Brain Simulator is one of GoodAI’s in-house software platforms that we use for our experiments. You can also check out Arnold Simulator.
Design your own AI brain on Brain Simulator: Click here to download our installer or access the project on GitHub.
Please note that using Brain Simulator requires an NVIDIA CUDA-enabled graphics card and 64-bit Windows.
Latest Brain Simulator update – School for AI: check out our Brain Simulator release notes and Brain Simulator documentation.
Create your own AI Brain
Brain Simulator is a collaborative platform for researchers, developers, and high-tech companies to prototype and simulate artificial brain architecture, share knowledge, and exchange feedback.
The platform is designed to simplify collaboration, testing, and the implementation of new theories, and to easily visualize experiments and data. No mathematical or programming background is required to experiment with Brain Simulator modules. GoodAI will continuously improve the platform based on its own research development and user feedback.
The ability to rapidly prototype architectures of artificial brains and instantly test different hypotheses is paramount to keeping our research dynamically moving forward. This is why we created Brain Simulator.
On this platform, a researcher can experiment with existing AI modules (e.g. image recognition, working memory, prediction, motion behavior generator, etc.), modify them, create new ones and link them together. The resulting AI agent can observe, interact, and modify the simulated environment. No mathematical or programming background is required to experiment with Brain Simulator. We replaced AI programming with AI designing.
AI agents designed in Brain Simulator operate on sensory input in a simulated world, interpret and change their environment, and can communicate with each other.
Apache 2.0 License
Brain Simulator and its products, including AI modules and brains, are under the Apache 2.0 license. Apache License is a standard open source license (OSI-approved). While previously Brain Simulator and associated modules could only “be used only for non-commercial, educational, research, and non-evil purposes,” this new open source license allows for commercial use, modifications in source form, distribution, and private use – as long as what you do with Brain Simulator and the associated modules includes the GoodAI copyright, license, notice, and states changes. Read more about the new license and full terms here.
Brain Simulator is currently being developed in C# with AI modules and architectures simulated on CUDA. We plan to make the project multi-platform (Windows, Linux, Mac, etc.), cloud-based, and cluster-based (multi-GPU/CPU).
Using Brain Simulator currently requires 64-bit Windows 7, 8, 8.1 or 10, .NET 4.5, and an nVIDIA graphics card with CUDA support (compute capability of 2.0 or higher).
The project can be accessed by downloading this installer, or through GitHub.
Our older proof of concept work is available in Brain Simulator as demos.
Our AI agent learns to play a Breakout game from the unstructured input of screen pixels and reinforcement signals from the environment.
- Signals are triggered by the agent’s interaction with the environment and encourage it to behave in a desired way
- The agent uses visual and attention modules to create a model of the world in its working memory
- The content of the working memory is transformed into a symbolic representation and passed to the goal memory and to the action selection network
- The agent devises a strategy for completing its goal and adjusts it to increase positive reward
Our AI agent completes a series of actions to achieve a final goal, represented as pressing a light switch inside a 2D maze environment
- The agent’s design combines a hierarchical Q-learning algorithm with a motivation model
- The agent switches between different strategies in order to reach a complex goal
- The agent initially rewards itself for any successful change of the environment, generates a set of abstract actions assigned to any possible change, and analyzes its “experience buffer” to avoid futile actions
- The agent creates a hierarchical decision model and is able to execute a complex strategy, working with a delayed reward