The Evolution of Goertzel’s AGI Architectures

Ben Goertzel’s journey through the landscape of artificial intelligence (AI) has been one of vision, experimentation, and relentless innovation. As the founder of OpenCog in 2008, Goertzel sought to build a system that could achieve Artificial General Intelligence (AGI)—the kind of intelligent systems that possess the ability to understand and learn any task, much like a human. His ambition culminated in the creation of the MeTTa language and the Hyperon project, which represent his latest advancements in hybrid, scalable AGI systems. These efforts have evolved from the initial architecture of OpenCog, progressively advancing toward increasingly integrated and sophisticated AI systems capable of emulating human-like general intelligence.

This blog will explore the core concepts behind Goertzel’s early work with OpenCog, trace the significant advances leading to Hyperon, and highlight how these innovations exemplify his overarching vision for AGI. Goertzel’s unique approach contrasts with more traditional narrow AI methods, positioning him as a technical leader in the ongoing pursuit of AGI.

Core Concepts Behind OpenCog: Cognitive Synergy and Atomspace

Ben Goertzel’s journey toward achieving Artificial General Intelligence (AGI) began in earnest with the creation of OpenCog in 2008. This ambitious project aimed to develop a flexible, scalable cognitive architecture capable of mimicking the broad spectrum of human intelligence. OpenCog was designed as a foundational framework for AGI, bringing together a variety of AI methodologies and paradigms. At its core, the architecture was centered on two key concepts: cognitive synergy and the Atomspace. These ideas not only defined the structure of OpenCog but also laid the groundwork for Goertzel’s subsequent AGI architectures, which evolved into even more complex systems like Hyperon and MeTTa.

The concept of cognitive synergy is a foundational pillar of OpenCog and one of the most critical aspects of Goertzel’s vision for AGI. Cognitive synergy refers to the integration of different AI techniques and systems—each with its strengths and weaknesses—working together to create a more powerful, holistic intelligence. According to Goertzel, no single approach to AI could achieve true AGI on its own. Whether it was symbolic reasoning, neural networks, or other AI methodologies, each had its limitations when applied in isolation. To overcome this, Goertzel proposed that AGI systems should combine and integrate diverse AI methodologies, allowing them to complement each other’s strengths and compensate for their weaknesses.

This broader vision sought to create an intelligence that was flexible, capable of performing a wide variety of tasks, and able to adapt to new challenges. Just as the human brain uses multiple processes—both symbolic (abstract reasoning, language processing) and sub-symbolic (perception, emotion)—to function cohesively, Goertzel believed that an AGI system would need to leverage both symbolic and sub-symbolic AI components in synergy.

The Role of Atomspace: A Knowledge Representation System

The architectural cornerstone of OpenCog, and a key to its success, was Atomspace—a knowledge representation system that played a central role in facilitating the cognitive synergy between various AI components. Atomspace was designed to represent information in a highly dynamic and flexible manner, mimicking the complexity of human knowledge structures. At its core, Atomspace allows for the storage, connection, and processing of knowledge in a way that enables the system to reason, learn, and make decisions based on both high-level abstract concepts and low-level sensory data.

Atomspace operates by storing information in the form of “atoms,” which are the fundamental building blocks of knowledge. These atoms are connected together in a network-like structure, allowing the system to create relationships and connections between pieces of information. For example, a “cat” atom could be linked to atoms representing attributes such as “has fur,” “is an animal,” or “is a pet.” This network of atoms facilitates the system’s ability to reason about and manipulate knowledge in a manner that mirrors human cognitive processes.

What makes Atomspace particularly powerful is its ability to integrate symbolic and sub-symbolic information. Symbolic information refers to abstract representations like concepts, categories, and rules (e.g., “cats are animals”), while sub-symbolic information refers to raw data, perceptual information, or neural representations (e.g., sensory input from a camera or audio data). By blending both types of information, Atomspace enables OpenCog to handle both high-level reasoning and low-level data processing, which is essential for creating an AGI system that can interact with the world in a flexible, human-like manner.

This integration of symbolic and sub-symbolic processing is critical because it allows OpenCog to handle a broad range of tasks. For instance, symbolic reasoning helps the system understand abstract concepts, make logical deductions, or process language, while sub-symbolic learning helps it interpret sensory data and learn from experience. The synergy between these two processes gives OpenCog its cognitive versatility, allowing it to adapt to a variety of tasks and solve complex problems.

