The Dictionary Transition for AI

There is this popular term people use "AGI" (Artificial General Intelligence). AGI is a milestone or threshold people talk about where AI can do anything a human can.

However I think this is nonsense. If you could make an AI which can do anything a human can, it would be better than us not equal.

Firstly because anything implies everything. Current AI tools like ChatGPT are complete polymaths, i.e. they have expertise in every domain. I can't imagine why the same wouldn't be the case once we make humanoid robots.

And AI has an inherent advantage over humans. Copying the human brain takes 18 years. And it still doesn't have it's parent's memories. Whereas copying an AI takes minutes or seconds.

Also humans are extremely bad at interfacing with computers. Transferring your memories into a computer requires to type or draw it using a software package which can take hours. Also most people are extremely awful at programming computers.

AIs don't have these problems. The current state-of-the-art AIs already surpass humans in their ability to interface with computers. This is because LLMs like ChatGPT have encoders that convert between images/text and embeddings in their neural networks.

In other words what takes you an hour (such as typing a blog-post) is done for ChatGPT automatically in the blink-of-an-eye by a piece of software called an encoder (tokenizer for text). It's like if you had the computer directly read the neurons in your brain and translate it instantly.

So I claim that the first AGI will:

• have expertise in every subject
• be infinitely copy-able
• be able to interface with computers at-least 1000-times faster than a human (probably a million times)

By the way LLMs already have all those three abilities. You can download Llama 3.1 and run it on any computer with a powerful GPU.

So the term AGI is essentially nonsense. What we will have will actually be ASI. Artificial Super-Intelligence.

So when this transition happens where AIs become smarter than humans, I allege it will be like the dictionary transition. Instead of AI being like a simulation of how humans think, it will become an authority on how to think. I.e. the transition from descriptive to prescriptive.

Steve Wozniak proposed a definition for AGI that it can walk into his house and make a cup of coffee (ref).

In that case I take issue with not just the "G" in AGI but also the "A". Why is it artificial?

If an AI makes a cup of coffee, is it an artificial coffee?

If an AI-powered self-driving car crashes, is it an artificial crash? I reckon if you were a passenger it it would feel just as real as if a human driver crashed the car.

If an AI robot slaps you in the face, is it an artificial slap? It probably hurts just as much as a real slap, so why do we call it artificial?

The only reason the AIs of today are "artificial" is just because they are disembodied. The moment you give AI direct control over the real world it stops being artificial immediately.

So by this point we started with AGI, replaced the G with an S, and subtracted the A. This leaves us with: SI = Super-Intelligence.

But the word "super" only exists to distinguish it from human intelligence, and if we're being honest most humans are pretty stupid anyway. So we can drop the S as-well and just call it "I" for intelligence.

Thus the Artificial Intelligence becomes the actual Intelligence in the same way that the Dictionary transitioned from descriptive to prescriptive.