Neurons arranged in the form of concepts

First of all, who knows exactly what intelligence means? Could be consciousness, the ability of learning, reasoning, general knowledge, creativity. But when talking about machines, how is intelligence defined? Back in 1956, a group of old white men described it. They decided artificial intelligence is the neurons arranged in the form of concepts. A set of well-defined problems tested mechanically to solve and give possible answers; a machine that executes activities and self-improves through time; a system that can abstract data, summarize, and identify it. Crazy right? But artificial intelligence, at least for me and what I understood from this week, is training.

Did you know? That our body releases "Myelin"? It is a fatty substance that wraps the axons in our muscles. The axons are in charge of transmitting information to our muscles for performing activities. Through practice and repetition, the myelin covering/ insulation grows, so the energy loss is less and travel of information is more efficiently.

Machines can't release a chemical compound, so they run tests until finding the solution and the most efficient way of getting it. There are two types of data from which a machine can learn: quantitative and qualitative. We mostly use quantitative (numbers, units), and when representing it, we unify and simplify experiences into one word/sentence/description. It is fascinating because we do it with ourselves too. When we use senses, we gather lots of information. For example, when we taste something. We define it just as sweet, acid, bitter, etc. But there are many other ways to describe the intensity of flavors. We abstract and, therefore, lose information. In systems is the same.

There are three types of machine learning:

• Supervised: Data set is split into training and test set. The training set is used to train a model (provided by the algorithm) from symbolic data, and the test set is used to measure how effective the model is ( regression (linear & polynomial), decision trees, or classification)
• Unsupervised: Models learn through discovering relationships and patterns in symbolic data. Labels cannot be created, associated ( clustering (dimension reduction))
• Reinforcement: Continuous iteration of supervised and/or unsupervised learning by a goal or belief. This example is typically used in autonomous control systems

I guess the hardest part of data and machine learning systems is ethics. Remember when a Tesla car killed a woman in the US? Or the machines used in North Korea trained to shoot everything that moves? In the end, someone is behind it; someone codes it. We can say they are just following orders from people above them, but how far does their ethic go?

___

What did I learn?

I learned about AI history, the main concepts used, terms, and a bit of data sets and ow to use them.

Learning process?

• Understand concepts
• Understand how to use Big ML and data sets
• Exercises + exercises