A computer program is said to learn from experience with respect to a class of tasks and a performance measure
P if its performance at tasks as measured by
P improves with experience beyond a baseline accuracy of: "Guessing the most frequently occurring correct answer."
Perform a market demanded task at an acceptable speed, accuracy and hostile-user tolerance, that it becomes a source automatic income such that the cost of a computer, electricity, software, internet + maintenance is less than the employment alternative. The money fountain comes from people who want to pay less for an adequate service.
3Blue1Brown, how trained neural networks simulate human abilities: http://www.youtube.com/watch?v=aircAruvnKk
Machine Learning is just computer programming. The annoying part is data transformation from a given arrangement to a needed arrangement. The reason
y=mx+b from Algebra and the derivative from Calculus I-III is the beginning of machine learning is because the slope of a tangent to a point on a curve is the extractor of information gain that separates a superior brain from the inferior. The algorithm that the universe uses to create life is "The partial derivative from Calculus, of 3 dimensions, with respect to one, deriving a hyperplane of max separation, nudging weights on the signal processing molecular engines in your neurons.
CGP Grey: Genetic algorithm simulates human abilities: http://www.youtube.com/watch?v=R9OHn5ZF4Uo
Machine learning principles have been well-known since the 1960's. The difference now is a budget computer with an NVidia 2080TI GPU, (or cheap EC2 instance) gives you more 32-bit floating point operations (10-20 TFLOPS) than ten million dollars worth of computer in 1990. So now a middle class amateur can isolate a model that outperform the models that even the best humans make for themselves at narrowly defined tasks.
To fast track machine learning: pay attention in all computer science and math themed classes:
Algebra(scatterplot of y=mx+b),
Trigonometry(expressing 3D rotation with theta, sine/cosine/tangent),
Discrete and continuous mathematics(Algebra of relational logic and causality),
Probability(Bayes, summing information gain),
Decision theory(Tree, Graph, HashMap),
Calculus 3(dy/dx derivatives) and get good at intuition for transforming math to computer code and back.
The reason machine learning is good is because it's the next step of evolution. The human brain grows like a cake over the centuries with the outer surface layer the newest. The next layer is man machine hybrids as Wozniak-2 hammers out physics-engine mirrored overlays between neural lace, bluetooth to Phablet, 4G to a stack of air conditioned Nvidia GPU's inhouse, or over a marketplace of the same in a data center.
In 2100, surplus cognitive energy from said machine hybrids join into a planet-sized brain that invents fusion+nanotechnology needed for: http://youtu.be/LRqsCB-K784?t=71