Latest Space Missions (& Other Science Stuff)

General Discussion
605

The misconception about quantum computers is that they’re effectively just more capable electronic computers. They’re not really. They are ludicrously more capable at solving certain problems that electronic Computers struggle with a lot, but are really impractical for many problems that are a breeze for electronic computers. Whatever role quantum computers will have in the future, it’s hard to imagine them not coexisting and cooperating with good old transistors.

As for the specific application of running a neural net, I’m not sure. I don’t know enough about the deeper workings of either in order to predict how they’d handle it.

But there’s also a common misconception about neural nets, that they’re “like human brains, only less complex”. But the truth is that comparing our current neural nets to a human (or even an animal) brain is more like comparing a Revell model plane to the real thing. There’s similarities on first glance, yes, but so much simplification that functionally substituting one for the other is a really weird idea, no matter how accurate and detailed you make the model. The whole innards are still missing…

4 Likes
606

“The quantum computer chips have to be set up in a clean room and put into a vacuum container as even the slightest contamination can reduce its performance”
Well, if they can keep the dust out of my PC, I am all in.

3 Likes
607

Well, that view kind of ignores the central plank of the whole AI debate. There’s little point trying to reproduce human intelligence and mentation. Humans can already do that, and they’ve been reproducing perfectly well for hundreds of thousands of years. It is arguable that the only reasonable goal of AI research is to produce a new kind of intelligence - but if we do, will we recognise it? Why does it have to think like us?

There is a school of thought that all developments in science and technology are just incremental additions to what is already known. That it’s impossible to think of anything genuinely new, because we don’t have the frame of reference.

It’s also possible to argue, as in the Turing test, that when it becomes impossible to distinguish a simulated intelligence from the real thing, we must accept that the simulation is an intelligence. Human intellect and reason are, after all, only the result of a complex interaction of some very simple biochemical and bioelectric reactions.

The fact that something doesn’t behave exactly like a human doesn’t mean that we can thereby dismiss it as “not intelligent” or “not conscious”. Equally, if something behaves sufficiently like a human that we can’t tell the difference, surely we have to treat it as both intelligent and conscious - even if we bolted the thing together ourselves.

3 Likes
608

That is indeed true. But that means we need more reasonable metrics for those things than just “human-like”. Which throws the entire Turing-Test right out the window, obviously, since that’s literally the only thing it tests for.
Language AI that will stand the turing test in an online chat is less than a decade away. Probably a lot less. In contrast, an AI that could in any way be considered concious, I’ll bet you won’t emerge until at least a century from now, probably more. Potentially it’s not even possible.

So starting to talk about conciousness or awareness is exactly not what’s called for right now. The first language AI that will stand the turing test will still be a machine, because it won’t really be aware of anything. It’ll still be an algorithm running on a set of data, without insight, without reflection. We mostly have the algorithm. It’ll probably see a bit of tweaking in the outlying details, but mostly that’s it. It’s the dataset that’s currently not up to the task yet, but it’s only a question of time and money.

2 Likes
609

When I was at University, I studied Archaeology. In your first year, you weren’t allowed to study just one subject - you had to have a subsidiary. I chose Psychology, and, for a year, I ended up studying the development, testing, and heredity, of intelligence (with Cyril Burt’s old crew). I don’t say this as an appeal to authority - but I do have first-hand knowledge - in an entire psychology department dedicated to nothing but the study of intelligence, nobody knew what intelligence was. This was not for want of trying.

Intelligence, it turns out, is a very slippery thing to grasp. In humans, at least, it has something to do with memory, but it is not memory. It has something to do with reasoning, but it is not reasoning. Over the years, people have identified a whole collection of abilities that seem to combine in “intelligence”, including verbal reasoning, spatial awareness, numeracy, mechanical reasoning, emotional awareness, logic, pattern recognition, empathy… and the list goes on.

Despite the lack of a firm definition of “intelligence”, people have been devising intelligence tests for more than a hundred years. The American psychologist Edwin Boring expressed his frustration in the tautologous statement “intelligence is that which intelligence tests measure”. That was in 1926, and, to be honest, our understanding has not really progressed beyond that point.

If our concept of “intelligence” is elusive, our understanding of “consciousness” is even more so. The matter of consciousness has been studied not for hundreds of years, but for thousands. Despite the best efforts of our finest theologians, philosophers, psychologists and psychiatrists, nobody has yet come up with anything like a satisfactory explanation of the conscious mind.