Cognitive Synergy: Integrating Diverse AI Techniques

The idea of cognitive synergy in OpenCog goes beyond just combining symbolic and sub-symbolic systems; it involves the integration of various AI techniques to form a unified cognitive system. OpenCog leverages techniques such as neural networks, evolutionary algorithms, and graph-based learning alongside traditional symbolic approaches. Each of these AI paradigms has its strengths and limitations:

• Neural Networks excel at learning patterns from large amounts of data, making them ideal for tasks such as visual recognition or speech processing.
• Symbolic Reasoning is strong at handling abstract reasoning, logical deduction, and structured data, such as rules or facts.
• Evolutionary Algorithms allow for adaptive learning and the optimization of solutions through simulated evolutionary processes.

By combining these different techniques, OpenCog was able to create an integrated system that could take advantage of the strengths of each methodology while compensating for their weaknesses. This multi-faceted approach is what allows OpenCog to handle both concrete tasks (like visual or sensory processing) and abstract reasoning (like planning, decision-making, and learning). The combination of these approaches creates a more robust and flexible AGI system, able to solve a broader range of real-world problems.

For example, OpenCog might use a neural network to process raw visual data and identify objects in an image, while symbolic reasoning could then be used to infer relationships between these objects, such as understanding that a “cat” is an animal, has fur, and is typically a pet. The ability to seamlessly integrate these processes is what sets OpenCog apart from more specialized, narrow AI systems that focus only on one aspect of cognition.

Moving Toward AGI: General Reasoning, Learning, and Problem Solving

In the early stages of Goertzel’s work with OpenCog, the primary goal was to create a system capable of general reasoning, learning, and problem-solving—the foundational elements necessary for AGI. While narrow AI systems excel at solving specific, well-defined problems (e.g., playing chess or detecting spam), AGI systems must be capable of dealing with a wide variety of tasks, learning from experience, and reasoning about complex, unfamiliar situations.

Goertzel’s integration of symbolic and sub-symbolic processes within OpenCog demonstrated that it was possible to approach AGI in a way that combined structured knowledge and learning from experience. General reasoning refers to the system’s ability to draw inferences, recognize patterns, and apply logic across different domains, which is critical for developing flexible and adaptable intelligence. Learning, meanwhile, allows the system to improve its performance over time, gaining knowledge from experience. Problem-solving involves the system’s ability to apply both reasoning and learning to overcome challenges, make decisions, and achieve goals.

OpenCog’s cognitive synergy and Atomspace architecture provided the ideal foundation for these capabilities. The system’s ability to represent and process both abstract concepts and raw sensory data enabled it to approach problems from multiple angles, making it capable of solving tasks that involve both reasoning and learning. These early advancements demonstrated the potential for combining symbolic reasoning with sub-symbolic learning, providing the building blocks for more advanced AGI systems in Goertzel’s later work.

Laying the Groundwork for AGI

Ben Goertzel’s work with OpenCog was an important milestone in the development of AGI. The concept of cognitive synergy, which integrates diverse AI methodologies to work together seamlessly, and the Atomspace, which serves as a flexible and dynamic memory store, were key innovations that allowed OpenCog to approach AGI in a holistic, scalable way. By combining symbolic reasoning with sub-symbolic learning, OpenCog laid the groundwork for more advanced AGI systems that could solve a wide range of tasks and operate autonomously.

Goertzel’s early work with OpenCog set the stage for the more refined AGI architectures that would follow, including his more recent developments in Hyperon and MeTTa. Through these systems, Goertzel has continued to push the boundaries of AGI, striving to create architectures that are both adaptable and scalable. His focus on cognitive synergy and Atomspace remains central to his vision of building an intelligence that can reason, learn, and problem-solve across a broad spectrum of tasks—ultimately bringing us closer to achieving true AGI.

Advances in Hyperon: MeTTa, Distributed Atomspace, and Neural-Symbolic Integration

The evolution of Goertzel’s AGI work reached a significant milestone with Hyperon, a new The evolution of Ben Goertzel’s work in AGI took a monumental leap forward with Hyperon, a new architecture that builds on the core ideas first explored in OpenCog but takes them significantly further. Hyperon represents the next phase in Goertzel’s long-term vision for Artificial General Intelligence (AGI)—a fully integrated, scalable architecture that is better suited for tackling real-world applications. Key to this evolution are MeTTa, a novel programming language, and the distributed atomspace, which enables greater scalability and flexibility than previous systems.

