With artificial intelligence becoming a major part of present day every life, and major developers gunning for artificial general intelligence (AGI), the big question is:
What should AGI look like?
To answer that, it helps to first understand what intelligence is:
Intelligence is a broad mental capability encompassing the ability to learn from experience, adapt to new situations, understand abstract concepts, and use knowledge to manipulate one’s environment
We, or at least "I" have come across multiple reports where key AI figures or builders within the industry are being quoted for saying they lack knowledge of what AGI should look like or if current models have reached there.
Understanding it is rather simple
Building it is the hard part.
Although these key AI figures would make statements that suggests that current models may have reached AGI and they are not able to tell, the truth is that it's rather simple to understand and acknowledge when we've reached AGI.
We will visibly see and feel the shift from viewing the technology as isolated tools but integrated infrastructures that solves real, immediate problems.
Think about image and video generation for a minute.
Given my recent deep involvement with AI research, seeing that it is inevitable, I've come across multiple instances where it has appeared that people are currently struggling to distinguish between what's real and what's AI generated.
This isn't about being able to determine if an AI wrote a text—people seem to be experts here these days—but about determining if an image or video is real.
We are seeing these concerns show up in political stories and every day events.
This struggle directly proves that AI has achieved an impressive level of advancement in image and video generation, because clearly the average person is struggling to tell the difference.
AGI should look no different
It's very crucial to understand that at the end of the day, the benchmark should technically be about measuring against the average person.
What is the general intelligence of an AI model against the average most intelligent beings on planet earth?
In some ways, that framing may trick people to believe we already have AGI since AI holds more knowledge—by a large margin—than the average human.
But that would be a flawed conclusion because possessing storage of information and pattern recognition in itself isn't general intelligence.
As the previously quoted definition reveals, AGI needs to be able to learn from experiences, be adaptable, understand abstract subjects not in its training data and also be able to use its knowledge (experience in real sense — not stored data) to manipulate it's environment.
This shift moves AI from tools to autonomous agents with consistent top performance across multiple tasks, across multiple domains, generating real economic value.
An advanced multimodal capability in one agent. AGI is where reality collapses and you start questioning—without attaining clarity—if the person you're talking to physically or digital, is running on some data center using up clean water or if they are running on red blood.