Breakthrough in AI can happen by the end of the year

21 октября, 2021 от Kinok Выкл

We are talking about

Recently Deepmind.

This method is indeed very promising. But it requires a significant amount of computing and powerful computer equipment. And this, unfortunately, is not very suitable outside the demonstration games when the goal is not to surprise the public, but the decision of practical tasks in the real world.

As a super test was chosen

The only way to win in such hellish conditions is somehow able to combine non-standard thinking, research skills and good luck. Such an AI agent should be able to combine the optimal application of existing knowledge with the ability to explore completely unexplored areas (i.e., to combine Exploitation & Exploration).

The idea of researchers from Facebook is simple.

Working in the NLE environment, participants will be able to spend more time to test new promising ideas, and not to wait for the results of long-term computing.

Before them is tasked to design and train your agent with how much existing or invented method — with or without machine learning, using any external information, any method of learning with any computing budget.

Among the alternative concepts of creating models of AI agents — IMHO, the most promising is the process model

The essence of this process in:

✔️ statistical generation of predictions (unconscious conclusions about the world around and oneself on the basis of the internal model);

✔- checking these conclusions based on sensory data;

✔️ and constant minimization of prediction errors.

Active conclusion and prognostic coding, combined by the fundamental principle of the free energy of Karl Volonon, I call the «Constitution of Biomathemamatics».

ActinFlab participants believe that the active output agent can succeed in this complex computing problem, which requires a subtle balance between the study of unexplored areas and the application of knowledge available.

To implement an effective active output agent in NLE environment

This competition is the ability to demonstrate the potential for applying active output algorithms to effectively solve open issues in computer sciences.

If you are interested in joining a team that will take on initial planning and subsequent stages of the development of this project, please send an email with a description of your interest at the address

These are exclusive video — recording of three marching speeches of Karl Volonon on

But it is worth it.