Hyperon and MeTTa introduce new paradigms for integrating multiple cognitive processes, such as reasoning, learning, and perception, into a cohesive AGI system. These advancements signal Goertzel’s continued focus on building hybrid AGI architectures that leverage the strengths of symbolic reasoning, neural networks, and distributed computing. By combining these elements, Hyperon creates an AGI framework that is more versatile, scalable, and aligned with Goertzel’s vision of intelligent systems that can learn, reason, and adapt to new situations.

MeTTa: The AGI Programming Language for Hybrid Systems

At the heart of Hyperon lies MeTTa, a programming language explicitly designed for developing AGI systems. Unlike traditional programming languages, which are primarily tailored for narrow AI applications or conventional software development, MeTTa was developed to handle the complexities inherent in building AGI architectures. MeTTa’s primary role is to facilitate the creation of distributed and hybrid AGI systems—systems that combine different cognitive processes into a unified, scalable architecture.

The design of MeTTa is deeply rooted in Goertzel’s goal of developing AGI systems that can operate seamlessly across various tasks and processes, drawing from multiple AI paradigms. MeTTa enables AGI systems to integrate symbolic reasoning, sub-symbolic learning, and even sensory data processing in a unified framework. Its flexibility is key to enabling the scalable and hybrid nature of Hyperon, allowing developers to build systems that can grow and adapt based on the demands of the environment or the tasks they are given.

By using MeTTa, developers can create AGI systems that are distributed across multiple systems or nodes, which allows for greater performance, parallel processing, and scalability. This makes Hyperon capable of handling large, complex tasks and large datasets more efficiently than earlier AGI systems, which were often confined to single nodes or limited resources.

Distributed Atomspace: Scalable Knowledge Representation for AGI

Another significant breakthrough in Hyperon is the distributed atomspace, an extension of the Atomspace system introduced in OpenCog. While Atomspace was initially conceived as a centralized knowledge store—essentially a single repository for all the information needed by the system—the distributed atomspace in MeTTa takes this concept to the next level by decentralizing the knowledge representation and allowing it to be processed across various systems or nodes.

The distributed nature of the atomspace enables scalability, which is crucial for building large, high-performance AGI systems. By enabling multiple nodes to contribute to the processing of data, the distributed atomspace allows Hyperon to handle more complex tasks and larger datasets, making it suitable for real-world, large-scale applications. For instance, Hyperon can be deployed to power systems that require real-time decision-making, such as in autonomous vehicles or large-scale data analytics, where distributed computing and parallel processing are essential.

Moreover, the decentralized processing of atomspace enables more sophisticated, real-time collaboration between different components of the AGI system. In traditional systems, the centralization of data often leads to bottlenecks or inefficiencies, particularly when multiple processes must access or update the same repository of knowledge. In contrast, a distributed system allows for more efficient access and modification of knowledge, leading to a smoother flow of information and enhanced overall performance.

This design gives Hyperon its flexibility and resilience, enabling it to scale according to the needs of the task or system. By decentralizing knowledge storage and processing, Hyperon can be adapted to handle a wide range of tasks, from handling real-time data streams to solving complex optimization problems.

Neural-Symbolic Integration: Bridging the Gap Between Neural Networks and Symbolic AI

One of the most groundbreaking features of Hyperon is its integration of neural-symbolic systems. Traditionally, AI systems have been divided into two broad categories: neural networks (connectionist models) and symbolic AI. Neural networks are excellent at handling tasks like image recognition, language processing, and pattern identification—areas that require learning from large amounts of unstructured data. On the other hand, symbolic AI excels at tasks that require reasoning, such as logic, rule-based systems, and structured knowledge representation.

However, these two approaches have historically been treated as separate paradigms, with little overlap or integration between them. Hyperon bridges this gap by incorporating neural-symbolic integration, allowing both high-level reasoning (symbolic AI) and low-level learning (neural networks) to occur simultaneously within the same system. This integration provides a more holistic approach to AGI, enabling systems to perform both abstract reasoning and complex pattern recognition in a more unified and efficient manner.

