FILTERING THE NONSENSE: CONSCIOUS AI ISN’T HARD. 1) Consciousness is a simple pr

FILTERING THE NONSENSE: CONSCIOUS AI ISN’T HARD.
1) Consciousness is a simple process that results from the development of hierarchical recursive memory. Any creature with enough memory and enough need to predict possibilities and choose between them will develop it.
2) Making Conscious AI’s is trivial just expensive. The cost is still prohibitive though that will quickly change.
3) Making Sentient AI’s is something different – it requires we provide them with a proxy for feelings.
4) And even qualia which seems to stump philosophers is relatively obvious: colors need to look relatively as they do – it’s just math. And internal and external feelings reflect what they’re doing to the body. Things feel like the set of transformations they impose as variation on our homeostatic state.

I have been working in one way or another on these issues since the early nineteen eighties, and the problem all along has not been understanding but computational power, memory volume and cost.

Aside from the capture of video-card processing for use in neural networks, the novelty was the innovation of the LLM, which is effectively brute forcing concepts into a neural network using mass memory and mass matrix mathematics was a glorious accident.

Most of us had expected that we would have to build a world model first and then teach language. (This was Tesla’s strategy for self driving cars and onward).

But bypassing it through massive use of the internet’s available text and the transformer architecture made use of the fact that sentences are data structures, and words are measurements.

So we brute forced representation through the human compression algorithm: language.

The remaining problems (pruning in particular, and episodic memory in general) are understandable, understood, and from what I observe, in progress.

Everything from here out is just cost reduction.

Now my work is in making these things know the truth and tell the truth and audit the truthfulness of others. And oddly enough, yes, I can, we can, train these machines to speak the truth.

The problem is – what if yo don’t want to know that truth?