What is ASI and how does it trump today's AI?

Artificial superintelligence is hypothetical and could be decades away from reality

Artificial superintelligence is expected to surpass human intelligence in every aspect. Reuters

Meta Platforms boss Mark Zuckerberg is at it again with artificial intelligence, this time pledging to accelerate the development of AI's future.

The chief executive of Facebook's parent company has doubled down on his plans to develop artificial superintelligence (ASI), a technology not quite here yet but that's seeds he's planting.

"For our superintelligence effort, I'm focused on building the most elite and talent-dense team in the industry," Mr Zuckerberg said in a post on Threads.

He added that Meta would spend "hundreds of billions" of dollars to build this AI of the future.

What is ASI?

ASI is still an entirely hypothetical software-based system but is expected to have an intellectual scope surpassing human thinking capability and going far beyond the capacity of today's AI.

Its highly advanced cognitive functions and developed reasoning skills would surpass those of any human, IBM says.

However, at this point, ASI is still a future technology and does not exist yet, as today's innovations are too limited to fully build it.

To put this into perspective, the current level of AI is considered as artificial narrow intelligence, or weak AI, which, even though it can excel in specific tasks such as those presently carried out by generative AI, it still relies on pre-programmed algorithms and needs human intervention to operate.

More importantly, weak AI cannot learn new skills or develop a deep understanding of the world, IBM explains, because "the [human] brain's workings are still not fully understood making it difficult to recreate via software and hardware".

"Not all thinkers are aligned on the feasibility of something like an ASI. Human intelligence is the product of specific evolutionary factors and may not represent an optimal or universal form of intelligence," IBM added.

What are the areas in which ASI would excel? Decision-making and emotional intelligence most notably, in addition to boosting what present AI can already do with creativity and problem-solving,

How does ASI compare to quantum AI?

Advanced general intelligence (AGI) is another advanced form of AI capable of performing any intellectual task that a human can. It's also at a theoretical stage.

Unlike current AI systems, which are designed for narrow – or "weak" – specific purposes such as language translation or image recognition, AGI aims to understand, learn and adapt to a range of jobs across various industries.

AGI and ASI, in some cases, have been used interchangeably. The key difference is that while AGI would possess human reasoning skills in a wide range of specific applications, ASI is expected to surpass human intelligence in every aspect.

Quantum AI, meanwhile, is powered by large quantum models are also advanced AI systems designed to analyse, process and generate quantitative data rather than natural language, which is the basis for the current large language models that power generative AI.

“Word-based AI is trained only on the words of the internet," Jack Hidary, chief executive of Alphabet spinoff SandboxAQ, told The National. "So if it's not on the internet, it doesn't know it … we produce data in a very different way [like] from actual equations."

What is Zuckerberg promising?

Much like his highly criticised metaverse plans, Mr Zuckerberg is, again, vowing a lot - including the big bucks.

On another Threads post, he said his company is "going to invest hundreds of billions of dollars to build superintelligence". To hammer down that point, he stressed that Meta has "the capital from our business to do this".

As of Tuesday, Meta has a market capitalisation of $1.81 trillion, which is the sixth highest globally and one of only 11 companies with a trillion-dollar value, data from CompaniesMarketCap shows.

The National explains Meta's decision to discontinue fact-checking

02:30

In its fiscal first quarter, Meta's revenue jumped 16 per cent annually to $42.31 billion, while profit leapt 35 per cent year-on-year to $16.6 billion - more than analyst estimates and defying worries over the effects of US-imposed tariffs.

Meta is scheduled to report its second-quarter results on July 30, with sales in the range of between $42.5 billion and $44.5 billion, according to chief financial officer Susan Li, in line with analyst projections.

The company and its boss have been obsessed with futuristic technology and have pledged to invest in both costs - billions, reportedly - and recruitment.

When is ASI expected to be realised?

This is where it gets a little sticky, as predictions on when ASI - or even AGI - will become a reality range from the near future to decades away.

A study conducted in February by AIMultiple Research principal analyst Cem Dilmegani found AI experts expect a 50 per cent chance that machines may achieve human-level intelligence anytime between 2040 and 2061.

Meanwhile, a 2024 study conducted by researchers from AI Impacts, Germany's University of Bonn and the UK's University of Oxford found the possibility "of unaided machines outperforming humans in every possible task was estimated at 10 per cent by 2027 and 50 per cent by 2047 ... if science continues undisrupted".

The updated analysis, conducted in February by Cem Dilmegani, principal analyst at AIMultiple Research, combed through about 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 is a 50 per cent probability that we would achieve human-level intelligence in machines at some point between 2040 and 2061, the analysis found.

Is there a body dedicated to ASI?

Yes - the Artificial Superintelligence Alliance, a collective that describes itself as "the largest open-sourced, independent entity in AI research and development", according to its LinkedIn page.

The group comprises Fetch.ai, SingularityNET and Ocean Protocol and Cudos, and aims to "accelerate the advancement of decentralised AGI and, ultimately, ASI".