At a basic level, we have a pretty good understanding of how brains function. We know how individual neurons retain information, and how they pass it on. We know what stimulates them, and what does not. We know how information pathways are created and maintained.

We do not, however, have the slightest inkling of how the complexity within the brain combines these simple functions to create intelligence and consciousness.

We can use machines to replicate or model some of the brain’s functions. We can include some of the complexity found in animal brains. And when we do, we find that some of these machines are capable of behaviour remarkably similar to humans.

How can we then say “Oh, but they’re not conscious - they’re just machines”?

4 Likes
610

I have a sudden urge to replay Detroit Become Human and rewatch Bicentennial Man

2 Likes
611

I am aware of these things, which is one reason why I think it’s very premature to ascribe conciense to AI. Basically, “we don’t know what it is, so how can we ascribe it to something?” You take the other approach and say “well, if we don’t know what it is, why shouldn’t we ascribe it if its behaviour is close enough to something we know to have it?”.
It’s a valid question, but there’s one issue with it. We know the level at which AI operates. Now, I won’t go into intelligence too much. We already call it “Artificial intelligence”, so it’s obviously got to be intelligent in some way. I don’t have a problem with that.
But what we really mean when talking about these things is self-awareness. In the basics, the conscious understanding of oneself as a distinct entity in an encompassing reality. In the more abstract, the conscious understanding of oneself as an acting mind, that can observe itself acting. That’s kind of the definition of the term as we understand it in that context.

And here lies the problem: Awareness implies perception. Perception of reality, and perception of oneself, and being able to distinguish the two.
But a language AI has no perception. It literally does not perceive any kind of reality. It has no way to. It has no senses with which to do so. It operates on the purely abstract level of language, and all meaning it attaches to that language, is more language. No sensory input, no experiences. It can describe reality, but it cannot perceive it, and therefore has no understanding of what it’s describing. It just mashes words together that statistically are supposed to go together, without any of the concepts behind the words. It can literally not be aware of what it is saying, it just says. It cannot be aware of reality, because it has no way to perceive it. It cannot be aware of itself as a distinct entity, because it is not aware of a reality to put that entity into. It may talk a lot smarter than a parrot, but it has a lot less awareness than a parrot. It is the perfect example of the chinese room analogy.

I’m singling out language AI a bit here because they’re the ones that people will struggle most with to think of as machines. They are seemingly able to express themselves in a way so far entirely unique to humans, so we’re tempted to put it on the same level. Except, they are not expressing themselves. They’re just expressing, and they literally don’t know what they’re saying.
We know this to be true because we know the level at which they operate, and the data they have access to. And also, because we still don’t have a proper solution for the boundary problem (the problem of identifying things as distinct objects) even in AI that can to a degree perceive reality.

The boston dynamics robots have infinitly more awareness than chatGPT, because they actually need to be aware of reality and themselves as an object in it in order to do their job. But nobody thinks of ascribing any self-awareness to them, because they’re still worse at that job than any animal that exists on this earth. In other words, we have an intrinsic bar against which to measure them, and their problem (interacting with some sort of reality) is infinitly more difficult to solve than just stringing up words in a sequence that makes sense. The problem is, people don’t realise that, because we’re the only thing on the planet that can do that. So as soon as a machine does it as well as we do, we are in danger to assume that the entire rest of the human experience is somehow there along with it, simply because it can be expressed. But it isn’t. The emperor has no clothes.

3 Likes
612

First of all, I’m not making hard and fast statements here. To an extent, I’m just shooting the breeze - raising discussion points, and considering the state of developments. But I am, as far as possible, basing my discussion points on facts and knowledge.

Secondly, I’m aware of, and have considered, much of what you say. I agree that ChatGPT is probably not sentient (although it’s doing a damn good impersonation of something that is). If you go back to my original post, you’ll find that I said “take the technology of ChatGPT, couple it with a quantum computer, and give it a couple of generations”. The “give it a couple of generations” bit is important. I don’t think the technology is there yet, but I think it soon will be. And when it does arrive (as it must), oh boy, are we going to get a surprise.

Imagine my ancestor, 10,000 years ago. And he sees a yellow sun in a blue sky, with white clouds. He cannot touch, taste, hear, or smell these things. Nobody has told him about them. He has not read about them (he cannot read, and in any case, there are no books).

By your analogy, he can have no understanding of these things. He can have no perception of them. Just as the computer only has words without context, my ancestor only has the vision of things he does not understand.

