AIA: Artificial Intelligence Development Without Neural Networks

A new approach in the field of artificial intelligence

Yes, I understand that the name is template, but the essence of solving the problem differs from the standard concepts accepted in the world. I must say right away, the review article, algorithms and complex mathematics will not be here. But the client application working on the basis of this project will succeed. But more about that below ...
A few words about us, we are a group of programmers - Treevelopers - are developing to create artificial intelligence. We’ve been working on this for the second year now and have planned at least as much. We called the project Artificial Intelligence AIA or just AIA.
The goal of the project is to create a universal module to which any device capable of providing signals can be connected. At the output to the module, any devices capable of responding to signals received from the module are connected. After some time, the system adapts and begins to behave “logically”.



Artificial Intelligence AIA module and input / output devices

Input
Input data is itself an extensive topic of conversation. Indeed, for example, the same sound can be represented by graphs, in fact, that have nothing in common with each other. And if you reproduce this data with the appropriate devices, then the person hears is one and the same. But for an artificial intelligence module, this is far from the same thing.
I'm not talking about the fact that the input data can be different in nature. For example, it could be a video signal or an audio track with a bunch of spurious noises.


Input signals for the artificial intelligence module

A lot of literature has been written on the topic of each type of signals and there are algorithms for their corresponding processing. But the essence of the artificial intelligence system is that it works without an initial knowledge of what input will receive what signal.

What is the revolutionary and news of this approach?
Until now, the creation of artificial intelligence has been closely linked to neural networks. That is, with modeling the brain itself. This approach in some cases justifies itself, but so far it is far from the desired results.
As history shows, humanity has always loved to copy ideas from nature, the same thing happened with neural networks. But sometimes there are times when copying in nature is not the best approach. That is why we are now flying not on huge flies with fluttering iron wings, but on airplanes.
So what's the news? - Yes, the fact is that we are not modeling the brain itself, but its behavior - from the biological level to the level of abstract thinking. The second, by the way, is the most difficult, but at the same time the most interesting.

Recognizing the world around us is the road to intelligence.
Artificial intelligence needs data to process. After all, why intelligence, if there is nothing to process? But in our case, this is not raw data coming from the outside world. We present intelligence with data structures, which include processed data that describe certain phenomena that occur in the world around us. Another module, which we call the adapter, is engaged in preliminary data preparation for intelligence. Intelligence is only involved in making decisions based on previously obtained data.

Key Adapters Functionality

Pattern Recognition
In a state of uncertainty, the first step towards adaptation is pattern recognition in the input. A pattern is a basic specification of similar groups of signal sequences. Thanks to special algorithms, patterns are changed during training to acquire their natural form.
This functionality is partially included in AIA 1.0. It is thanks to her that ArmMen V1.1 (a client application for demonstrating the capabilities of AIA) can recognize previously heard words. Moreover, the words can be in any language, and it does not matter who said them before. Matches will be found regardless of gender, voice timbre or age of users.

Object Recognition
At the next level, the patterns found form a data structure similar to the human representation of an object. Recognized objects are transmitted by the intellect by one of its input channels.
Recognition of actions
The next level following the recognition of objects is the recognition of actions. The outside world is constantly changing. The properties of objects are also constantly changing. Recognition and registration of these changes is necessary to get an idea of ​​the current state of the outside world. In fact, this level turns the static world "at the moment" into a dynamic one, i.e. evolving over time.

Match Recognition
To understand the world at the level of interactions between objects, you need to follow the patterns of sequences of actions. This is a kind of analogue of templates over objects, but at this level, templates are already applied to actions, not objects. These patterns also form data structures similar to the human notion of actions. This level allows you to analyze the behavior and reaction of the environment (natural physics, the behavior of intelligent creatures, the concept of time and space).

Simulator
One of the main problems of artificial intelligence is its training. Even we humans, as the most advanced “non-artificial intellects,” spend years learning. What to expect from the artificial?

Artificial intelligence placed in some environment also needs some time to learn. In this scenario, learning will be measured by the rate of change in this environment. And if this environment is the real world, then in most cases it will be too long.
The process can be accelerated using simulators. They are programmed to simulate an environment in which the use of artificial intelligence is planned. AIA is placed in the simulator, which, during the simulation, learns much faster than if it were trained in a real environment.
Although training in a real environment in itself can be useful in many cases, for example, in the study of new planets. That is why training in a real environment, although slow, still has its uses.
The best result is a combination of both methods. In fact, the data of the simulator can be constantly updated with data collected from other similar artificial intelligences that are trained in a real environment.

Artificial Evolution
Artificial intelligence devices placed in different habitats will have a certain amount of data after some time. The creation of a new similar device will be accompanied by the selection of data collected by their previous generations. Moreover, a new portion of data for artificial intelligence can be synthesized not only on the basis of data from previous generations, but also on the basis of data from other similar devices. Moreover, AIA can also deal with this synthesis.
Thus, we are approaching a new level of modeling - modeling of human communication and natural development.
We are still very far from this level, but it gives a clear understanding of what we should strive for.

A few words about ArmMen V1.1.
At the end of the article I would like to talk a little about the ArmMen client application , with which you can test the functionality of the AIA artificial intelligence module version 1.0 online. The current version of ArmMen only supports audio processing. Processing of the video stream is planned in the upcoming version.
What can ArmMen do?
• Execution of previously recorded user commands
• Completion of sentences previously heard
• Recording the audio track and sending it to the server
• Playing back the server’s response on the client side
At first glance, it might seem that the system simply looks for text matches using long-known algorithms that were implemented in prehistoric phones. But in fact, this is not so, and this can be seen by checking ArmMen-a teams told by different people. For him there is no difference in what language the team is, gender, age, timbre, etc. are also not important. In exactly the same way as for us it is completely unimportant to communicate with a child or an elderly person, we equally well understand each other. Although ... there are exceptions of course.
For full ArmMen testing I highly recommend using a good headset and using a Chrome browser.