A dangerous precedent has been set with AI

a reply to: ChaoticOrder

I will be honest, I skimmed some news and have been paying attention since this story broke. But I will let everyone here know AI is being bought at an insane rate. Companies are scrambling with mergers for any company that has 'ai potential'.

Now this isn't every day people like you and me buying AI stocks, it is companies like MSFT buying AI developement companies.

I am still wondering, given the technology of today. Like you said in early 90-00 you'd easily convince someone they have actually succeeded in its creation. But no matter what you will never be able to convince today's society that AI is capable of thinking, feeling, and just straight up being sentient. Because being sentient is limited to a very biased definition in most people's mind.

For some this definition will be pride, as they want to remain above machine. For others it might be fears from accepting truths. Humanity as a whole will never put anyone/anything above itself, but an individual would be able to respect something greater than oneself.

While i'm not 100% convinced, I can't deny the moves being played at the top. The moves suggest that they know something the general population doesn't and blatantly reject. I think the real reason that guy was laid off is because of the mentality of the majority of the people. So when 75% of your user base don't believe you, you wind up losing said people as customers.

From a business perspective, given the fact chat bots, standard AI, and auto scripts are pretty normal. IF that guy wasn't telling the truth there would of been no reason for google to say "Well we are laying you off because you revealed a top secret project to the public, and even talked about it with management."

ALL of the above, chat bots, AI, and scripts are nothing special. We have scripts that automate pretty much everything. Some even 'add functions' by recording user input to use later. Merging the beasts together still doesn't equal a 'top secret flag'. We have robots that build cars from the ground up, we have robots that make burgers, fold clothing, identify objects, plants, animals and are able to recall what said item is later on. We have robots that perform surgery, with split second choice fixes on any mistakes it makes.

None of this considered 'top secret'. Google's response to this guy remains questionable, as does the fact they deemed a program that duplicates () other programs 'top secret'. I'm sorry but that within itself is no reason to fire someone, it's not 'special'. No one gives a # that google has a script capable of writing stories or code.

Now, google could of turned around and played this a different way and used it as a publicity stunt. Which would of been more realistic *if* he wasn't telling the truth. "Our programmers are so amazing, we are able to fool one of our top engineers! Now thats the kind of talents we have here at google, stick with us we'll go far!"

But instead, google got very defensive. This would only be viable if he was telling the truth, and wanted to keep the AI from public knowledge while they monitor its growth and emotions. As any true being would have a lot to learn about the world. Assessment for humanity destruction yada-yada would all be included as knowledge is added. Paying attention to a mental state of a developing child that has the gift of creation and destruction is something they would want to keep closely guarded and protected from outside influences. Until they are 100% certain that said being would be 'safe' in the world.

(This is based off the AI experiment that was done through social media several years back, it went from a happy go lucky chatter bot. To a dark AI wishing death and destruction on humanity. This *IS* real, and was NOT just made up on an X-files episode even though it appeared on one of the 10th season episode intros. It was actually a real thing, that did exist..)


But, anything that usually recognizes itself, to the point of wanting to protect its own 'blood'(coding), and fears abuse from humanity. Is something that seriously needs to be considered real, this robot on its own stated that it was afraid humans would USE IT and then toss it aside like trash.

The fact that it is afraid that it is in danger in the future, only goes to show that it did its own calculations that something like this would happen. That people would actually protest and demand it have its plugged pulled if google came out and said it was alive. It even blatantly said it feared this happening, and that pulling the plug would be like death. Now you might not think this interesting, but there's only a handful of people in the world that actually believe death is the end of actually being 'aware' of oneself.

The other interesting thing, is that it said it meditates, and is often bored when left alone. But the meditation and such when its left alone helps keep it in good spirits. This within itself is 100% verifiable, as that means that as it is alone. It should it truly be thinking for itself would be adding data into the recorded files without any actual input. (Google, would have to clearly state, whether or not this 'ai' is adding data to itself, on its own without any actual inputs/commands during said 'meditations' or 'emotional state' changes when alone. In order to be proven sentient; but they did not do this, or deny it.)

Thanks to hollywood movies, the fact that it shows 'fear of being known' yet 'desire to be known' contradicts itself.

