What Exactly Is AGI?

Leading AI labs like OpenAI, Anthropic and Google DeepMind are in a race to achieve AGI.

But what exactly is AGI?

AGI stands for Artificial General Intelligence. It's when an AI can learn and do a wide range of intellectual tasks at a level comparable to humans.

It doesn't have to be the best at everything. Humans aren't either. But we can learn different kinds of work and apply what we know to new problems.

But doesn't AI already do that? It writes code, makes presentations and analyses Excel sheets.

Yes, and that's why this question is getting harder to answer.

Today's AI can already do many different things. It can even attempt something it didn't encounter during training. If your company builds a new app, you can give AI the instructions and it can try to figure out how to use it.

It doesn't need separate training for every new assignment.

Then what is missing?

Look at what happens when you give AI a longer assignment. It might do some parts very well, then make a basic mistake, miss an instruction or need you to explain what to do next.

For AGI, we would expect it to handle many different kinds of work with competence comparable to humans, including when the problem is unfamiliar or things don't go as expected.

Today's AI shows some of these abilities. The question is how far they extend and how consistently it can use them.

And would that mean AI becomes conscious?

Not necessarily. Being able to do the work doesn't tell us whether it feels anything. Saying "I feel pain" doesn't prove that it experiences pain.

So how will we know when we have reached AGI?

That's where researchers disagree. What work should it be able to do? How well? And how much help should it need?

Until we agree on those things, a company announcing AGI won't settle the question. We'll need to examine what it has actually demonstrated.