Y’know: come to think of it that’s the same three things our own brains are mostly trained on… I’m sure that without mammalian proprioception and weird bits like saccades the result is just sparkling cognition though.
You have to admit the prospect looks more likely than ten years ago.
At least LLMs include Model in the name, which is an accurate description of the tech
And actually, I've heard ppl just using the word "models" by itelf and that might be the best way to go
Why is it better to refer to something in terms of the technology it's based on?
Intelligence is not a property of the material universe and will not 'emerge' in some special way from machines. That's a good science fiction story device (and actually the idea extends back in human storytelling before sci-fi) but ppl thinking that the idea carries through into the real world are suffering from what I call "Science Fiction brain". (Or they are simply fraudsters, happy to talk rubbish to prop up a bubble.)
Or on a technical note because certain brain structures etc. might not exist in an LLM and stuff like a default mode network and sensory input and a world to move in, is missing?
Basically, I'm not a materialist. (Which was a widely held position in philosophy across millennia and across cultures up until just recently. I treat consciousness as fundamental, not the material universe . Thus 'conjuring intelligence' into machines is simply not possible.
BTW, machine learning is a totally valid concept from where I stand. But it's 'dumb' learning.
Like the people who got certain brain injury and their whole persona changed or certain capabilites were lost?
There also research papers that certain brain structures have emerged in ML models.
At any rate, sure, not being at odds with the existence of consciousness might be a trait of the more reasonable strands of materialism (if there's even such a thing), but they will always see it as emergent/derivative and often even epiphenomenal, otherwise they wouldn't be materialist stances.
So am not sure where you see the logical misstep here, but am happy to hear. The point of the previous comment is that framing an ontology whose primitive is C, is in no way "esoteric", term that is usually used in contraposition with "rational". Taking C as epistemologically or even ontologically prior is the rational stance, not only on accounts of parsimony.
A human is born without any sensory connection to their body, except hearing. They can't feel their body or anything on their skin. They're blind. They can't taste or smell or touch. They can only hear, and make clumsy sounds in response, with no feeling of what their mouth is doing.
Are they intelligent?
AI is here doing literaly the same job as a human who needs to be intelligent.
I see your stand as hubris from your side.
For such a strong viewpoint, you deliver not any context of why you think like this.
I would say yes it understands things but it was a causios 'understand'
I think what's really underlying the "fear of calling something intelligent" we have, is more about who we are as a species.
We historically used intelligence as a yardstick for racial and species superiority. We're always judging whether something or someone deserves respect, deserves consideration. We judge and decide who is superior.
We do it amongst ourselves so of we are projecting that shit onto AI now. And we are afraid of anything that threatens the superiority complex we have as a species.
Why? Because our behaviours (call it intelligence, whatever) is how we survived, how we "won" in the fight against nature. We're at the top now. It's safe here.
So anything that threatens that position is going to be called an enemy because we're still a scared creature jumping at shadows in caves. Can't escape that, it's where we are from.
That's why we are both so fascinated by, and mortally afraid of, something else having "intelligence". It's our only weapon, our only defence.
The TV show "Humans" from ages ago made a good story about what AI brings out in us as a species. We both crave connection with something, but at the same time are horrified and incredibly afraid.
Anyway my point is I think we do know what intelligence is, if we just start being brutally honest with ourselves. It's not that complicated really.
I'm not sure how this is. The amount of things machine learning can do now is pretty wild vs. what it could do a year ago. It's not even in the same realm as what it could do 5 years ago.
If you can't measure or observe it then you're just making it up and at that point everything's valid and right provided it's what I believe. And if you disagree with me then clearly you are not adequately human and we will need a crusade to rid the world of your kind.
On the contrary, we have rarely been in so much existential danger as we are now. There was that one ice age that nearly got us, but we are messing with the foundations of our ecology right now. When the bugs go, we go soon after.
A 12 year girl from a third world country with little education and a normal cloud ai would far surpass me at work if I did not have access to AI, yet comes close if we both had it. AGI already passed a few months ago. SI or happy ai or whatever the us wants to call it may or may not come but if it does it’s skill set may drown out my understanding of what is even occurring.
Here's a nice test: Can an AI fully substitute you?
I don't mean replacing you at your job. I mean, if you take a human body and route its higher brain function to an AI, would it be able to perform convincingly?
I would say currently we're very far from such a scenario.
When I consider the option to buy a shirt, the problem is considered in a large amount of dimensions intuitively (is it warm enough, does it fit, does it clash with my wardrobe, will my girlfriend like it, will my friends laugh at me for wearing it, does it make me look older, is it a good price, are sales happening soon, is it annoying to clean, can I afford it, will carrying a bag with the purchase annoy me for the rest of the day).
Currently AI can offer good answers for any of those dimensions if explicitly asked, but it is pretty bad at the "intuitive creativity" of surfacing those points of consideration when relevant. It will at most chose a couple of criteria and fixate on them.
I think this ability, however it's defined, is a missing prerequisite for AGI. Without it, you can have something that outperforms humans at every individual task and still sucks in real world decision making.
Why do people insist on talking about AI like it’s some supernatural, otherworldly creation? No matter how advanced it is, it’s still human technology.
Ask a model whether a human purchasing clothes would care about clothes fitting well, or the price, or the opinion of their romantic partner. The model will in all likelihood get the answer right. Familiarity with human preferences is not what's missing, which is hardly surprising considering an LLM is trained on human-made content.
What it lacks is the ability to notice and surface these non obvious constraints when relevant. This is fundamental to real world decision making.
Maybe we can get something that resembles intelligence but in reality is another thing. Maybe intelligence is emergent but not the only thing that can emerge, maybe there are things that share some characteristics with the intelligence we know and aren't intelligence.