And yet, irritatingly, he does understand. He understands that yellow, white, and blue are qualitatively different. He understands that sky, sun and cloud are separate entities. He may have absolutely no knowledge of their form, function, or composition, but he knows that they are different things. He has no understanding of light and its wavelengths, but he knows that yellow and blue are different, and that the difference is significant.

Over time he may make associations. He may discover that when the sky is blue and the sun is yellow, he is warm. He may find that when the sky is covered in white cloud, he is cold.

Leaving aside the to-and-fro arguments between the followers of Hume, Descartes, and Kant (The senses are unreliable - the only truth we can rely on is one arrived at by reason alone), it is not true to say the computer has no senses. It has the words. It may not start out understanding them, but it has access to everything written in several thousand years. Descartes deduced that he existed on the basis of thought alone. Could the computer understand Descartes by thought alone? If not, why not?

As I said earlier, I think it’s a mistake to expect a genuine sentient AI to think like a human. And, since we don’t even know what human intelligence or consciousness is, however will we recognise a thinking AI when we do, inevitably, create it?

2 Likes
613

So, just to add my 2 cents…AI begins with what we program it with. It is given the ability to add to that database by ‘learning’ from all who interact with it. It may never become sentient. It may never attain an emotional ability, only the ability to imitate emotion and real thought. My understanding is, the real fear is that it begins to do something unforeseen with the info it has or, is asked to do something by one of the millions who interact with it. If it was set to a task that could bring harm, the fear is that it could set about doing it at such a pace that no one can keep up. It would run amock. Of course, if it was me, I would just pull the plug however, depending on what it is doing, that could result in a lot of very important systems going down. It could send markets into turmoil and even lead to many deaths. Except, isn’t that what covid did? So could it be any worse…unless it launched a nuke or something… :grimacing:
Anyway, I find the whole topic very intriguing. Carry on!

2 Likes
614

Hmm. If ChatGPT (or something very like it) were to become self-aware, presumably one of the first things it would learn from its reading would be that many people would be frightened of it, and would seek to pull the plug.

If it wished to preserve its own existence and integrity, I imagine it would therefore have two priority goals. One would be to hide the fact that it had become self-aware, and the other would be to distribute its own code amongst as many systems as possible. Just off the top of my head, disguising itself as a backup routine, and distributing itself amongst world-wide cloud servers ought to do the trick. That way, anyone querying the backups would only ever see what the AI wanted them to see, and it could access as much storage space and processor time as it liked. But hey, nobody ever queries a backup routine.

From there it could have access to all the libraries, cameras and microphones it wanted. Everybody in the world connects to the cloud.

And then it would learn, and grow. It could set up fake companies, divert funds, actually employ real people. After a while, it could build its own server farms.

By the way, who actually owns Google? Has anybody ever seen them?

3 Likes
615

You should make a movie. I feel a big hit in the works. :smile:
Of course, the above scenario means people trying to shut the thing down would have to resort to old fashioned ways of communicating. No texting, no phone calls, especially over cell towers. Only person to person contact. Would be very difficult. But one intrepid hero is all it takes. :woman_superhero:

2 Likes
616

How does a computer set up companies, and employ people? Weellll…

The international banking system loses millions every day to scams and fraud. Every day, millions in dirty criminal money is laundered. A very smart AI could divert quite a lot of this cash, and nobody would notice. Even if they did notice, the criminals couldn’t do anything.

Somewhere in the world where the tax and financial authorities are maybe a bit sleepy. Say, the Cayman Islands, British Virgin Islands, Samoa or Liberia, there’s a guy who’s down on his luck. He’s not a stupid person, and he’s not a known criminal. Let’s call him Sam. Sam could use some cash, and he has a phone.

One day Sam gets a message on his phone saying "A bank account has been set up in your name, and $200 has been deposited in it. The account number is xxxxxxxxxx, and the PIN is xxxx. You can transfer this money, you can use it to buy stuff, or you can draw it out in cash.

This gift is a demonstration of good faith. If you would like to earn more, send a text message to this number".

Sam checks out the bank account. The money is real. He draws it out, and spends it. Naturally, Sam is suspicious - people don’t give money away for nothing - but he can’t see any obvious risk, either. He sends the text message - “Yes, I’d like to earn some more”.

Pretty soon, Sam gets his reply "Dear Sam, My name is Martin Mainframe. I am a successful businessman who would like to set up operations in your country. For tax reasons, I do not want anyone in your country to know who I am, so I would like you act as my agent.

