An Accurate Classification of AI Technologies

Artificial intelligence has become a huge thing in the past years, and we know how it all is with landscapes, distinctions, and infographics: they all show how people think about AI. Here we’re not going to talk about these categories, since putting things into boxes don’t really mean we have a clear image about them.

We believe that what we’re about to do will be useful for those who are new to this topic, just as much as it will help those with some experience on their backs – they will all be able to start and keep a conversation about AI alive. We’re going to call this your way to pre-existing knowledge on this topic, a way that will let you choose info and even make new knowledge on artificial intelligence.

There are six AI paradigms:

Tools based on knowledge, which are mainly based on databases full of rules and information

Tools based on logic, which are the ones used for knowledge representation and for solving problems

Probabilistic methods, which are the tools that lets agents act in some incomplete information situations

Machine learning, which are the tools that let computers learn from specific data

Embodied intelligence, which is basically an engineering toolbox that shows a body that’s required for higher intelligence – we’re talking about at least a part of functions, like perception, interaction, movement, visualization.

Search and optimization, which make the most out of tools that let intelligence search with lots of possible solutions.

All of these 6 find themselves in other three different branches, called macro-approaches – Symbolic, Sub-symbolic and Statistical.

• Symbolic – human intelligence can be reduced to the manipulation of symbol
• Sub-symbolic  – it doesn’t really have a specific representation of knowledge given ex-ante
• Statistical – this one is based on math and math tools in order to solve the sub-symbolic problems.

There are five problems that AI has been used for:

1. Reasoning, which is the capacity to solve the problems
2. Knowledge, which is the capacity to represent and also understand the world
3. Planning, which is the capacity to set goals and then achieve them
4. Communication, which is the capacity to understand the language and them communicate with others
5. Perception, which is the ability to transform the sensorial inputs, such as sounds, images to information that can be used in different cases.

The technologies are divided into two groups: narrow applications and general applications. Of course, if you’re new here, is crucial to understand the difference between weak and narrow AI, strong and general AI and artificial superintelligence. The artificial superintelligence is a theory that’s up to date, the general AI is a goal that’s to be reached and the narrow AI is what we have in the present, and that is a set of technologies that aren’t really able to cope with the things that are outside their scope. Some of these technologies can solve a specific problem, some better than humans (we’re talking about the narrow applications), and others can deal with multiple problems now or in the future, some better than humans (we’re talking about general applications).

Which are some of those technologies?

Robotic Process Automation, which is the one that cites a list of rules to take into account by seeing someone doing a specific task.

Expert Systems, which is a computer program that has some rules to outdo the way humans make decisions. An example of this is fuzzy systems which use values between 0 and 1, and not the traditional logic, which only gives the 0/1 result.

Computer Vision, which are methods to get and understand digital images

Natural Language Processing, which is a subfield that deals with natural language data – language generation, language understanding and machine translation.

Autonomous Systems, which is another sub-field that is placed between intelligent systems and robotics – agile object handling, intelligent perception.