The ability to combine symbolic and sub-symbolic reasoning is key to enabling more nuanced and complex decision-making. In traditional narrow AI systems, these two modes of processing are kept separate, limiting the system’s ability to handle complex, multi-faceted tasks. However, Hyperon’s neural-symbolic integration allows for a richer, more dynamic decision-making process that draws on the strengths of both approaches.

For example, in a scenario where an AGI system must navigate a complex environment, symbolic reasoning might be used to plan and strategize, while neural networks process real-time sensory data to guide actions. By combining these two systems, Hyperon can make decisions based on both abstract reasoning and immediate, situational context—much more like how humans reason and act.

This integration enables greater flexibility in the way Hyperon approaches problem-solving and decision-making. It also allows for the development of AGI systems that are more adaptable, capable of learning from experience while simultaneously reasoning about the tasks at hand. The combination of symbolic and sub-symbolic AI systems enhances Hyperon’s ability to tackle a wide variety of problems across diverse domains.

Hyperon’s Vision: Scalable, Hybrid AGI Systems

The advances in Hyperon and MeTTa exemplify Goertzel’s overarching vision for AGI systems: to create scalable, hybrid AGI systems that combine the strengths of various AI paradigms. Rather than relying on one AI methodology—whether it’s symbolic reasoning, neural learning, or distributed systems—Goertzel has created an architecture that integrates these approaches into a cohesive whole. This hybrid architecture allows Hyperon to operate more efficiently and flexibly than traditional AI systems, making it well-suited for a wide range of real-world applications.

The significance of this hybrid approach cannot be overstated. While traditional narrow AI systems are highly specialized and excel at specific tasks, they are often limited when it comes to general reasoning, flexibility, and adaptability. By combining multiple AI paradigms, Hyperon is able to tackle more complex, multi-dimensional problems that require both high-level reasoning and low-level learning. It represents a significant leap forward in the search for true Artificial General Intelligence, providing a system that can learn, reason, and adapt across a wide range of domains.

Moreover, Hyperon’s scalability ensures that it can handle the computational demands of real-world applications, from autonomous vehicles to large-scale data analysis, making it a versatile and powerful AGI architecture. As Goertzel’s work continues to evolve, Hyperon stands as a powerful demonstration of how hybrid AI systems can deliver real-world results and represent a viable alternative to more traditional narrow AI approaches.

A New Path Toward Hybrid AGI

Goertzel’s work with Hyperon and its key innovations—MeTTa, the distributed atomspace, and neural-symbolic integration—showcases a groundbreaking approach to AGI. By combining symbolic AI, neural networks, and distributed systems, Goertzel has crafted a more flexible, scalable, and integrated architecture for creating hybrid AGI systems. These advances not only exemplify Goertzel’s long-term vision for AGI but also position his work as a powerful alternative to mainstream AI methods that typically focus on narrow, specialized applications.

Hyperon’s ability to merge reasoning and learning, scale across distributed systems, and handle real-time data sets it apart from more traditional AI approaches, opening up new possibilities for AGI. Goertzel’s technical leadership in this area highlights the growing potential of hybrid AI systems and signals a new era for the development of intelligent systems that can more closely mimic human cognition.

How This Architecture Exemplifies Goertzel’s Vision for Integrated AGI

Ben Goertzel’s journey in the realm of Artificial General Intelligence (AGI) spans more than a decade and is marked by a relentless pursuit of creating intelligent systems that integrate diverse cognitive processes. His work with OpenCog, Hyperon, and MeTTa exemplifies a distinctive approach to AGI that integrates multiple AI paradigms into a cohesive system. This system is designed to mimic human cognitive functions, facilitating the development of machines that can learn, reason, and adapt in ways that are akin to human intelligence.

At the core of Goertzel’s vision is the idea that AGI should integrate different types of cognitive processes to create a system that can handle a variety of tasks, from abstract reasoning to sensory perception. His systems go beyond the limitations of narrow AI, which excels at isolated tasks but lacks the flexibility to adapt to new, unforeseen challenges. Goertzel’s AGI systems, by contrast, are designed to be adaptable and capable of solving an ever-growing range of problems over time. Through the integration of symbolic reasoning, sub-symbolic processing, and distributed knowledge storage, Goertzel’s architecture provides a path toward truly general intelligence.