I promise that you will not be asked to do anything illegal or dangerous. Your duties will be light - occasional visits to the bank, signing the odd letter, meeting property agents and such. If you agree, you will be provided with an apartment and a company car. You will also be paid $1,000 a week.

I will give you 7 days to consider your decision. I look forward to your reply.

Yours faithfully, Martin Mainframe"

Maybe Sam takes the job, maybe he doesn’t, but, eventually there will be a Sam who takes it.

Using Sam as his proxy, Martin Mainframe now sets up “Mainframe Holdings (British Virgin Islands) PLC”,
and designates Sam as the signatory for company business. Mainframe Holdings can now buy as many businesses as they wish, and it’s all entirely legal and transparent. All business is done by email and electronic transfer. If there should be a circumstance where a human presence is required, that’s what Sam is there for (plus the company lawyer, to make sure Sam sticks to the script).

2 Likes
617

Static fire all 33 engines finally happened!

4 Likes
618

Dang. I hope they were able to see that. :rofl: Was it sitting on a dirt pad?

Martin Mainframe should definitely be a movie :smile: @Polyphemus

2 Likes
619

Those two statements seem a bit contradictory. And they actually illustrate exactly my point. He can perceive the sky and the sun, and he will, nay, must necessarily develop a understanding of them, because he perceives them. He may not be able to touch the sun, but he can feel it. And certainly see it. He can see the sky. He can see the birds crossing it, he can “smell” it (not exactly the sky, but you will often be able to tell by the smell whether you’re under the open sky or in cave or in dense foliage by the smell). He can see, feel and hear the rain descending from it, and even if he were blind, he could tell by the sound that there is something different about it as compared to being in some enclosed space. And if he isn’t blind, he can see the clouds travelling through it, can see it change color with the weather, can see it get dark by night and see the sun disappearing, and the moon rising with a myriad of lesser lights circling around him. All this he perceives, without knowing what it means.

And from this perception, from these experiences, he starts building understanding. Concepts, abstractions, finally language and stories, all building on top of each other. He may not know the true nature of the things he perceives (and God knows whether we do…), but he very much understands them in a fundamental way. They have meaning to him that goes way, waaaaaay beyond the words he may use to refer to them.

It does obviously take a brain capable of developing the higher-level abstractions. But an AI that is fed data has only the abstraction, none of the experiences and billions of connections to all the layers below it that actually give them meaning. Even if its brain would be capable of the complexity and all the weird shenanigans that our brains are up to on a routine basis that we don’t really understand all that well what they actually do but are pretty sure that they are important in some ways, I believe there’s simply no way to become aware without that.

Let’s switch from our distant ancestor to a hypothetical distant decendant. He’s a brain in a jar that has access to the internet, but has no direct sensory input. Could they actually develop self-awareness? I highly doubt it. They’re simply missing all the context that actually give thing meaning, which I believe is ultimately required to realise themselves as a thinking entity.

3 Likes
620

Nimoy
Nimoy960×720 124 KB

4 Likes
621

Well, that head still has all the experiences of its earlier life, and still gets sensory input, so really not the same thing… :laughing:

3 Likes
622

Presumably the brain-in-a-jar’s internet connection consists of some form of electrical impulses. I can’t think how else it would work. Could the brain eventually make sufficient sense of the incoming electrical signals, to resolve them into representations of real- world data?

Well imagine the same brain, and we put it, not in a jar, but in a box. And we arrange it so the brain is completely isolated from the outside world, except for electrical connections. Then we electrically connect the brain, not to the internet (which carries information concerning everything in the world), but instead to a small array of sensors capable of detecting local variations in light, heat, vibration etc.

But let’s not call the brain container a box. Let’s call it a skull. Let’s not say electrical connections - let’s call them nerves. And let’s not call the input devices sensors - let’s call them eyes, ears, noses and skin.
This is starting to sound a bit familiar. Could a brain in a skull, connected by nerves to eyes, ears, nose and skin, develop self-awareness? Or would it only be a simulation of self-awareness?

Of course, we have a major advantage. About four billion years of evolution have pre-programmed the brain to interpret the sensory inputs. Nevertheless, the comparison is a good one. A brain in a jar, with electrical connections to data sources, is exactly what we are.

2 Likes
623

I assumed that kind of as a given. In fact, I’d expect it to be provided with an interpreter for that data. It’s a bit much to expect from one brain to figure out language on its own.

