What is AI

The 3 eras of AI

AI has embraced very different technologies during its history:

classical code if → then → else
expert system using human-made rules
statistical algorithms machine learning the rules

AI is machine learning

Knowledge is extracted from data. Machine learning is a combination of:

statistical algorithms
systematic experimental process

Basically, it's experimentation with algorithms.

Principles of machine learning

The model computes predictions. E.g:

linear regression
decision trees
SVM The optimizer tunes the model to reduce the prediction error. E.g:
gradient descent
genetic algorithms

Differentiable programming

Machine learning by gradient descent is an optimisation of a differentiable function:

A differentiable function allows to compute the error gradient
Each iteration, the gradient shows how to modify the model parameters to reduce the error.

Note

If the function we want to optimize is not differentiable, we use other optimization algorithms (such as random search, genetic algorithms, bayesian optimization, ...), it is then called black box optimization.