Chapter 3 – Apply Machine Thinking to a Human Problem

1. Can a human beat a chess engine? (Yes | No)

The answer is no. Today, the highest level chess tournaments are not between humans but between chess engines. Each chess engine software editor prepares for these competitions by making their algorithms faster and requiring less CPU. In fact today, a top chess engine running on a smartphone can beat humans. In human-to-human chess competitions, the level of chess has reached very high limits of complexity. Humans now mostly train against machines.

2. Humans can estimate decisions better than machines with intuition when it comes to large volumes of data. (Yes | No)

The answer is no. The sheer CPU power of an average machine or even a smartphone can generate better results than humans with the proper algorithms.

3. Building a reinforcement learning program with a Q function is a feat in itself. Using the results afterward is useless. (Yes | No)

The answer is no. While learning artificial intelligence, just verifying the results are correct is enough. In real-life applications, the results are used in databases or as input to other systems. 

4. Supervised learning Decision Tree functions can be used to verify that the result of the unsupervised learning process will produce reliable, predictable results. (Yes | No)

The answer is yes. Decision Tree functions are not intelligent, but they are very efficient in many cases. When large volumes are involved, Decision Tree functions can be used to analyze the results of the machine learning process and contribute to a prediction process.

5. The results of a reinforcement learning program can be used as input to a scheduling system by providing priorities. (Yes | No)

The answer is yes. The output of reinforcement learning Q function can, in fact, be injected as input into another Q function. Several results and be consolidated in phase 1 and become the reward matrix of a phase 2 reinforcement learning session.

6. Artificial intelligence software thinks like humans. (Yes | No)

The answer is yes and no. In early days, this was attempted with neuroscience-based models. However, applying mathematical models is presently far more efficient. Pretending otherwise at this point is hype. Who knows what will happen in future research? But for the time being, deep learning, the main trend, is based on mathematical functions.