Present-day Artificial Intelligence (AI) solutions focus on well-defined domain specific problems, clearly defined objectives, and are unable to efficiently generalize across domains. The AI solution that’s applied to self-driving cars cannot be applied to stocking shelves. Human beings can perform both tasks and many more as they maintain general intelligence.
Artificial general intelligence (AGI) is the intelligence of a machine that can successfully perform and generalize across tasks in a manner similar to Human beings. AGI allows for a single solution to be applied to multiple problems across multiple domains. With respect to learning and applied learning, critical understanding gathered from one domain and experience can be ‘internalized’ and applied to another. To achieve this capability, one must address consciousness: An awareness which allows one to perceive, think, convey thoughts, embody intention, and construct generalized understanding across domains. Consciousness is essential to Human general intelligence and must be addressed in an equivalent Artificial General Intelligence solution.
Monad.ai aims to establish such a solution with stated capabilities via a foundational AGI software framework that embodies awareness. Monad.ai’s framework requires less training iterations than present AI solutions as it can reason independently and generally. It requires less training data as it can intuitively construct and reason through variances in input data and local memory. As the platform is ‘aware’, users and administrators are able to interact with the AGI framework via natural language and higher level languages thus broadening the range of individuals who can program, tune, and task it. Problems such as AI’s language problem will be truly resolved.
At the core of the architecture is a computational model of consciousness that maintains generalized perception and cognition allowing for generalized learning, communication, and goal driven interaction. Monad.ai’s architectural approach utilizes 3 years of fundamental research centered on information theory, physics, biological systems, neuroscience, and cognitive science. It brings a paradigm central to biological life to the computational domain. Development of the AGI software framework will be done in phases of capability. The first phase will proof a basic level of end-to-end system functionality. Subsequent versions will enhance overall capability, features, and awareness level.
AGI
Monad.ai’s AGI software framework is composed of a core, inner region, and a cortical region. Various layers of supporting systems and subsystems structure around the core in the inner region and outer cortical region. At the lower levels, the framework functions on a deeply revised and biologically inspired neuron model and network. Sensory information flows into the framework via the host interface whereupon it is processed further and perceived. Motor commands are sent from the core, outward to the host interface, and arrive at the host where they are manifested. For compatibility, AI solutions external to Monad.ai’s framework can wire into it via a dedicated interface segment in the cortical region.
Development Framework
The development framework consists of three-components: Environment (Env), Host, and AGI. The Environment component reflects an interactive space-time dimensioned domain based on universal objective laws. The host component is present in the environment as a physical construct that manifests the AGI’s capabilities. The AGI component resides in the host and serves as the brain. Sensory receptors present in the host receive information from the environment and transmits it to the AGI for perception and further processing. The AGI performs generalized perception, cognition, learning, and further interacts with the environment via command transmissions to the host. The host then manifests the commands in the environment via its respective output facilities. Human beings interact with the outside world in a similar manner.
Context
Development and operation can occur in both real-world and virtual contexts. An example of an interaction loop in the real-world context consists of the AGI component interfaced to a robotic platform equipped with sensory and motor facilities. Throughout the proofing stages, development and operation will occur in a virtual context. A virtual context is achieved using simulation software which generates a virtual space-time domain paired with a physics engine which applies physical laws. In a virtual context, the host component is resolved as a virtual object with sensory and motor capabilities. The AGI interfaces with the virtual host and environment though the host interface as it would in the real-world context. Given this generalized host-interface, it is possible to wire the AGI into and out of different contexts and domains without conducting any new software development.
Interaction
The Environment, Host, and AGI components connect to each other through tightly controlled interfaces consisting respectively of physical interaction and exchanges of data and control packets. By default, the AGI functions in open-explore fully-autonomous mode with a generic goal set to explore, experience, and learn. Beyond general interactions, it is capable of interacting with users, bystanders, and other other AGI in this mode. It can communicate its internal thought process when requested, convey what it has more specifically set its goals to, and perform a whole host of other functions in an interactive manner.
The AGI also functions in semi-autonomous mode whereby its primary goal(s) are set by an Administrator (Admin). If it is unclear about a given goal or, if there is a conflict, it can inquire and converse to further establish an understanding before it proceeds. In the real-world context, the Admin is able to gather run-time information and send commands to the Host and AGI components. In a virtual context, the Admin can interface to and command all three components (Environment, Host, and AGI). The capability to command all three components is essential during development, test, and proofing. Administration is conducted through a control application that operates in normal or developer mode. The administrative interface to the AGI supports both proprietary data protocols and natural language.