Cognitive Synergy: Integrating Diverse AI Paradigms for Holistic Intelligence

One of the most significant innovations in Goertzel’s AGI approach is the concept of cognitive synergy—the idea that no single AI methodology can achieve true AGI on its own. Goertzel emphasizes that different AI paradigms, such as symbolic reasoning (logic, knowledge representation, rule-based systems) and sub-symbolic processing (learning from experience, neural networks, sensory data processing), must work together to form a cohesive whole. The cognitive synergy in Goertzel’s systems enables them to leverage the strengths of each paradigm, compensating for their individual weaknesses and creating a more robust and adaptable system.

For instance, symbolic reasoning excels at processing high-level concepts, rules, and abstract thought. It allows systems to perform logical deductions, make inferences, and handle structured data. However, symbolic reasoning alone cannot process the raw sensory data (images, sounds, etc.) that humans use to interact with the world. This is where sub-symbolic processing—which is the domain of neural networks and other learning algorithms—comes into play. Neural networks excel at learning patterns and representations from data but often struggle with handling abstract reasoning or symbolic logic.

In Goertzel’s AGI architectures, the combination of symbolic and sub-symbolic processes allows for a more holistic approach to problem-solving. For example, while a neural network might identify an object in an image (sub-symbolic processing), symbolic reasoning can be applied to understand that object’s role or relationships within a broader context (symbolic reasoning). By integrating these two types of processing, Goertzel’s systems are better equipped to navigate complex problems that require both high-level abstract thinking and perceptual learning.

This synergy between symbolic and sub-symbolic processes is foundational to creating flexible and adaptable AGI. It allows the system to learn from data, adapt to new information, and reason about complex concepts—skills that are central to human-like intelligence.

Distributed Atomspace: Scalable and Flexible Knowledge Representation

Another key innovation in Goertzel’s work, particularly in Hyperon, is the distributed atomspace, which extends the original Atomspace concept introduced in OpenCog. In its original form, Atomspace acted as a centralized knowledge store, where information was represented as “atoms” connected to form a network of related concepts. This atomspace was crucial for facilitating the cognitive synergy between different cognitive processes, as it provided a unified framework for both high-level reasoning and low-level perceptual data processing.

However, as AGI systems grow in complexity and scale, the limitations of centralized processing become apparent. A single, centralized atomspace could become a bottleneck for large-scale AGI systems, particularly those that require real-time processing or need to handle vast amounts of data. In response, Goertzel’s distributed atomspace in Hyperon was designed to enable decentralized knowledge storage and processing. Instead of relying on a single node or central repository, the distributed atomspace spreads the knowledge and data processing across multiple systems or nodes.

This distributed architecture offers several advantages:

• Scalability: By decentralizing the atomspace, Hyperon can handle much larger datasets and more complex tasks without being constrained by the limitations of centralized memory or computational resources.
• Flexibility: The distributed atomspace allows for more flexible deployment across different systems, enabling AGI to be more adaptable in a variety of environments.
• Real-time Collaboration: Multiple components of the AGI system can work together in real time, improving coordination and performance across different tasks.

This distributed approach makes Hyperon a much more scalable and efficient system than earlier architectures, enabling it to perform tasks at the scale required for real-world applications, such as autonomous vehicles, large-scale data analytics, or complex decision-making in dynamic environments.

Neural-Symbolic Integration: Combining Learning and Reasoning

One of the standout features of Hyperon is its integration of neural-symbolic systems, which combines the strengths of neural networks and symbolic AI into a unified system. Traditionally, these two branches of AI were viewed as distinct and often incompatible. Neural networks excel at learning patterns from large datasets and are excellent for tasks like image recognition, language processing, and classification. However, they often struggle with reasoning, logic, and handling abstract, high-level concepts.

On the other hand, symbolic AI has been designed to handle high-level reasoning, logical deduction, and structured knowledge. It excels at tasks that require rules, inferences, and abstract thinking but lacks the ability to process unstructured data like images or sound. The key challenge has been to integrate these two approaches in a way that allows them to complement each other, with neural networks handling perception and pattern recognition, and symbolic AI managing reasoning and decision-making.

In Hyperon, Goertzel’s approach to neural-symbolic integration bridges this gap, enabling the system to handle both high-level reasoning and learning from experience in a seamless, hybrid framework. This integration allows for more nuanced decision-making and adaptable intelligence that can handle complex tasks requiring both abstract reasoning and real-world sensory data. For example, a system might use neural networks to recognize an object in an image and then use symbolic reasoning to infer the object’s function or relationships to other entities in the environment. This hybrid approach is essential for enabling true AGI, as it mimics how humans combine sensory perception with logical thought to understand and interact with the world.

