AI notes: The natural evolution of the algorithm

Machine Learning is the next natural step for artificial intelligence, it has been described as applied statistics, advanced algorithms, but is a mix of all of this. A.I. is about processing data to understand it and react intelligent to it. In some way M.L. is the natural evolution of all we have been done.

When we write software, we usually build the decision making directly into the program. For example if we want a program to recognize my face, we could say: Look for black hair,  look for brown eyes,  look for glasses, etc. But in M.L. we build a model that could look at a bunch of images over time and we say: This face is Victor, this face is Brian, this face is Paco, this face is John, this face is Carlos. Then, we give to the model a new image. And based on the previous experience the model can recognize who is. So the key of this is the data we build makes M.L. very useful. M.L. can be applied to every domain: Improve business decisions, identify diseases, distribute water in cities, coordinate  traffic lights, search planets, etc (a big).

There are several difficult problems in AI

– –  intelligent models have limited resources
– –  Computation is local, but problems have global constraints and influences, how to deal with that?
– –  Logic is deductive but  many problems are not
– –  The real world is dynamic, but the knowledge is limited
– –  Problem solving, reasoning and learning are complex problems, but explain and justify that problems are even more complex

Characteristics of AI problems

– –  Knowledge and data often arrives incrementally, so the scope changes
– –  Problems exhibits recurring patterns, many of them hidden from human view
– –  Problems have multiple levels of granularity
– –  Many problems are computationally intractable
– –  The world is dynamic but you face the problem from the static point of view, because your knowledge of the world is static or at least does not change fast as the problem you want to resolve
– –  The world is open – knowledge is limited

In general AI is about manage the uncertainty, therefore make sense use AI when you don’t know what to do.

Intelligent behavior process

We have 3 forms of process which are intrinsically connected to perform intelligent behavior. These are, reasoning, Learning and Memory.


– –  Understand natural language sentences
– –  Generates natural language
– –  Make decisions
Reasoning is a fundamental process of knowledge based on data


– –  Answer the right questions and store the answer somewhere
– –  Connect and amend the wrong answers.


– –  When you learn something, that knowledge that your learnt has to be stored  somewhere (memory)
– –  If you are going to reason using knowledge then, that knowledge has to access from memory
– –  The memory will store what we have learnt as well as provide access to knowledge for reasoning

The process of reasoning, learning and memory, constitute a unified architecture. Each form of process feedback in both directions, therefore: We learn because we can reason and then, the more we know, the more we can learn, and so on.  There are many knowledge theories based on AI that unify reasoning, learning and memory. The idea behind to put this 3 elements together is called deliberation. And deliberation is only 1 part of the overall architecture of AI agents knowledge.

Naming few technologies:


Analize large amout of data and determine patterns from that data. they often are optimal, but necessarily human like


Autonomous Vehicles, they act normal in the real world, but they also need to act optimally


Intelligence Robots can dance with music and behave like humans


Smart Web is a new generation of web technologies, in which the webpage understand the information on it.

AI notes, are my personal notes on machine learning. Mainly from the course Machine Learning Nano Degree course from Udacity, and other sources.
You can follow the experiments and the research in my GitHub account:

Leave a Reply