I am amused by giant multi-billion dollar company leaders still throwing around the term AGI with no real plan to get there. Hype mongers the lot of ‘em.
No one has clarity on the proper approach because no one has a reasonable north star.
So, in this post I’ll share my viewpoint on what the goal should be.
Our goal should be machine consciousness. When I say consciousness I simply mean the capability:
- to oversee an ensemble of other models that together predict the world and
- the ability to alter that ensemble to produce better predictions
In fact, machine consciousness already exists — it is the engineers who gather the data, view the results, and then tweak the model to improve it. We need to do is shift our focus to building macro structures that perform this meta process.
What this gives us is self-directed learning. The machine must identify for itself where it is lacking and generate “ideas” for how to close the gap. That is the essence of a “machine intelligence” and will lead to more general problem solving that we can ladder up.
Checkout The Ingredients for Limited Machine “Consciousness”.