(original) (raw)

OpenCog Hyperon: realizing the AGI dream

The open-source software framework bringing together multiple AI paradigms in a cognitive architecture oriented toward AGI at the human level and be yond

Join us on our quest to build the future of artificial general intelligence

(and to build it based on sound ethical principles and democratic decentralized governance)

AI has advanced tremendously in recent years, but the next AI revolution will be the biggest by far: the transition from narrow and training-data-bound AI systems to human-level AGI. This will also be the last AI revolution to be engineered centrally by humans — the next transition after this, to Artificial Superintelligence, will be largely led by AGIs rather than humans.

The deep neural architectures currently so popular and successful have some basic limitations, and are not in themselves suitable to power the next phase of the AI revolution.

OpenCog Hyperon provides a fundamentally more capable architecture, able to re-organize and improve itself to equal and ultimately surpass human-level general intelligence.

This is an initiative with deep roots in mathematical and cognitive theory, and also building on decades of practical engineering and prototyping and experimentation experience, with earlier versions of OpenCog and their predecessor systems.

The free and open source nature of OpenCog Hyperon means any developer with the time and expertise can contribute to the epochal goal of creating real AGI.

Please join us! We are doing this because we feel there is nothing more interesting going on here on Earth at the moment — and also nothing with clearer potential for broad benefit to humans and other sentient beings.

OpenCog Hyperon is...

OpenCog Hyperon is the successor of the original version of OpenCog (sometimes called OpenCog Classic), which also has an active developer community and a number of commercial applications.

Hyperon has an almost entirely new codebase, designed with radical scalablity in mind, and with a goal of leveraging new mathematical approaches to represent diverse cognitive algorithms in a unified way, so as to maximize the “cognitive synergy” between them

The underlying cognitive theory integrates ideas from numerous disciplines, including formal logic, knowledge representation, natural language processing, evolutionary learning, probabilistic reasoning, attractor neural nets, functional programming, process calculus, nonlinear dynamical systems and much more.

Key Components of OpenCog Hyperon

Atomspace Metagraph

The Atomspace Metagraph is weighted labeled graph-like structure known used by Hyperon to represent and store data.

Atomspace is a flexible and dynamic knowledge repository, which enables Hyperon to organize and access a variety of knowledge of various types (declarative , linguistic, mathematical, procedural, attentional, goal-related, sensory, motoric, emotional, intuitive, etc.)

Atomspace lives in RAM for use in real-time cognitive processing, and also lives on disk and in distributed form across multiple machines, thanks to the Distributed Atomspace implementation.

The MeTTa programming language

MeTTa (Meta Type Talk) is a novel programming language designed specifically to meet the needs of AGI. It is intended both for humans to script portions of AGI cognitive processes, and for the programming activity of AGI-related learning and reasoning algorithms themselves.

MeTTa programs combine elements of functional, logical and process-calculus-based programming. Each MeTTa program is represented as a subgraph of an Atomspace metagraph, and carries out its business centrally by querying and rewriting portions of Atomspaces.

MeTTa handles highly abstract constructs like run-time self-modifying code simply and naturally, but is also designed for efficiency both in running its own natively implemented AGI-related algorithms, and in interfacing with external processes invoking non-OpenCog AI algorithms such as deep neural nets or SAT solvers.

OpenCog Hyperon is....

SCALABLE

OpenCog Hyperon is designed to make effective use of distributed computing architectures to judiciously spread computational tasks across a large number of machines or clusters.

It is also designed to exploit either traditional cloud or on-premise architectures or decentralized, blockchain-based deployments (leveraging tools like SingularityNET, NuNet and HyperCycle).

MATHEMATICAL

The semantics of the MeTTa language, and the mode of implementation of different AI algorithms in MeTTA for use in Hyperon, is guided by a body of mathematical work connecting cognitive processes like learning, reasoning, creativity, concept blending, perception and motivation with a common set of mathematical structures related to metagraphs and associated algebras, categories and symmetries.

COGNITIVE

The OpenCog Hyperon toolkit can be used to implement a great variety of different software systems embodying a diversity of cognitive architectures; however, we have paid special attention to be sure that it is suitable for scalable and simple implementation of cognitive architectures resembling our current best understanding of the human mind.

Where can you start your journey with OpenCog Hyperon?

Starting your journey with OpenCog Hyperon as an AI developer is an exciting chance to explore the limits of Artificial General Intelligence (AGI). To get started, there are several key steps you can take:

Selected papers

Latest AGI Video Discussions

A few of the Innovators behind OpenCog Hyperon

Get Involved - Join the OpenCog Hyperon Community

Joining the OpenCog Hyperon community will give you the chance to meet other like-minded people, researchers, and developers.

Join us in insightful discussions, collaborate with subject-matter experts, and let’s build the future of AGI – together.

© 2024 Hyperon OpenCog All Rights Reserved.