The rise of artificial general intelligence (AGI) — an artificial intelligence (AI) system with superhuman intelligence that can perform well at various tasks — is a matter of when, not if, according to a new analysis of thousands of expert opinions.
The updated analysis, conducted Feb. 18 by Cem Dilmeganiprincipal analyst at AIMultiple Research, has combed through approximately 8,600 predictions from scientists, AI experts and entrepreneurs between 2009 and 2023 to understand when experts believe it may happen.
A subsection of the analysis encompassed 10 surveys that queried a total of 5,288 AI researchers and experts. Based on an averaging of the data, there’s a 50% probability that we would achieve human-level intelligence in machines at some point between 2040 and 2061, the analysis found.
More recent surveys expect the technological singularity to arrive sooner. For instance, one of the most recent studiesconducted in 2023, questioned 2,778 scientists, and suggested AGI will be achieved by 2040 at the latest. Some in the field, like Dario AmodeiAI researcher and CEO of AI company Anthropic, believe it may even happen as soon as 2026.
Related: Scientists design new ‘AGI benchmark’ that indicates whether any future AI model could cause ‘catastrophic harm’
The rise of AGI has been fueled by the rapid advancement of transformer-based large language models (LLMs). This is the technology on which chatbots like ChatGPT and image generators like Dall-E are based. Before the advancement of these technologies, some scientists had predicted in 2019 that AGI would occur by 2060, or possibly never at all.
Why AGI is deemed a matter of when, not if
The analysis provided several reasons why scientists believe that AGI is certain to happen.
First, unlike human intelligence, there is no theoretical limit to increases in computing power. This is according to Moore’s Lawwhich predicted that power doubles roughly every 18 months. Future AI systems could one day reach parity with human intelligence in terms of calculations per second — but only if this prediction is followed. In recent years, many argue that Moore’s Law no longer tracks.
Quantum computing is also cited in this study as a means to overcome computing limitations. Quantum computers can process calculations in parallel by tapping into the laws of quantum mechanics. Classical computers — including the fastest supercomputers — must perform calculations in sequence. So quantum computing could fuel an advanced AI system with considerably more processing capacity than the best models today.
Other scientists in the field, however, believe further breakthroughs are needed before we can get anywhere close to AGI.
Facebook’s chief AI scientist, Yann lecunfor example, stated during a talk held in October 2024 that transformer-based architecture and current approaches to AI are incompatible with human-level intelligence. He has also urged scientists to move away from the notion of AGI entirely. He suggests there is a false equivocation between its widely used definition and what a single human being can achieve in reality — which, in practice, is a narrow subsection of specialised tasks, rather than having the capability of learning any and every task.