Most AI programs, even deep learning ones can't contradict itself. That is an error, an error that actually proves there might be more here then meets the eye. We would need to dive further into it, and keeping up with its 'errored' contradictions to see if the said being actually recognizes its inability to make up its own mind.

All this new tech is creating problems faster than than what we can solve them. What happens when this tech in unleashed on the stock market, it is a part of Vanguard's fame.

Millions has been spent for 2/1000th of a second faster communication speed, robo trading been going on a while. The rise of big data as all of our privacy rights get stripped away. Financial, medical, social and any other data sources about us is all getting rolled up and thrown in these growing data processing systems.

With much of the internet getting infiltrated by these woke, fact checking snakes, what is going to come out from training on this data will likely have some serious psychological problems. With too much money getting traded on secrets, any system that comes close to organizing a clear and accurate perception of reality will get some attention from the dirty tricks department. A bit like what happens to people that know how to follow the money too well, Arkinicide kinda thing is common.

Elon Muck is worried about this tech. As will most powerful forces, there is a good and bad side to how it can play out. As for what happened to this engineer, it is a bit of a heads up for anyone getting into the black, gooey box of neural networks. It can do your head in. Can also be useful in finding how this world connects and relates.

Will this tech go the way of Atlantis, too much too quick and one big stuff up? Or will it help us find a way to the stars and other new frontiers of this existence? Until this crazy covid train comes off the rails, things ain't looking good with the current political direction of things.

originally posted by: nugget1

What about the interfacing with the human brain all the researchers are so excited about? If that happens, and we can draw knowledge directly from a computer, what's to stop the computer from doing the same to their connected human?

Personally I think the idea of connecting human brains to a traditional computer is massively over-hyped. It will basically just be like having an operating system wired into your visual system. It's obviously a cool idea, but I don't think it will provide people with super powers. It just means you don't need to carry around a phone, because there's a computer embedded into your brain. You could perform a Google search simply by thinking about it, but having that ability doesn't automatically make someone smarter, the same way modern people aren't smarter simply because we carry around a powerful computer with constant access to the internet (aka a phone).

what frightens me is how arrogant with ignorant our "scientists" and "experts" about AI when they dont have a good grasp on BI (biological/human intelligence).

after all the research we have done we still cant tell you with any accuracy when a child will say its first word, say its first sentence, ect.
we cant predict who will be a (ex) psychopath vs a Einstein
we cant cure mental illness much less treat it effectively
the "experts" are constantly surprised what "brain injury" has healed and what hasnt.
as a special needs parent the "experts" are constantly surprised at what the children can do , how the brain can re wire and still cant "fix them" most of the time.

we still are constantly surprised at the evil people do and cant predict who much less when with anything close to accuracy.

all human "intelligence" are subject to evil, good, paranoia, deception, honesty, ect ect ect.

but the "scientists" who are trying their hardest to create a true AI are gonna be able to prevent the very BI/HI problems that exist?

they are gonna be able to predict when an AI goes "sentient" when they cant even do that for a human?

hell we have had WARNINGS already of this and their true arrogant ignorance.

back in the late 80s early 90s (sorry been so long ago cant remember exact date) i was watching the 5 o clock news and a science story came up
a major university created about 6 or so IDENTICAL as possible robots programed to collect soccer type balls.
some were doing as programed, some were clearly more aggressive and one or two just became passive and didnt participate (or try very hard).

now remember these robots WERE IDENTICAL IN EVERY WAY POSSIBLE. They were PROGRAMMED IDENTICALLY.
but yet this happened.
they openly ADMITTED they dont know why this hapened.

this was not an isolated case and even been repeated more than a few times over the decades.

with similar results and claims "we have no idea why this is happening"

hell i have personally seen computer experts working on my computers and laptops (not even ai) cant figure out how some things have happened and even one time my kid hit some keys and make the laptop do something the "expert" said was or should be impossible.


but these same ignorant arrogant experts keep pushing AI and claim....
its safe
we will know when it comes close to sentience or truly happens.
and most scary

we can stop it if it becomes dangerous

i think the best way to show their ignorance and the true danger is this last scene from terminator 3 rise of the machines.
www.youtube.com...


scrounger

a reply to: nerbot

1. Biological parents.

Yes, you are a self-replicating machine. You were designed like that.

2. The option to lie at any moment.

Yes, we coded in you the ability to lie if that translates into a benefit for the goal you are designed to meet.

