There's no universally accepted definition of AGI. You can try and google about it, read it on Reddit, scroll through X - and you wouldn't really find one single common artificial general intelligence definition. The ambiguity is not merely academic - it highlights the complexity to replicate or surpass human intelligence within machines.
What is clear is that we aren't there yet, at least that's our stance. We may be at a point where it's possible to tick off the various milestones, instantiated as part of the journey towards AGI, but it's clear there is another that needs fulfilling: that of setting a task and rationalising it.
Wikipedia, has an interesting definition of AGI as 'a hypothetical type of intelligent agent which, if realized, could learn to accomplish any intellectual task that human beings or animals can perform.' However, the use of 'perform' in this context opens a Pandora's box of semantic and philosophical challenges. Performing is inherently subjective - varying drastically across the human experience.
This variability begs the question: Can we develop AGI without anchoring it to human reference points? And if so, does the human benchmark become irrelevant in defining AGI's scope and potential?
The comparison to human ability is a double-edged sword. On one hand, it provides a tangible framework for monitoring AGI's development. On the other, it raises the issue of variability among humans themselves. Here's a thought experiment: if some humans can't grasp complex theories while others can, to which standard do we hold AGI? This discrepancy highlights the challenge of establishing a clear benchmark for AGI, leading to a broader, more fundamental question:
Is it necessary for AGI to fulfill human criteria to be considered genuinely intelligent?
To answer this question, let's get into the evolving definition of the term.
The origin of the term Artificial General Intelligence is often traced back to a military technology research paper called 'Nanotechnology and International Security' that refers to systems that rival or surpass the human brain in complexity and speed. The term today has evolved and often incorporates aspects of consciousness, self-improvement, and even the singularity - a theoretical point where machine intelligence surpasses human intelligence.
The vox populi definition of AI has always been tasks humans can do that computers can't do - but the tasks been continuously moving goalposts since decades. Initially, tasks like playing chess were deemed the exclusive domain of human intellect. Yet, as AI systems like Deep Blue and later, more versatile models emerged victorious in these arenas - the benchmarks for what constitutes true intelligence shifted. This progression underscores a key insight: Our understanding of intelligence: both human and artificial - is in a constant flux, shaped by ongoing discoveries and technological breakthroughs.
This evolution in understanding reflects the ever-changing nature of the AI field, where distinctions between narrow AI and AGI blur as advancements are made. Today's AI excels in narrow domains: for instance, Fetch.ai's AI agents and GPT's linguistic capabilities highlight significant advancements. However, even these advanced systems lack the generality of human intelligence. They are confined to their programming and training and do not tackle tasks they weren't specifically designed for. AGI requires a leap beyond this: a system that can adapt, learn, and innovate across any intellectual terrain. This necessitates not just advancements in computational power and algorithms but a deeper understanding of how intelligence itself can be abstracted and replicated in silicon.
The challenge is not only technical but conceptual, requiring breakthroughs in our understanding of cognition, learning, and the neural underpinnings of intelligence.
This shifting landscape suggests that AGI cannot be pinned down to a static definition. Instead, it represents a spectrum of capabilities, constantly expanding as new horizons in AI research are explored. The distinction between AGI and narrow AI becomes less about the tasks they can perform and more about their adaptability, learning capacity, and the breadth of their applications.
The scope of AGI transcends merely mimicking human thought, opening avenues for innovation in creativity, analytical thinking, and comprehension. Pursuing AGI transcends the boundaries of mere technical endeavor: it invites us into a philosophical exploration, urging a reevaluation of what constitutes intelligence, awareness, and the involvement of artificial entities in shaping our tomorrows.
The journey towards realizing AGI is complex and multifaceted, weaving through advancements in technology, ethical considerations, and the collective ambitions of society. As we progress through this varied landscape, the evolving benchmarks of intelligence highlight the continuous interaction among human understanding, machine capabilities, and the changing matrix of challenges and prospects that spearhead our pursuit of AGI.