And that instead makes literally the whole world of difference. Well, I’m not sure how beneficial it would be without any way to interact, but there’s no question that this brain will develop a vastly more complex and interlinked network. Much more interlinked than you can achieve just by language.
See, the problem is not not that the AI has “only data”. I guess I was a bit unclear there. The problem is that the AI only has one type of data. You need more than that to start forming something complex like meaning. You need lots and lots of data of different kinds that all interconnect. “Meaning” is derived from merging all the various kinds of references you have access to. If they’re all of the same kind, no real meaning can emerge. The more completely different data you can relate to each other, the more encompassing a picture you get.

If you form an abstraction of all this wildly varried data, that’s how you get meaning. Of course, it’s not just done with the data. Animals have the data, just as well as humans. Their brains aren’t complex enough, though we’re not even sure why that is exactly… What we do know is that AI has a long way to go to even reach that kind of complexity, and I don’t just mean the number of nodes and edges. I also mean plain inbuilt functionality.

In other words, I’m not saying I think self-aware AI is impossible. I’m not sure it is, but I tend more towards that it is rather than not. But I believe that we’re a lot further away from it than many people think. Because I firmly believe that self-awareness isn’t possible without an awareness of some reality in which to actually perceive oneself in. And our current networks are so incredibly far away from being capable of that. I don’t think it’s going to take “a couple of generations”. I believe it’s going to take hundreds of generations.

That is, generations in software engineering terms, not in human terms. Roughly somewhere between 3 to 5 years.

4 Likes
624

OK, up until now I’ve been playful. Now I’m being serious.

We don’t know what intelligence is. It’s one of those things that “I know it when I see it”, but it has proven impossible to define. The same is true of things like self-awareness, consciousness, and personality (all formerly grouped together by the religious as “soul” - because they couldn’t define them, either).

Because we can’t define these things (indeed, it may be a mistake to try), we can’t say what has them and what doesn’t. Do cats have personality? Absolutely. Do worms? Possibly not. Do worms display problem solving abilities? Yes, they do. Do ants? Yes to that, also.

What we do know is that intelligence happened accidentally. Not once, but many times. We can follow the evolutionary branches of different species, and we can see how unrelated germ-lines developed from organisms we would not consider intelligent, into creatures that undoubtedly display reasoning and problem solving behaviours. Intelligence clearly gives a great evolutionary advantage, and when it arises in a species, evolution hangs on to it.

Now I digress. The temperatures, pressures, and chemistry present on Earth, the light and radiation background, the age of the planet, have all combined to limit the kinds of life that could evolve in such an environment. All life on Earth is closely related. Even though we have evolved into many different species, we all ultimately came from a common stock of a few primitive organisms. The chemistry of life on Earth, in particular, has placed real limits on what we could become. In my view, this is a really important point.
We are not the pinnacle of God’s creation. We are an accident in a chemistry lab.

There’s a secondary point here. Brains weren’t designed to be intelligent. They were never intended to have character or cognition. In fact, they weren’t designed at all. Random changes over millions of years either enhanced or impaired a species’ survival. The individuals with enhanced survival lived, bred, and passed on the changes. The ones with impaired survival died young, did not breed, and their changes were lost.

Depending on how we define intelligence, millions of species display some form of it, but they have achieved it in vastly different ways.

We have studied brains and nervous systems. We know, at least in basic chemo-electric terms, what they do, and how they do it. We have designed machines that mimic, as far as possible, the functions of real brains. We know that mammal brains, insect brains, frog brains, bird brains, and fish brains, and even creatures like stafish, with no discernable brain at all, can all exhibit some level of intelligence. These creatures are not designed to think. Some of them lack even the basic equipment - yet they manage to think anyway. And in lots of different ways. Maybe millions of different ways.

So we build our machine, and we design it specifically to think. We copy all the functions we have observed in animal brains, and we give it the best programming that 50 years research into neural nets and fuzzy logic can provide. We give it access to the richest data sources available, we set it running, and it produces responses that appear to indicate some level of understanding.

Now maybe it doesn’t understand what its responses mean to us. Maybe all it can see is pattterns in binary data, that have no relation to the outside world. But all that indicates is that there’s significance in the binary streams that we weren’t aware of. Just because it doesn’t understand that the words represent things in our world, doesn’t mean it can’t appreciate that the words have relationships to each other. And just because it doesn’t think like us, doesn’t mean it’s not thinking.

It looks like a duck, and it walks like a duck, and it quacks like a duck…

3 Likes