Summary
Presently, AI is predominantly a branch of applied mathematics, not of neurosciences. You must master the basics of linear algebra and probabilities. That's a difficult task for a developer used to intuitive creativity. With that knowledge, you will see that humans cannot rival machines that have CPU and mathematical functions. You will also understand that machines, contrary to the hype around you, don't have emotions; although we can represent them to a scary point in chatbots (see Chapter 16, Improving the Emotional Intelligence Deficiencies of Chatbots).
A multi-dimensional approach is a prerequisite in an AI/ML/DL project. First, talk and write about the project, then make a mathematical representation, and finally go for software production (setting up an existing platform or writing code). In real life, AI solutions do not just grow spontaneously in companies as some hype would have us believe. You need to talk to the teams and work with them. That part is the real fulfilling aspect of a project—imagining it first and then implementing it with a group of real-life people.
MDP, a stochastic random action-reward (value) system enhanced by the Bellman equation, will provide effective solutions to many AI problems. These mathematical tools fit perfectly in corporate environments.
Reinforcement learning using the Q action-value function is memoryless (no past) and unsupervised (the data is not labeled or classified). MDP provides endless avenues to solve real-life problems without spending hours trying to invent rules to make a system work.
Now that you are at the heart of Google's DeepMind approach, it is time to go to Chapter 2, Building a Reward Matrix – Designing Your Datasets, and discover how to create the reward matrix in the first place through explanations and source code.
