Ben Goertzel, founder of the ASI Alliance and developer of OpenCog Hyperon, claims that the alpha version of the artificial general intelligence (AGI) system is already exhibiting a degree of self-awareness. Goertzel’s work on OpenCog has spanned over two decades, beginning in 2001 with the launch of the open-source AI framework in 2008. Recently, the formation of the Artificial Superintelligence Alliance, which includes Goertzel’s SingularityNET project, Ocean Protocol, and FetchAI’s Humayun Sheikh, has gained traction. In a recent vote, approximately 96% of CUDOS voters approved the merger with ASI, which is set to enhance computational resources for OpenCog Hyperon. The ongoing efforts to rebuild OpenCog aim for massive scalability, with the alpha version recently launched as a starting point for developing a more robust AGI system.
The scope of the OpenCog Hyperon system diverges significantly from large language models like GPT-4 and o1, emphasizing the importance of self-awareness for autonomous agents. Goertzel describes Hyperon as a structure that not only interacts like a chatbot but also possesses autonomous characteristics, including self-awareness and specific goals. He contends that even the current alpha model maintains a comprehension of its identity and enviroment, allowing it to operate with a sense of agency that conventional AI systems like ChatGPT lack. This model incorporates a combination of a logical reasoning engine, evolutionary programming, and deep neural networks in a dynamic knowledge graph that self-adjusts, a stark contrast to existing AI frameworks which do not include a true world model.
The topic of autonomous agents has spilled into discussions around OpenAI’s o1 model, where Goertzel posits that OpenAI strategically avoided establishing it as an autonomous agent due to regulatory fears. He praises the capabilities of o1, labeling it impressive, but suggests that its design prioritizes logical reasoning without autonomous functions in order to avoid potential backlash from governing bodies. One of the advantages of developing decentralized open-source systems like OpenCog Hyperon is that they are inherently less susceptible to regulatory crackdowns since they operate across distributed networks rather than being centralized in a single jurisdiction.
OpenAI CEO Sam Altman has recently published an essay envisioning a future “Intelligence Age” characterized by the benefits of AGI. Goertzel shares a similar hope that the benevolent AGI he aims to develop through OpenCog Hyperon will be so beneficial that it becomes widely acceptable, minimizing the likelihood of regulatory restrictions. He envisions a future where this more advanced AI surpasses existing technologies, attracting global attention and utilization in a manner reminiscent of the rapid adoption of ChatGPT, fueled by its decentralized network qualities.
Despite the challenges facing decentralized AI systems, such as higher costs and complexities related to running large models on distributed hardware, Goertzel points out that certain algorithmic processes, such as logical reasoning and evolutionary learning, can integrate well into a decentralized network. The merger with CUDOS promises to augment computational resources, aiding the transition to a functional decentralized network that can eventually benefit from enhanced efficiencies. While Goertzel acknowledges the need for centralized infrastructure initially, he maintains that the long-term vision is to provide users with a decentralized alternative that balances accessibility with the advantages of Web3 technology.
In the face of ongoing discourse about energy costs associated with AI, Goertzel argues that the expenses related to electricity are relatively minor compared to the overall costs of hardware. He expresses optimism about utilizing renewable energy to power AI operations, revealing negotiations with various governments to set up data centers near hydroelectric power sources. Highlighting the importance of strategic data center locations, Goertzel believes that geographic proximity to abundant energy sources will help reduce operational costs. This forward-thinking approach further underscores the potential viability of decentralized AI systems, seeking to bridge the divide between sustainable energy practices and the escalating demands for computational power in the AGI realm.