he made the correct guess

The Learning Problem

the game we just played right now, is presicely what every ML/DL/RL/AI algorithm does

  1. take data that has been cleaned/processed (into a puzzle)
  2. find a pattern in the data
  3. use the pattern on new data (it has never seen before)
  4. hope it gets appriciated

but keep in mind that not every problem can be solved this way

because...

  1. not every problem has (enough) data on it
  2. not every data has a pattern in it

some of them also/already have a (not so) simple mathematical answer to them, you don't need rely to finding patterns

but for the rest, they can probabily (maybe not efficiently) be solved by having a computer find a pattern AKA learning from data

How does a Computer Learn from Data

short answer, you solve the learning problem 😝

long answer to solve any learning problem (ML, DL, RL)👇

the learning process

  1. unknown solution: think of this the true pattern/source of all the data we have
  2. training data: data that we have collected and processed so a computer can find a pattern
  3. learning algorithm: the actual algorithm that refines a model so the model learns from the data > eg: Perceptron Learning Algorithm, Backpropagation Algorithm, etc
  4. hypothesis set: a set of all possible models that can find a solution > eg: Perceptron, Neural Networks, etc
  5. final hypothesis solution: a pattern with the heighest accuracy

while this IS the long answer to solve any learning problem

this long answer lacks details...
the details which would be dealt with in the forthcoming blog articles

originally posted on my personal blog

thank you for reading
do let me know what you think in the comment section
✌️

Top comments (0)