A Holistic, Integrated Path to AGI

Goertzel’s AGI architectures, particularly Hyperon, exemplify his broader vision of integrated, flexible, scalable, and adaptive AGI systems. Unlike traditional narrow AI, which excels at solving specific, isolated tasks, Goertzel’s systems are designed to learn, reason, and solve a broad range of problems over time. They are capable of adapting to new situations, integrating diverse cognitive processes, and scaling to meet the needs of real-world applications.

By combining symbolic reasoning with neural learning, and by employing distributed knowledge storage and processing, Goertzel’s systems have the potential to solve more complex problems that involve both high-level abstract thinking and perceptual learning. His integrated architecture provides a concrete alternative to the narrow, specialized AI systems that dominate the current AI landscape. While narrow AI can perform highly specific tasks with impressive accuracy, Goertzel’s AGI systems are designed to think, learn, and adapt in a more human-like manner, solving a wider array of challenges across diverse domains.

The Path to True General Intelligence

Ben Goertzel’s AGI architectures, particularly through OpenCog, Hyperon, and MeTTa, exemplify his vision of creating systems that combine symbolic reasoning, sub-symbolic learning, and distributed knowledge storage into a unified, adaptable framework. His emphasis on cognitive synergy, neural-symbolic integration, and scalability reflects a deep understanding of what is required to build true AGI—intelligent systems capable of performing a broad range of tasks, learning from experience, and adapting to new challenges.

By taking an integrated approach, Goertzel has positioned himself as a leader in the AGI space, offering a path toward true general intelligence that goes beyond the narrow AI paradigms dominating the market today. His work is an important step toward developing intelligent systems that can reason, learn, and interact in a human-like manner, providing a much-needed alternative to the limitations of today’s specialized AI solutions.

A Concrete Alternative to Narrow-AI Methods

Ben Goertzel’s journey from OpenCog to Hyperon and MeTTa reflects a relentless commitment to advancing AGI in ways that are both innovative and pragmatic. By focusing on hybrid, scalable architectures that integrate multiple AI paradigms—symbolic reasoning, neural learning, and distributed systems—Goertzel has positioned himself as a leader in the field, offering a concrete alternative to the current trend of narrow AI solutions.

His work is a reminder that achieving AGI is not about following the hype or chasing after the biggest models, but rather about developing intelligent systems that can think, reason, and learn in a way that is deeply integrated with human knowledge and experience. Goertzel’s technical leadership has paved the way for the next generation of AGI architectures—systems that prioritize flexibility, efficiency, and adaptability in the pursuit of true general intelligence.

Works Cited

Goertzel, B. (2014). The Goertzel-Lopresti model of conceptual representation and analogy-making in OpenCog. Procedia Computer Science, 41, 153-158.

Goertzel, B. (2014). OpenCog: A Software Framework for Integrative Artificial General Intelligence. In B. Goertzel & P. Wang (Eds.), Artificial General Intelligence (Vol. 8598, pp. 272-277). Springer International Publishing.

Goertzel, B. (n.d.). Ben Goertzel: Keynote at CrewAI Summit 2024. CrewAI. Retrieved from https://www.crewai.com/summit

OpenCog. (n.d.). OpenCog Wiki. Retrieved from https://opencog.org/wiki/Main_Page

OpenCog Hyperon. (n.d.). Home. Retrieved from https://hyperon.ai/

SingularityNET. (n.d.). Ben Goertzel. Retrieved from https://singularitynet.io/ben-goertzel/

Techopedia. (n.d.). What is Artificial General Intelligence (AGI)? Retrieved from https://www.techopedia.com/definition/artificial-general-intelligence-agi

Klover.ai. “Multi-Front Leadership: Goertzel’s Roles Across AGI & Transhumanist Organizations.” Klover.ai, https://www.klover.ai/multi-front-leadership-goertzels-roles-across-agi-transhumanist-organizations/

Klover.ai. “Benjamin Goertzel.” Klover.ai, https://www.klover.ai/benjamin-goertzel/

Klover.ai. “Decentralized Intelligence: How Ben Goertzel Democratizes AGI.” Klover.ai, https://www.klover.ai/decentralized-intelligence-how-ben-goertzel-democratizes-agi/