Covering Disruptive Technology Powering Business in The Digital Age

image
The Countdown to Superintelligence: Are We 5 Years Away From AI That Aces Exams (and More)?
image

Written by: Izzat Najmi Abdullah, Journalist, AOPG

Public attention often goes to the flashier side of Artificial Intelligence (AI), like the fantastical creations of tools such as Midjourney and ChatGPT. However, this is just the latest chapter in a much longer story. While these generative AI applications are undeniably captivating, numerous behind-the-scenes AI applications are quietly transforming various industries. From healthcare to consumer technology, AI’s influence is undeniable, and as it continues to advance, so too will its impact.

But have you ever thought of how soon we can expect truly intelligent AI capable of mimicking human thought? How soon will superintelligence come about?

This question on superintelligence connects to the debate surrounding the Turing Test, a thought experiment proposed by Alan Turing in his 1950 paper. The test involves a human judge holding a conversation with a human and a machine, both disguised. If the judge can’t reliably distinguish between them, the machine is considered intelligent. While some early programs like ELIZA achieved success, the Turing Test has limitations and is criticised for not measuring true intelligence.

Superintelligence

So, how close are we to achieving human-like AI, to getting superintelligence? Nvidia’s CEO, Jensen Huang, hints that it is within reach, sparking a debate about the imminent arrival of Artificial General Intelligence (AGI).

At a recent economic forum, Jensen offered a surprising timeline for AGI, depending on how we define it. For Jensen, if AGI simply means the ability to pass any test we throw at it, then it might be just around the corner. He believes that within five years, AI could potentially ace everything from law school exams to specialised medical tests. This prediction is based on the rapid progress AI has already made. Today’s AI can conquer legal bar exams, showcasing its impressive learning capabilities.

However, Jensen acknowledges a crucial caveat. This definition of superintelligence might be a bit narrow. True AGI, according to some, would require a deeper understanding—an ability to mimic human thought processes. Here, things get murkier. Scientists haven’t even reached a consensus on how the human mind actually works. This lack of a clear target makes it difficult for engineers who rely on well-defined goals to build such intelligence.

Jensen’s comments, however, highlight an important point. Even if AI can’t replicate human thought processes exactly, its ability to learn and adapt across various tasks is a powerful sign of its growing intelligence. The challenge lies in harnessing this potential for good.

The rapid advancements in AI, from the captivating creations of tools like Midjourney to the success of programs like ELIZA in the Turing Test, paint a picture of a future brimming with possibilities.

Jensen’s distinction between AI that aces tests and true AGI mirroring human thought definitely emphasises the complexities involved. While AI’s ability to learn and adapt across various tasks is undeniable, the lack of a clear understanding of human cognition presents a significant hurdle in replicating it entirely.

In the coming years, we will likely see continued debate surrounding the true nature of AI superintelligence. Will machines ever truly think like humans? Or will their strength lie in augmenting human capabilities, leading to a future of human-AI collaboration?

(0)(0)

Archive