About to create super-human artificial intelligence?

Ben Goertzel believes we are on track to create an artificial intelligence as smart as humans by 2027

Ben Goertzel, a computer scientist and CEO of SingularityNET, made the claim during the closing keynote address at the AGI 2024 Summit in Panama City, Panama. He is known as the “father of AGI” after helping to popularize the term artificial general intelligence (AGI) in the early 2000s.

The best AI systems deployed today are considered “narrow AI” because they can outperform humans in one domain, based on training data, but cannot outperform humans overall. These narrow AI systems, which range from machine learning algorithms to large language models (LLMs) like ChatGPT, struggle to reason and understand context like humans do.

However, Goertzel notes, AI research is entering a period of exponential growth, and evidence suggests that artificial general intelligence (AGI) — where AI is capable of matching humans in some domains independent of the initial training data — is within reach.

Goertzel suggests that 2029 or 2030 may be the most likely years for humanity to build the first AGI agent, but that it could happen as early as 2027.

If such an agent were designed to have access to and rewrite its own code, it could quickly evolve into an artificial superintelligence (ASI) — which Goertzel defines as an AI with the cognitive and computational capabilities of the entire human civilization.

“No one has created human-level artificial general intelligence yet. It is reasonable that we could achieve human-level AGI within the next three to eight years,” says Goertzel.

He points to three pieces of evidence to support his thesis, the first of which is computer scientist Ray Kurzweil's modeling in his book “The Singularity is Near” (Viking USA, 2005), which has been distilled in his forthcoming book “The Singularity is Nearer” (Bodley Head, June 2024).

In his book, Kurzweil builds predictive models that suggest AGI will be achievable by 2029, focusing primarily on the exponential nature of technological growth in other fields.

Goertzel also points to the improvements made to LLM over the past few years, which have awakened many people around the world to the potential of AI.

The third proof of concept lies in the work to build the “OpenCog Hyperon” infrastructure, as well as the related software systems and the upcoming AGI programming language, dubbed “MeTTa,” to support it, Goertzel said.

OpenCog Hyperon is a form of AI infrastructure that involves combining existing and new AI models, including LLM as a component.

However, Goertzel admits that he could be wrong and might need a quantum computer with a million qubits or something.

“Once human-level AGI is reached, full superhuman AGI could be achieved within a few years — unless AGI threatens to stunt its own development out of conservatism,” Goertzel added.