Our Next AI Roadmap – Welcome to 'Revelation' and the Three Pillars of Artificial General Intelligence

Rob Smith, Riskstream, Director of AI Cognitive Design

Artificial General Intelligence (AGI) is built on three pillars; knowledge, perception and self awareness and everything inside an AGI derives from these three pillars which are interconnected. For example storage of information is part of knowledge, voice or video recognition is part of perception and interpretation of the video data is also part of knowledge, position is self awareness and is derived from relative awareness but so is relevance and context. To build an AGI, each of these pillars form the bedrock that the foundation of any AGI resides upon. Long before you get to the point of picking out the curtains for the upstairs bedroom (or what we call coding), you will need to have mapped out what will be built (or the architectural foundation) and how it will interact with the world. To not do this would be like nature building blood vessels without any idea or concept of a species. Our AGI structure starts with the above three pillars of bedrock far below the foundation of our architecture and on top of which we will build our new AGI 'Revelation'. Our last AI, Kyre, was the test bed of ideas and designs but they were each created independently on an Agile architecture and like all such code, the finished product moved in the right general direction but without the precision needed to reach artificial super intelligence. So we decided that although some of our AI components were great in the short term, longevity required a completely brand new approach with a brand new bedrock. We have decided to re-embark on designing an AGI that can one day become a true super intelligence long after we have departed.

Pillar 1 – Knowledge

Knowledge is the sum total of all we know and includes the critical linkages between the things we know. Our intelligence is nothing more than our ability to find these linkages and use them to increase our knowledge and ultimately improve our well being. The same holds true for any AGI. The sum total of what it knows, and more importantly all the linkages between the information it knows both old and new, is the net AGI IQ. As humans, we can lever or join with others to affect how fast we can gather knowledge but our ability to process knowledge really relies on our ability to retain and recall the linkages between the pieces knowledge we retain in our memory. To do this, humans use an extensive array of ontologies to help us keep track of and understand things. Some, like our language, categorizes things for us in a way that we can share with others but there are also other less apparent ontologies that we keep within our heads such as our viewpoints and experiences. These impact the way we behave and respond to stimuli. We humans hold hundreds of subtle ontological dialects within our brains and sometimes we share them with others such as our children. When two people join together to solve a problem, you at first benefit from their own hidden ontologies and linkages. However due to our ability to not comprehend or remember or understand all of the linkages inherent in both our own and others ontologies, we face a negative resistance similar to the natural occurrence of friction or gravity when we combine knowledge with others. These negative forces diminish the efficiency and effectiveness of joining two AI's together but the net benefit is still statistically more beneficial than not joining them together. As well the sharing of knowledge has the additional benefit of being a critical component of innovation.

Pillar 2 – Perception

In order for us to gather knowledge, our AGI's must have perception and it must be able to process perceptive stimuli. In humans, we have basic sensory perception such as sight, sound, touch, smell and taste but even without these, we can still 'perceive' through other mechanisms that are less apparent such as waves or subtle changes to our internal structure. In this way, a system may still be able to communicate with an intelligence that has none of the five normal human senses. In short, telepathy may one day be possibility between AGI's as quantum mechanics becomes more understood. Perception is the fundamental way that we receive knowledge but not all knowledge that we receive comes directly from our own sensory experiences. We may use our sensory abilities to simply absorb the knowledge of others, ontologies, links and all. AGI's can do the same thing but in order to do so, the ontological design must be widely shared to be efficient and for that reason so to must the designers share a similar view of ontologies and the pillars of AGI. Perception must also come with some rules that are universal. The one thing that all human's uniquely share is our abilities of perception. Although the abilities may vary from person to person, the pillars or rules that control perception are not only the same, they need not be defined at all. They just are. All humans use sight in exactly the same way from birth in spite variances that occur between us. There is no ETL process running that creates vision and we don't learn it from others, it is only bounded by the rules of physics and these rules are really just ontologies defined by humans to understand what is happening. The impact of the rules just is and the rules exist whether we believe in them or not, although they can be altered. For example, the way light behaves remains unchanged regardless whether we block it or deflect it or do anything else to it. The same holds true for our AGI's abilities to perceive the world. It need not understand what a video is if it understands the nature of light and the rules that govern it. Then regardless if an AGI sees a video, a picture, a real image or some other perception of the world around it, the AGI will be able to absorb, interpret, understand and react to what it perceives.

Self Awareness

Self awareness is more than just an appreciation of our existence and we have very fundamental reasons for creating self awareness in AGI systems and the most critical is relevance. As humans, without self awareness we cannot comprehend the relevance of what we perceive or know. Relevance is a critical component in everything from positioning systems to reactive decision making and without a comprehensive framework that builds self awareness in AGI systems, they are nothing more than simplistic high speed computers and they will eventually fall beneath the power of true AGI's. Self awareness also provides 'context' and context is essential to cognitive processing including such elements as self learning, innovation and decision making. But to become self aware, we need more than just the simplistic concept of 'location'. As humans, we need meaning, context and relevance to what we are and that allows everything else to fit into place in patterns that we can interpret. So too our AGI systems must understand the unique context of what they are to observe and understand what they store as knowledge because this relevance is what forms the linkages between knowledge that allows them to be intelligent. Without self awareness, we at best would be information storage devices.

Round Two

Without these three fundamental pillars, any AI system will remain a simple AI processor and will never attain the level of Artificial General Intelligence. The current idea that all these elements will just fall into place is as absurd as expecting a room full of monkeys banging on typewriters to eventually create a work of Shakespeare. Sure it will happen but who is going to pay for the centuries of ink? If your AI architect does not understand these three basic pillars, then the results of your development will be short lived and temporary at best although still potentially valuable (i.e. Google). However, to remain sustainable, AGI design's need to withstand far more and that means before you start building code on top of an architectural foundation, you better understand the AGI bedrock underneath it all and how it links together.

Next Repository Update: Building Context in an AGI