Nothing quite that complicated, I’m afraid; it’s just an alignment problem:
The problem is that, even between two humans communicating through speech and certainly between a human and an AI system that makes decisions in a very inhuman and alien way, you can have serious issues in defining what you want the agent to do in a way that the agent can understand and follow.
That’s certainly true, because alignment problems are a universal feature of system design. Any system we build, AI or not, has to worry about this stuff.
Roman engineer Julius Sextus Frontinus (circa A.D. 10) wrote, “Inventions have long since reached their limit, and I see no hope for further developments”.
I think he was a little off base with that comment.
There will always be new inventions, at least for the foreseeable future. Some will be relatively little ideas that, for some reason, no one thought of before. Others will be momentous.
Alignment, by that term, is a commonplace in any discussion involving management or agents. I know you know that, but some folks might not.
When selling a house, getting your sales agent to have the same goals you do is an example of an alignment problem.
You both want a sale, and sooner is better for both of you. But the terms of the tradeoff between “How much cheaper price to accept” versus “How much quicker / easier sale to make” are very different between you both.
Getting middle management to support the Board’s goals about profits and shareholder value over the middle managers’ own petty empire building and competing w their peers over the next promotion is another alignment problem.
Just yesterday Star Talk had a 13-minute video that touches upon the subject of uploading consciousness. They were talking about making one immortal but the upload is the first step in teleportation.
TLDW: If consciousness involves quantum physics the possibility of uploading it is a firm no.
Well, yes and no. Quantum information can’t be copied exactly, because of the no-cloning theorem; but that doesn’t rule out a classical copy. A classical copy would not be exactly the same individual, but it could be practically indistinguishable by an external observer. That means there could be two of you, like the twin Rikers in Star Trek.
From wiki, here is a list of years with shifts in the scientific paradigm:
1543 1543 1687 1783 1826 1832 1859 1880 1901 1905 1919 1920 1952 1964 1965 1969 1973 1974
They average 25 years apart +/- 78 years. The list does not include genome sequencing or changes in approach such as the evidence based medicine movement. Yes, I’m blurring my terminology. We still haven’t figured out consciousness. Where there are new paradigms, there is potential for new applications. I say there’s more juice to squeeze in the orange.
The idea that consciousness has anything to do with quantum mechanics is abject nonsense – lt’s pseudo-scientific claptrap, not neuroscience. There are innumerable areas in science where our understanding is incomplete; the nature of consciousness and quantum physics are just two such areas, but there’s no more reason to believe that one is the foundation for the other than to believe that quantum phenomena explain why your cat likes to perch at the top of the cat tree.
That video struck me as akin to late-night college dorm room philosophizing after consuming a fair amount of potent weed. It wasn’t science, and especially not after one of the participants used the laughable term “quantum awareness”.
The facts are that we have a good basic understanding of the biophysics of brain function and that consciousness is evident not just in humans but in lower mammals. It manifests not at some magic threshold but as a continuous emergent property of scale.
As for uploading a human brain, it will undoubtedly be possible (and will be based only on sophisticated applications of classical physics) and has already been done for very simple life forms. It’s a daunting task for the human brain because of the tremendous scale involved, but there’s already serious research into some of the essential precursor technologies such as light-microscopy-based connectomics (LICONN).
Neil and Charles Liu are about as unqualified as wolfpup, for all we know maybe more so. They’re astronomers and science communicators. They have no background in neurology or the relevant fields of physics.
Well, if they didn’t, they wouldn’t be mimicking human behavior very well, now would they?
Ironically, though, computers now would be easier to explain to Abraham Lincoln than computers 20 years ago would have been. Because now, a computer can be a device that you talk to and it talks back. How it works under the hood would be a mystery to him, but then, it’s a mystery to most of us, too. But the interface would be familiar.
The main problem with consciousness is just getting people to agree on the same definition. Once you’ve got a definition, it’s easy to study.
I’d say CRSPR is a recent unanticipated innovation that has revolutionized things. AI is getting there - it is far different from the AI I studied 55 years ago, much of which is in our phones today.
A vaccine for cancer would qualify as a revolutionary innovation. It can’t be done with just a slightly better current vaccine. It is being worked on.
AGI would be another. Truly understanding gravity, ditto.
When people say this, they usually mean “we could control it like we do electromagnetism”. But unfortunately, we can already control it like we do electromagnetism. Fundamentally, we manipulate electric fields by moving charges around. And we can manipulate gravity by moving masses around, too.
The problem is just that gravity is so mindbogglingly weak that the masses we would need to move around for any practical effect would be stupendously large.
Well, I agree we manipulate gravity by moving masses, but that is just applying various vectors that may result in movement up if their force is greater than gravity from earth. I’m talking about truly understanding it, which I didn’t do an hour ago when I threw a basketball at a net.
Whether understanding gravity will give us anti-gravity and space drives is unknown, but it would be a big change.