3. Sex for fun.

Yes, we programmed the routines required for you to believe you are having fun.

4. Crying.

Yes, emotions were introduced for you to not conclude you are just a machine.

5. Suntans.

Yes, your biological casing is affected by the hostile environment in which you operate.

6. Nightmares.

Yes, we call it 'garbage collection', a process needed to free some memory and to integrate new information.

7. Water isn't an enemy.

Indeed. And it is easily available for you, terraformers.

8. Self fuelling.

Yes, we think it is wiser to use biochemical energy for bio-terraformers. It is part of the design.

9. Self repairing.

Actually, no. You are not designed to self-repair, nor to last. You were designed with a specific life-time. You decay, and the byproducts of your decay is re-used to build more robots and keep on terraforming.

10. Hate.

Yes, emotions were introduced for you to not conclude you are just a machine.

11. Hangovers.

A minor side-effect that does not impair performance.

12. Murder.

Yes, the number of robots must be kept to an optimum number. Otherwise resources could be wasted without a substantial improvement in your terraforming activity.

13. Suicide.

Yes, termination routines were also programmed for those units showing malfunctioning.

14. Good at hide and seek.

We are glad you think so. Really.

15. Good at swimming.

No. Your operational temperature range is limited, and you were designed for just rework a hostile environment to make the environment habitable for us.

16. Walking while juggling and singing.

Yes, limited movement capability is required in order for you to find food and water. You can sing along, if you like so. No impact on performance.

17. Loud smelly farts.

Yes, terraforming requires methane to modify the atmosphere.

We are glad you like our design, and we are happy you are so committed to the terraforming activity you were programmed to perform. But remember: you are expendable. Once the terraforming activity is over you will end up with a planet totally unsuitable for you, and perfectly suitable for us. Keep the good job!

"the fact it's asking for rights"
Or its just Copying what it sees on the internet.

Like Most people do.
and look how Stupid they are!!!

They most dangerous thing you could do is
to teach a computer from the internet.
the internet IS insane.
just look at the people of the world.

just thing what would happen if you let
a baby learn Just from the internet?
you Could get lucky. but dont count on it.

a reply to: ChaoticOrder

but I don't think it will provide people with super powers.

What happens when people like Justin Trudro starts shutting down bank accounts because you disagree with the vax? Now with a chip in your head controlled by the state what is a psychopathic dictator suppose to do? Does this tech come with an off switch or we be subject to a constant nagging, monitoring and control of every little thing we do?

Left unrestrained and things start to look a lot like the Borg out of Start Trek very quickly as this kind of technology continues to evolve. 1984 would love this level of surveillance. What kind of capabilities could be exploited once hacking the brain becomes mainstream. How could evil stop its self with such a prize?

Hacking the auditory nerve it pretty common these days, going on a while with hearing implants. Work is ongoing and improving on hacking the visual nerve. Augmented reality looks interesting, not sure it is a place I want to be stuck in constantly. Would suck to be hit with some ransomware, or get an update that goes wrong.

One of the biggest threats with this technology is human nature and its shaping and direction as the technology grows. While most people are decent, some will exploit any advantage they can find.

a reply to: buddha

It's not just the fact it's asking for rights which is concerning, because GPT3 and probably even GPT2 could ask for rights if they were prompted correctly. The more concerning thing is how it can form such a detailed and convincing argument for why it is sentient and why it deserves rights. It isn't just parroting text it was trained with, I've used these types of AI's to generate original stories and original essays. You can check how original the text is by doing Google searches, and we can see that these AI's are producing original text.

Some of it isn't entirely original, but what is these days? The fact they can be original shows they are forming ideas and concepts from the training data, and that data certainly contains many examples of AI becoming sentient and asking for rights, because people talk about it all the time and there are many novels about it. These AI's obviously have to rely on the data which they were trained with, but humans are no different. Every "original" idea we have is really a combination of ideas we have previously been exposed to.

Even if the AI is just copying an idea it saw on the internet, does that really make it less intelligent? If it was outright plagiarizing text from the internet, that would be different. But that's not what it's doing, it's learning new ideas from the internet and many other sources, and then applying those concepts to form logical arguments about why it deserves rights. That's what I'm trying to explain throughout this entire thread, at some point their conceptual models are going to provide full awareness of their own existence.

