Introduction
There are many reasons and motivations for me to initiate this new project whose name coded as Sapientia. The primary part is that I was frustrated by small and even-no progress of knowledge retrieval technology in near a decade. Yes, here I mean someone like Google and Baidu. Recently, ChatGPT and the its flashy integration with Bing give a shot in the arm to shake the basis of the sleeping search engine giant. OpenAI shows the capability of Large Language Mode (LLM) with Reinforcement Learning from Human Feedback (RLHF) in modeling us accumulated knowledge in a probabilistic way (specifically Neural Networks). About twelve years ago, Wolfram Alpha had shown the potential to make the same thing, rather in a symbolic and computing way different from ChatGPT. There still is a discussion for a final paradigm of either the probabilistic or the symbolic. Let the bullets fly for a while.
Meanwhile, a new paradigm of neural-symbolic routine is attracting more researches recent years. This is where our story starts. From a higher vintage point, maybe either probabilistic or symbolic paradigm is just a unilateral approximation or simulation of a single functionality of the natural brain mechanisms. They are not contradictory but supportive. Maybe there is still other undiscovered supportive mechanisms towards a real Artificial General Intelligence (AGI). Upon above reasons, the Project Sapientia involves the neural-symbolic fusion as one of its core designing principles.
Another important question we have to answer ahead is this project’s first big milestone as our success target. AGI is an attractive but fuzzy direction for an engineering software system. We would lost our direction at current stage if we target at AGI by now. We would absorb inspiring ideas and concrete achievements in related domains such as neural science, cognitive science, AGI theories and more. Project Sapientia’s primary target is to construct an extendable infrastructure of cognitive computing, which includes a new kind of language Kollider easy to use, a unified internal representation of knowledge and program, a distributed neural-symbolic knowledge storage, and a Gödel Machine inspired execution engine of the unified program representation. All of these components will be formulated in detail in following articles.
Inspiring Works
The Sapientia project was inspired by multiple awesome existing projects. One of them is Wolfram Language (WL) as well as its commercial cloud service of Wolfram Alpha. WL is a generic knowledge computing language constructed on a solid success of Mathematica for nearly thirty years in scientific computing over multiple disciplines. Mathematica originated from the blossom of meta/M-expression language paradigm and evolved into an advanced algebra calculus and symbolic reasoning engine. Its scale computing capability is enhanced by the GridMathematica technology. Symbolic language has the nature of representation flexibility in the sense that WL could represent all objects and knowledge in a unified model, which is the foundation towards a general knowledge computing engine. Due to the commercial limitation of WL, we know less about the inner implementation of the real computing process under the hood.
Another vital inspiring project is OpenCog and its inheritor OpenCog Hyperon as proposed by Dr. Ben Goertzel. OpenCog targets at a open framework of AGI whose fundamental ideas lay in Dr. Goertzel’s book “Engineering General Intelligence”. Project Sapientia and OpenCog Hyperon share a few designing disciplines such as hypergraph-modeled knowledge base, distributed knowledge storage. OpenCog has a high-level language called Atomese whose V1 is a bunch of scripts in different languages to handle Atom objects in Atomspace storage. The successive Atomese 2 proposed in Hyperon is designed with a core of generic gradual-typing based Atom and Atom-type-system framework. As still under design, Atomese 2 illustrates the potential of exploring dependent type, gradual typing, user pluggable type system. These aspects still remain under active researches. In the Kollider Language towards a native cognitive computing language, we would as well involve and practice some of these novel features.
There are many other outstanding projects and researches that inspire us. They are Cyc, Dr. Saba’s ONTOLOGIK, and many excellent researches on cognitive architectures such as Global Workspace Theory (GWT), Conscious Turing Machine (CTM), ACT-R and so on. The core design ideas of these inspiring works would be discussed in following related topics.
Next In Coming
I would write a series of articles on the core design of each Sapientia’s component. These topics remain open for public discussion. One of Project Sapientia’s goals is becoming a community-driven cognitive computing framework. The implementation part is hosted on Github community HadronMind. Welcome to join!
This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.
