Ai Dreams Forum

Member's Experiments & Projects => AI Programming => Topic started by: pawel.biernacki on May 20, 2021, 12:06:43 am

Title: Dorban - a demo for the Svarog AI library
Post by: pawel.biernacki on May 20, 2021, 12:06:43 am

I have a new project - dorban, it is a demo for my Svarog library. Dorban can be downloaded from (, and Svarog from ( The whole project is described in (

In the demo there are five cities connected as a graph. There is a vampire in one of the cities. Apart from Dorban (my intelligent agent) there is also Pregor controlled by the player. Dorban can ask Pregor to accompany him, since he believes this will make his chances against the vampire greater. When accompanying Dorban please use "following orders"  menu item, doing anything else means Pregor quits him.

This demo is using the so-called precalculated knowledge. In dorban-0.0.1/svarog there is a file dorban.svarog that specifies the Dorban's behavior in the Svarog language. 
Title: Re: Dorban - a demo for the Svarog AI library
Post by: infurl on May 20, 2021, 01:44:20 am
Welcome back Pawel. I watched your video and had a look over the documentation that you have written. It looks like good work. Your system looks similar to Non Axiomatic Reasoning System (NARS) in some ways. That system also uses beliefs and is designed to cope with limited computing and memory resources. I will follow your work with interest. (
Title: Re: Dorban - a demo for the Svarog AI library
Post by: pawel.biernacki on June 07, 2021, 02:12:00 pm
I have taken a look at the OpenNars. I think I am not in a position to compete with dr Wang who is an acknowledged scientist. I am just a programmer, not even a scientist. Apparently you do not wish to check my program. You may be right – I am aware of its limitations. For instance it does not work well when we have multiple intelligent agents. My algorithm resembles the minimax without the “min” part, it works just for a single agent when dealing with uncertainty and a stochastic environment. It is also an optimizer, not an inference engine. It attempts to maximize the expected value of a payoff function in the next n steps. So my algorithm tries to optimize the agent's situation in a game. It does not answer a single question like "is X a bird", given some knowledge. Both systems use beliefs, that is true, but in Svarog the beliefs represent a probability distribution over a set of possible states. I will try to improve it.