VOSAI - AGI
VOSAI AGI is the worlds first AGI with its intent of ensuring it benefits humankind while avoiding all negative impacts.
The intent, purposes, direction, features and enhancements of the AGI are determined by VOSAI Owner, Members and limited Contributors.
VOS means YOU in Latin. This is your AI not some large tech company or government. Therefore - your privacy, safety, and ethics are of utmost concern to us. We will ensure at all costs that no data acquired by the AGI is sold or shared with 3rd parties.
VOSAI AGI is an artificial general intelligence system designed to learn on its own and over time. The AGI system does not modify its own code base but does have the ability to learn code. This is not the goal of this AGI system.
VOSAI AGI system leverages all modern day commonly used machine learning techniques combined with hybrid data stores wrapped up in an enterprise grade architecture.
VOSAI AGI is targeted at natural languages initially. However, the same technique also applies to all mediums including but not limited to visual (e.g. images, video), audible (e.g. sounds), environmental (e.g. gases, heat, flame). The system at its most basic learns much like that of a human - with stimulus and reactions. Much like that of any living creature - the AGI requires extensive interactions with external stimulus in order to learn.
In this section we set forth the technologies leveraged to create the VOSAI AGI. These technologies are subject to change at any time but dictate a starting point for development purposes.
- Intel and ARM based CPUs
- NVidia and AMD based GPUs - no code may be written specifically for one GPU manufacturer
- DPU based processing
- All storage must be SDD or better - no spindle drives can be used
The following diagram illustrates the infrastructure logically according to the architecture of the AGI. The diagram only illustrates the primary hardware and not the supporting hardware. For example, it is assumed there is an infrastructure that supports promoting code (Jenkins), server maintenance (Ansible) and so on.
- APP X (ARM) - used for highly parallelized processing that requires little horse power
- APP X (INTEL) - used for normal component processing
- COMPUTE X - used for training (e.g. neural network processing)
- QUEUE X - used for exchanging messages between systems
- DB X - a combination of an eclectic set of data stores supporting the AGI
The AGI is designed in such a way that it can easily be configured to run within a cloud provider (e.g. AWS, Azure, GCP), within a data center, within the AGI Client running on a desktop PC, any combination of these together or separate.
- Programming languages permitted - Python, C, C++, Assembly
- Ansible and Jenkins for continuous integration / delivery
- Rabbit MQ for message queuing
- Memcached for message and data store caching
- Arango DB as a data store
- Couch DB as a supporting data store
- MySql as a supporting data store
- TensorFlow for distributing machine learning algorithm processing
- SciKit-Learn for non TensorFlow applications embedded within specific components
- Celery for task queues
- Convolutional neural networks for language interpretation
- Recurrent neural networks for predicting responses and dynamically linking nodes within graphs as appropriate
- Autoencoders for unsupervised learning