'I Worked on Google's AI. My Fears Are Coming True'

FWIG one materialist view of sentience is that it's an emergent property - it is a way of describing a state of a complex system.

The question, then, is: what are the properties of the system that the term describes?

In this case, it looks kinda like: memory, self-awareness, and understanding. Memory isn't a question - it's possible for these things to remember (even if sometimes that memory is in the form of access to data on the internet). Self-awareness is trickier - but by most measurable schema, the machines seem to be doing as well as humans do in this regard (they can describe themselves and talk about their place in the world in context). Understanding is a topic of hot debate - but I think it tends to fall to the same problems as that of sentience generally; that is, we have this concept of what understanding is, and we know enough about how LLMs work to say that they're not the same, but if you start poking at how to measure it, we're back to square one vis a vis what's apparent might not be what is.

Yes, but I don't think it's the case.

we give a lot of those people cabinet positions.

>>sentience is just high-order analysis of lower-order processes with the ability to impact the lower-order processes....we will see it within the next 10 years.

Definitely plausible timeline, but we're not there yet.

I think it'll take a good number more layers with equal or greater sophistication, between and on top of each other.

The current text and graphic tools make only associations, they have zero conceptual knowledge. Enough associations make it look good, but they don't make an abstract concept. E.g., shown enough pics of an astronaut, it draws a good astronaut. But, it can also generate pics of a bikini-clad woman with both her bum and her torso & face facing us (literally putting the bottom half on backwards). Or, if asked for a pic of a specific person, responding with the same picture, background and all, when there were not high numbers of examples input. It isn't figuring out what is the person's image, and how body parts go together, and how the subject is different from the background the way even children do when drawing.

Same for text, where we can ask it "Mike's mom has three kids named Alice, Bob, and Chris; what is the name of her fourth child?" it says "Not enough info". When told that the answer is in the question, it doubles down on explaining how there's insufficient info. It posts entirely fabricated references.

These are not even close to intelligences; they're only fancy statistics which are a fantastic parlor trick and sometimes useful.

Add another one that understands the 4D physical world, then combine it's understanding of billions of 2D photos, and trillions of words, and MAYBE it'll be able to properly abstract the concept of a person, with the right number of limbs, in the right relationship with each other, and the right relationship with the chair, and the right relationship with the skin, hair, eyes, clothing, and background etc., and an appropriate understanding of the input text to reliably generate an original output that actually makes sense in the world. Add several layers on top of that, and maybe we're approaching sentience...?

Two things here:

1. In the sense presented by the media it is clear they want to point toward the idea that sentience AI is a conscious AI => using this idea your dog is not sentient

2. Dogs are legally sentient beings but in this conversation should make a difference between making a law like this to protect them (which we should do as the legal system needs to name things to work) and the idea that something has a level of sentience that includes consciousness that can be comparable with humans.

In my mind, this is more about levels of complexity. Say we do a scale of matter aggregations that creates a level of complexity that will create new abilities out of this complexity.

This would be a scale and not a 0/1.

Thus we and animals are let's say inside an interval from 50..100. And sentience is from 50 to 100 in that imaginary scale. Humans are close to 100 (as we are the ones defining the scale) and animals start at 50 some of them are at 70, some at 80, and maybe some at 90.

Now, what is the main purpose of these articles, the underlying idea: is that AI is at 100 or that is close to being at 100.

And my main point was that AI is not there. I am not even sure AI is at 50, for me AI is between 0 and 50 on that scale. It is evolving but it is still very simple in how it experiences it's external and internal world. It talks like a being that is on that scale over 50, but it is not there.