What should ChatGPT output for us to call it an AGI? I used to think it was ‘new knowledge’. But I was corrected by a Twitter user and AGI researcher Carlos De La Guardia (@dela3499): there is nothing preventing an AI from creating new knowledge. It is constantly outputting words and there may be new knowledge somewhere in there. What then is the difference between AGI and AI?
To view this problem more clearly let’s compare AI to existing GIs: Us. Humans are general intelligence.
What are the differences between humans and ChatGPT? The differences fall into a few categories:
Creation.
Criticism.
Motivation.
Inner life.
Creation: Humans can create new knowledge out of existing knowledge. This knowledge is not merely information: it is useful. ChatGPT can create new knowledge but it won’t be useful knowledge. The useful knowledge that it outputs is already existing knowledge. Why can’t ChatGPT create useful new knowledge? This brings us to point 2.
Criticism: Humans criticise the ideas they conjecture to improve them. This makes them useful. ChatGPT does not criticise knowledge with the purpose of making it useful because it is not trying to solve any problem. Why?
Motivation: ChatGPT does not have motivation. It merely responds to a user’s input. Humans are motivated by problems. We solve problems and by solving problems we create new explanations about the world. Once the user’s input is over nothing happens inside ChatGPT. Because…
It does not have an inner life. What is an inner life? If we knew exactly what it was we could code it into ChatGPT. But we don’t. It is akin to a permanent computer that is constantly creating and criticising knowledge, forming new theories, and discarding old ones. An infinitely complex web of ideas constantly changing each other, disappearing, appearing, qualia, memories. And yet despite this, we function. How? If you know you can make the first AGI.
Can ChatGPT ‘scale up’ to an AGI? No. It is a qualitative difference. It is like asking if a calculator can ‘scale up’ to a human. Put simply we do not know what makes us general intelligence. And till we do we cannot make one.