Within a short period of just three years, humanity can be expected to reach a revolutionary milestone in the development of artificial intelligence (AI), creating agents with cognitive capabilities comparable to those of humans.
This prediction was shared by Brazilian scientist Ben Goertzel, computing pioneer and leader of SingularityNET, in his closing talk at the AGI 2024 Beneficial Summit, which took place on March 1, in Panama City. Goertzel, often cited as the “father of general AI,” has been a central figure in promoting this concept since the early 2000s.
In contrast to current AI systems, which are classified as “narrow AI” and have superior abilities to humans in specific domains thanks to their specialized training, they still cannot match the versatility and contextual understanding of humans.
These systems, which range from machine learning algorithms to complex linguistic models like ChatGPT, still face significant barriers to human reasoning and perception.
Ben Goertzel, an authority on artificial intelligence, highlighted that we are on the brink of an era of accelerated advances in the field. According to him, the possibility of achieving artificial general intelligence (AGI) — a form of AI with multifaceted abilities, comparable to those of humans and not limited by initial training data — is closer than ever. This theoretical milestone, known as “singularity”, could represent an unprecedented transformation in the field of AI.
According to the artificial intelligence expert, the year 2029 or 2030 marks the most promising period for the creation of the first artificial general intelligence agent (AGI), although this achievement could emerge as soon as 2027.
An IAG agent with the ability to modify its own code could, theoretically, quickly ascend to the artificial superintelligence (AIS) stage. This advanced form of AI, as described by Goertzel, would have cognitive and processing power equivalent to the sum of all human civilization.
“No one has yet created human-level artificial general intelligence; no one has a solid understanding of when we will get there. I mean, there are known unknowns and probably unknown unknowns. On the other hand, it seems quite plausible to me that we could achieve human-level AGI within, say, the next three to eight years,” said the Brazilian scientist.
As we approach the tipping point known as the singularity, Goertzel highlighted “three strands of converging evidence” that underlie his view. The first aspect is based on the projections of Ray Kurzweil, an illustrious computer scientist, whose predictions are detailed in the book “The Singularity is Near” and will be expanded in the next work “The Singularity is Closer”. Kurzweil anticipates that artificial general intelligence (AGI) could be a reality by 2029, emphasizing the exponential acceleration of technological progress.
Furthermore, Goertzel highlighted the significant advancement of large language models (LLM) recently, which have revealed the true potential of AI to the world. He noted that although LLMs are not sufficient to achieve IAG, due to their lack of authentic understanding, they are fundamental pieces in a more complex and interconnected architecture that could lead to the achievement of IAG.
The third and final line of evidence that Goertzel presents in support of approaching the singularity lies in his own contribution to the construction of an advanced infrastructure, called OpenCog Hyperon. Additionally, he is developing related software systems and a new programming language for IAG, called MeTTa, that will support this infrastructure.
OpenCog Hyperon is an AI system that seeks to integrate existing and new AI paradigms, including LLM as one of its components. The ultimate goal is to create a large-scale distributed network of AI systems, operating on diverse architectures, that can simulate different aspects of human cognition, from content generation to complex reasoning.
The methodology in question has the endorsement of renowned experts in the field of artificial intelligence. Among them is Matei Zaharia, CTO of Databricks, who recently co-authored an article on February 18, published on the Berkeley Artificial Intelligence Research (BAIR) portal, reinforcing this perspective.
However, Goertzel admitted that he “could be wrong” and that we might need “a quantum computer with a million qubits or something.”
“In my opinion, once artificial general intelligence reaches the human level, it is plausible that within a few years we will witness a breakthrough to a radically superior AGI, unless AGI itself restrains its development due to a conservative bias. I firmly believe that when an IAG is capable of insight into its own mind, it will have the ability to perform engineering and scientific tasks at human and superhuman levels. This ability would make it possible to create successive versions of IAG that are increasingly intelligent, thus triggering an explosion of intelligence. This scenario could result in an exponential increase that would exceed even Ray Kurzweil’s expectations,” he concluded.
