What is Artificial General Intelligence (AGI)?

Artificial General Intelligence describes a hypothetical type of AI that is capable of matching human intelligence or, someday, even surpassing it.

The exact origin of this phrase is a little hard to pin down, being a natural follow-on from John McCarthy’s phrase “artificial intelligence.” However, it has been noted that the phrase likely reemerged into popularity (or was invented) via the work of Ben Goertzel and Shane Legg back in 2022.

What are the characteristics of AGIs?

While the term AGI is understood in a general sense, the exact criteria that would define “intelligence” in the case of AGI are not agreed upon. Broadly, it is believed that this intelligence would have to possess the ability to plan, learn, apply knowledge, and reason regardless of uncertainty.

Beyond the obviously very complex nature of trying to program a computer to be able to do these things, it also requires that we have a fairly strong grasp of what these criteria actually mean. Frustratingly, even though we do have a good notion of these concepts, scholars and academics hotly debate many of our presumptions about “intelligence” as a concept.

Hence, many academics will also claim that a true AGI would have to be able to recognise threats and perform imaginative acts, which are also very difficult to define. Computers are fundamentally logic-based machines that use a lot of rather simple binary math to complete very complex tasks.

This makes things like innovation, imagination, and rationalization in the face of threat or uncertainty very difficult to compute or even to emulate convincingly.