Why we need a new paradigm to achieve AGI

From the moment ChatGPT created a new reality around AI (artificial intelligence), everyone began speculating that we are near an AI apocalypse.

Terms like AGI, ASI appeared and everyone counted backwards until news of the end-of-computing-as-we-know will happen. But, it didn’t happen and as it seems, it will not happen soon.

Like a combustion engine

When ChatGPT appeared, everyone was asking me, what will happen. You know, as CTO, people around you, think that you now these things, as long as they involve computers.

At the time, I told people,

It is like the combustion engine. First, the actual performance was below 20%, which basically meant that it created more heat, than kinetic energy to move the car. That changed in time and the performance rose to 30–40% (I do not know the specifics), but it will never exceed this. This is the same thing with LLMs (Large-language models). It will be better and hugely positive in some areas. It will disrupt the way we are doing things, but it will reach a plateau and then we will have to innovate again.

Why?

Basically, it is the computational model. Even those wonderful technologies, are implemented and have to follow an if … then … else approach.

These models, have calculates weights that are used to do tons of calculations with each pass and in the end predict the next token. By predict we refer to the result with the highest probability. This is it, more or less.

It works wonders, but it cannot escape its fate.

A new paradigm

So, it is statistics on steroids. But whatever we do, it cannot escape the fact that it is if … then … else. We need a new programming paradigm, not technologies to do matrix multiplications faster. Ok, this is useful also … but still.

I do not what this will be. Nobody knows and this is the main reason we will not reach AGI soon. Yes, we will have glorified machine learning in production. Yes, it will disrupt our lives and it will be everywhere. But it will not be AGI.