[TheAceOfHearts]

> At the core of most definitions you’ll find the idea of a machine that can match humans on a wide range of cognitive tasks.

I expect this definition will be proven incorrect eventually. This definition would best be described as a "human level AGI", rather than AGI. AGI is a system that matches a core set of properties, but it's not necessarily tied to capabilities. Theoretically one could create a very small resource-limited AGI. The amount of computational resources available to the AGI will probably be one of the factors what determines whether it's e.g. cat level vs human level.

[pants2]

While we're posting our favorite definitions of AGI, I like what Francois Chollet once said:

"AGI is reached when it’s no longer easy to come up with problems that regular people can solve … and AIs can’t."

[dr_dshiv]

That’s like Peter Norvig’s definition of AGI [1] which is defined with respect to general purpose digital computers. The general intelligence refers to the foundation model that can be repurposed to many different contexts. I like that definition because it is clear.

Currently, AGI is defined in a way where it is truly indistinguishable from superintelligence. I don’t find that helpful.

[1] https://www.noemamag.com/artificial-general-intelligence-is-...

[mlyle]

I think "being able to do as well as a 50th percentile human who's had a little practice," on a wide range of tasks, is a pretty decent measure.

Yes, that's more versatile than most of us, because most of us are not at or above the median practiced person in a wide range of tasks. But it's not what I think of when I hear "superintelligence," because its performance on any given task is likely still inferior to the best humans.

[fragmede]

That seems like a personal definition for super intelligence. I don't think I'm alone in assuming super intelligence needs to be greater than all humans for it to be considered super vs "pretty good intelligence".

[mlyle]

> > I think "being able to do as well as a 50th percentile human who's had a little practice," on a wide range of tasks, is a pretty decent measure.

> That seems like a personal definition for super intelligence.

I was giving a definition for artificial general intelligence as distinguished from super-intelligence, since the poster above said that most definitions of AGI were indistinguishable from super-intelligence.

To me, a computer doing as well as a practiced human at a wide swath of things is AGI. It's artificial; it's intelligence, and it's at least somewhat general.

[fragmede]

Ah. I misread you then. Apologies.

[dr_dshiv]

Remember, that AI is a “jagged frontier.”

AI is already better than a 50th percentile human on many/most intellectual tasks. Chess, writing business plans, literature reviews, emails, motion graphics, coding…

So, if we say “AI is not AGI” because 1. It can’t do physical tasks or 2. it can’t replace intellectual human labor yet in most domains (for various reasons) or 3. <insert reason for not being AGI>, then it stands to reason that by the time we reach AGI, it will already be superintelligent (smarter than humans in most domains)

[mlyle]

> then it stands to reason that by the time we reach AGI, it will already be superintelligent (smarter than humans in most domains)

> > Yes, that's more versatile than most of us, because most of us are not at or above the median practiced person in a wide range of tasks. But it's not what I think of when I hear "superintelligence," because its performance on any given task is likely still inferior to the best humans.

> AI is already better than a 50th percentile human on many/most intellectual tasks. Chess, writing business plans, literature reviews, emails, motion graphics, coding…

Note the caveat above of "with some practice." That's much less clear to me.

[dr_dshiv]

> That's much less clear to me.

But you must see that, if this point falls away, so does the argument.

“With some practice” surely means less 100 hours of practice. Not extensive practice, or PhD level practice or master-level practice.

Will the average person, after 100 hours of practice, really be rated better than AI at music, writing, visual art, coding, research, etc?

[mlyle]

??? most of us have a lot more than 100 hours of practice at things like this. Probably "some practice" means roughly the secondary background most people have, then undergraduate plus a couple of years of industry work?

If we're thinking about what human activity it will or can displace, it's probably unfair to compare it a seasoned doctorate. But it's also probably unfair to compare it to a kid with a few months of music lessons, too.

[Dylan16807]

That definition gives me a headache. If it's not up to the level of a human then it's not "general". If you cut down the resources so much that it drops to cat level, then it's a cut down model related to an AGI model and no more.

[turtletontine]

What does this even mean? How can we say a definition of “AGI” is “correct” or “incorrect” when the only thing people can agree on, is we don’t have AGI yet?

[tim333]

Of course AGI is just an abbreviation for artificial general intelligence and stuff like GPT-5 is artificial, somewhat intelligent and somewhat general.

I think the goalposts for "AGI" will keep moving so current AI doesn't match it.

I've thought of it as human level but already people are saying beating the average human isn't enough and it has to beat the best and be nobel prize worthy.