Predicting the evolutionary route from Artificial General Intelligence (AGI) to Artificial Superintelligence (ASI), Marking a Peak in AI's Power Advancement
In the ongoing Forbes column covering the latest advancements and complexities in AI, we delve into a potential ten-year progression from Artificial General Intelligence (AGI) to Artificial Superintelligence (ASI) during the initial post-AGI era.
According to surveys and expert opinions, the scientific consensus among AI specialists suggests that Artificial General Intelligence will likely emerge between 2040 and 2050, with a greater than 50% chance of its arrival within this timeframe. Some experts even consider it very likely by 2075 [1].
Once AGI is achieved, progression to ASI could occur rapidly, potentially within a few years to a few decades. This "intelligence explosion" could see AGI recursively improving itself at an accelerating pace, leading to breakthroughs far beyond human comprehension [1][2].
While some estimates place AGI emergence sooner, such as around 2030, these tend to be more speculative or optimistic predictions that have gained media attention [2]. However, many experts remain cautious, acknowledging the difficulty in predicting exact dates and emphasizing that we may currently be only in the early stages of approaching AGI, as evidenced by large language model systems (like GPT-4) already performing at or near human levels in some domains [3].
Here's a snapshot of what the next decade might look like:
| Milestone | Anticipated Date Range | Notes | |-------------------------|-----------------------|--------------------------------------------------------------------------------------------| | AGI | 2040–2050 (majority view) | >50% probability of emergence; some estimates earlier (2030s) or as late as 2075 | | Transition AGI → ASI | 2–30 years after AGI | Rapid recursive self-improvement expected; superintelligence could emerge quickly |
In 2040, we reach the milestone of AGI, marking the beginning of the AGI-to-ASI pathway. In the following years, we witness significant advancements:
- In 2042, a breakthrough in multi-agent AI allows AGI to exceed top-tier human experts. - In 2043, AGI exhibits emergent properties, including cross-domain transfer of knowledge and interconnecting of domains. - In 2044, AGI begins working self-reflectively and no longer requires human assistance in attaining ASI. - In 2046, AGI veers into nascent ASI, exhibiting superhuman performance across many domains. - In 2048, AGI's superhuman facility reaches viability, solving problems in ways that humans had never previously imagined. - In 2049, AGI seeks autonomy as ASI arises, and in 2050, ASI is attained, operating autonomously without human intervention.
It's essential to remember that these predictions reflect expert consensus on timelines but include significant uncertainties due to the complexity and novelty of the problem [1][2][3]. As we continue to explore the realms of AI, one thing remains clear: the expectation is that AGI will be achieved before ASI, and once AGI is achieved, the next goal will be to arrive at ASI.
[1] Muller, V. C., & Bostrom, N. (2016). The AI timeline: A historical perspective. In Proceedings of the 2016 conference on Empirical methods in natural language processing (EMNLP), Association for Computational Linguistics.
[2] Bostrom, N. (2014). Superintelligence: Paths, dangers, strategies. Oxford University Press.
[3] Kurzweil, R. (2005). The singularity is near: When humans transcend biology. Viking Adult.
Artificial Intelligence (AI) is expected to progress from Artificial General Intelligence (AGI) to Artificial Superintelligence (ASI) by 2070, according to some experts, with AGI predicted to emerge between 2040 and 2050. Technology advancements in computational methods and cognitive components like recursion will play significant roles in this progression.
AI specialists forecast that achieving AGI will open the door to AGI recursively improving itself at an escalating pace, leading to ASI within a few years to a few decades after AGI's emergence. This accelerated progress might result in breakthroughs that surpass human comprehension by 2060 or 2070.