Hackernoon logon5 advantages of the top-down approach in the creation of AI by@AlexanderHolovan

n5 advantages of the top-down approach in the creation of AI

Author profile picture

@AlexanderHolovanAlexander Holovan

Nowadays, the progress in the development of the neural networks has shifted the focus in the creation of AI towards the «top-down» approach. At the same time, we may notice a certain decrease of the advancement pace towards this approach.

Our team, mainly, consists of the psychologists and the psychoanalysts. For many years we have been modeling the mental processes in various IT-products and this predetermined our choice of strategy in the creation of AI. We have chosen the «top-down» approach. Similar views on this issue of such authorities in AI technologies as Marvin Minsky and Seymour Papert strengthen our confidence.

In the article I will constantly compare these two approaches. We assume that among the solutions which must form the basis of a strong AI have to be such that founded on the first, and rest upon the second.

In the course of work on the AI creation, we have already received the confirmation of the advantages of the method of copying the human psyche’s structures and functions:

1. The teaching process

In order to receive the high-quality result it must be conducted the labour-intensive training process of the neural networks, but only the highly specialized experts can prepare the training data. But the teaching new information of the AI, which is created according to the top-downapproach, realizes in the absolutely natural form for the person — by means of conversation.

In addition, the standard method of teaching new knowledge and building skills has both — a developed methodology and a large number of trained specialists.

For example, to train our AI, we, basically, attract the simple teachers. The very process of training passes in the form of the procedure that is very close to the child’s education.

2. Understanding.

From our point of view, one of the most important features of the strong AI is the ability to understand a person.

The methods used to teach neural networks do not allow us to assume that AI, built only on their basis, will come to the same understanding of beauty, greatness and sin in the way as a person understand them.

Using the top-down approach, the way of expressing the structures of the human psyche in the AI assumes the forming of these categories in its conceptual field. The AI acquires the ability to identify and to operate them as a person does. This is particularly noticeable in the AI’s understanding of the context of the speech.

I am pleased to see how our AI, after seeing several phrases of the interlocutor or reading the text, makes the same conclusion that I would make. Despite the fact that I perfectly know all the algorithms which form the reaction of the AI, I feel an absolutely irrational joy.

3. The ability to use the information, a person has, to solve the problems without a preliminary preparation.

Thanks to that AI operates the same categories as people do, it gets the opportunity to use the enormous amount of knowledge, accumulated by mankind, about particular objects, people and phenomena. The content of this information, the form of its receiving, storage and reproduction — all this is «adapted» for a person and all this can be used by the AI almost without any restrictions.

The present level of development of our technology already allows us during the dialogue to give information and to receive the correct answers to subsequent questions. Moreover, for the teaching we already use the texts written for children.

4. The ability to monitor the development of AI.

The neural networks represent, in general, a black box. There is no effective way to control the process that takes place inside the network during its learning. We are only able to get the results obtained at the end of the process.

The method of copying the structures and processes of the psyche, on the contrary, makes it possible to monitor them, since the content of these processes and structures correspond to human ones.

In our work, the quality of the information comprehension verified during debugging procedures, the content of the conclusions made by AI, allow us to identify even the minor deviations from the human thinking. Understanding of the structures and processes in AI makes it possible to predict with a high probability its response to the specific information as well as, in general, the prospects for its self-development.

5. Creative work

It’s not about solving problems and maintaining a dialogue. It is exactly about the creative work, the creation of the fundamentally new. The presence of the familiar elements in the new works gives us a possibility to evaluate such creations. The common basis, consisting of the identical concepts and processes, permits us to hope that the AI’s creations can be appreciated by a person and enrich his world.

The conclusion about the possibility to assess the creative work of AI, created by the top-down approach — is more an assumption. We make this assumption on the base of the understanding of the structure and features of our solution’s functioning. We have become accustomed to the fact that AI produces the understandable reactions, albeit rather primitive, as a child of the corresponding age. We observe how these reactions become more difficult — the child grows. This formed our expectation that when it will become more adult, we as well will be able to understand it and appreciate the results of its work. I, for instance, do not presume its answer «42» in the future. But so far we had no opportunity to test this hypothesis because of the «young age» of our AI.

We see that the «top-down» approach in the creation of AI has significant advantages: it makes up a number of the deficiencies in the «bottom-up» method and we assume that it, along with the neural networks, will make its own contribution to the creation of a truly strong AI.

Tags

Join Hacker Noon

Create your free account to unlock your custom reading experience.