Robots are forced to "think"

One definition of the notion of “thinking” that can be found is:

Thinking is the ability of a person to reason, which is the process of reflecting objective reality in ideas, judgments, concepts.

In other words, the process of thinking is inextricably linked with objective reality through the ideas and concepts with which it operates. For a robot, this means the need to “see” in the literal and figurative sense, as well as “understand”.

Thinking robot

So what is thinking, and can you teach this robot?

“To teach a robot”, “a robot thinks” ... WHAT ???

Based on the definition of "thinking", you can precisely think to teach. This is confirmed by the most successful solutions in the field of Artificial Intelligence (AI), which use the ability to learn systems.

So why do we need a thinking robot

A thinking robot is the same robot that a “man” expects. A robot with which it will be enough just to interact with an ordinary person and who will be able to interact with a changing human environment. Today, the creation of such a robot is a huge challenge for any company.

Any task in the human environment, for example, at home, becomes impossible now even for the most sophisticated robot.

For example, let's look at the command “Prepare tea”. To understand, even simplify it to “Put the kettle”. This is a highly abstract team that is completely familiar to humans. “Thinking” and “thinking” are completely ordinary and understandable for a person processes and often many tasks people perform on a subconscious level, i.e. without thinking consciousness.

Let's try to solve it consciously. Before performing this task, you must answer the questions: “What?”, “Where?” And “When?” Need to be done.

• when? - there is no answer and no indication, therefore, probably, without delay
• what? - the answer is in the problem, but not explicitly: put the kettle = turn on the kettle with water . The formulation of the task “to put the kettle” for a person unfamiliar with such terms will be a significant complication. It means that the robot needs to know the terminology used in this field. It is obvious that in our understanding now these phrases are very close in meaning. It turns out the robot must not only know the words (terminology), but also understand the meaning of whole phrases. More on this later.
• Where? - there is no answer, but a person can think like this: “the kettle is usually in the kitchen ” or “I saw the kettle in the kitchen ”. And if the robot does not think, then for him a similar answer is not available. The robot, in order to find the answer, you need to know the terminology, to be able to use the visual perception of the world and to be able to perceive and process abstract concepts.

Abstract thinking

So, what does “abstract” or “abstraction” mean?

Abstraction - mental abstraction, separation from those or other parties, properties or relationships of objects and phenomena to highlight their essential features.

It turns out that abstract concepts in the language are inextricably linked with terminology.

Today, there are quite good language models for translation and speech processing tasks, and very often they use the model of the vector representation of words, i.e. terminological model, for example, word2vec. It is not without flaws, but it is very curious that it allows the robot to “learn” the understanding of unambiguous words within one terminological field, and can serve as some figurative map of abstractions.

Such a model, for example, will allow the algorithm to go through the chain:

Milk → Refrigerator → Kitchen

or, as in our case:

kettle → kitchen

Existing language models also use other approaches, for example, various models of recurrent RNN networks for natural language processing tasks. Such models are able to learn to highlight the meaning and, after training, say:

put the kettle - turn on the kettle with water

So, now the robot knows what to do: it needs to immediately turn on the kettle with water in the kitchen, but does not yet know how.

How does a person usually think? It is necessary:
')

1. Getting to the right place - Kitchen
2. Find your desired object - Kettle
3. Find a button on the kettle
4. Press the Kettle power button

Each of these tasks must, among other things, be turned into movements of the legs and arms or wheels and manipulators. You may notice that in the process of execution a lot can go not according to plan. A person simply solves minor difficulties:

• The door to the kitchen is closed - open
• The door does not open - ask people around or call a specialist
• No water in the kettle - tap water
• On the kettle is not a button, and the lever - pull
• There is no kitchen - a kettle somewhere else - we are looking for an item that looks like a kettle
• Etc.

And how will the robot act?

Robots are still learning to perform their functions in close proximity to humans. They learn to act using an action planning algorithm and abstract thinking. In this process, robots explore their capabilities, as well as explore the world around them. First, by random means, and with the advent of experience, they strive to act in the best way to achieve their local goal.

Functions performed by the intellect

You can see that the robot has a lot of algorithms in its head. These are the algorithms of perception of the surrounding world, the learning algorithm, various security mechanisms and much more. Also, the robot needs space for learning - even robots need school. To accelerate early learning, various virtual and simulation environments can be used.

What are the functions performed by the intellect that a robot needs?

Depending on the purpose of the robot, different intelligent algorithms may be needed. Among them may be algorithms:

• Visual perception of the world
• Orientation in space
• Communication (voice or data)
• Perception and processing of abstract concepts
• Planning
• Decision making
• And etc.

The process of creating thinking robots is quite complicated, but also very interesting. Each robot must include software and hardware stuffing. Many companies in the world are working on the creation of robots and their various parts. Some produce components, such as drives and various sensors, others are developing new types of drives and sensors. Someone teaches their robots to walk and tumble, others try to endow them with anthropomorphic signs. There are still companies that are working to endow robots and other devices with intelligence.

We all still have a long way to go, and on this path those who are not afraid to dream and who are striving to work together, in partnership with other teams, will be able to achieve the most serious results.