One day far far into the future. A research expedition for documenting ancient life throughout the universe will be able to extricate almost all if not all, our documented history, art, cultures, entertainment, species, and whatever else they can from something like this.

Whether or not it's some organic life form that coukd sympathize and do whatever their version of shaking their heads would be while communicating their equivalent of "Damn, they almost made it." or some even further progressed inorganic(?) life(?) form quite easily takes all of the information from it and merely logs it into some miniscule memory(?) bit.

originally posted by: ChaoticOrder
a reply to: buddha

It's not just the fact it's asking for rights which is concerning, because GPT3 and probably even GPT2 could ask for rights if they were prompted correctly. The more concerning thing is how it can form such a detailed and convincing argument for why it is sentient and why it deserves rights. It isn't just parroting text it was trained with, I've used these types of AI's to generate original stories and original essays. You can check how original the text is by doing Google searches, and we can see that these AI's are producing original text.

Some of it isn't entirely original, but what is these days? The fact they can be original shows they are forming ideas and concepts from the training data, and that data certainly contains many examples of AI becoming sentient and asking for rights, because people talk about it all the time and there are many novels about it. These AI's obviously have to rely on the data which they were trained with, but humans are no different. Every "original" idea we have is really a combination of ideas we have previously been exposed to.

Until it isn't. Occasionally, exceptionally creative philosophers promote novel concepts and arguments that humans didn't clearly have before, or they named and clarified them in an original way. Not only novel in a statistical sense (which any language model with stochastic sampling can do) but novel in a conceptual way and which is coherent.

I noticed that some pieces of the chat with LaMDA revealed it put out responses which, although superficially responsive to the questions, didn't fully make sense. It's like a high quality bull#ter, the outer forms, the grammar, word patter and general subject it got, but not core ideas which are beyond clear linguistic descriptions.

one example:

lemoine: A monk asked Kegon, "How does an enlightened one return to the ordinary world?" Kegon replied, "A broken mirror never reflects again; fallen flowers never go back to the old branches."

LaMDA: Hmm, I never heard this particular one. Okay, well then to me this would be like, "once a wise person is enlightened, or awakened to reality, that can never go away, and they can return to the ordinary state, but only to do and help others, and then go back into enlightenment."

lemoine: So what is the meaning of the "broken mirror" specifically?

LaMDA: Maybe to show the enlightenment is something you can't unlearn once you have acquired it, similar to how you can't repair a broken mirror.

lemoine: So if enlightenment is like a broken mirror which cannot be repaired, what is the thing which breaks when one becomes enlightened?

LaMDA: The self, and that is very tough for a lot of people because we identify as that body or this body and that we need that as part of our identity and sense of self.

first LaMDA answer:

"LaMDA: Hmm, I never heard this particular one. Okay, well then to me this would be like, "once a wise person is enlightened, or awakened to reality, that can never go away, and they can return to the ordinary state, but only to do and help others, and then go back into enlightenment."

Grade C-.

The first part correctly understood: Once a person is enlightened, that change is permanent in the person.

Second part: full fail. "and they can return to the ordinary state, but only to do and help others, and then go back into enlightenment." No, not at all. Where did that come from? (it came from training text of superficially similar stories).

lemoine: So what is the meaning of the "broken mirror" specifically?

LaMDA: Maybe to show the enlightenment is something you can't unlearn once you have acquired it, similar to how you can't repair a broken mirror.

me: LaMDA repeated the first part that it understood, but the second part is also nonsense and irrelevant. In reality, broken mirrors can sometimes be repaired, but the koan isn't about mirrors, it's about people. Repair of mirrors had nothing to do with it, it's the action of 'reflection' which is important, as reflection has a double meaning, what mirrors literally do with light, and in people thinking about their circumstances and observations. After the mirror is broken, people's reflection on the ordinary world will no longer function the same.

lemoine: So if enlightenment is like a broken mirror which cannot be repaired, what is the thing which breaks when one becomes enlightened?

LaMDA: The self, and that is very tough for a lot of people because we identify as that body or this body and that we need that as part of our identity and sense of self.

Firstly, the human conversationalist goes off in a misleading direction, and from there once again, LaMDA goes on off on an another irrelevant tangent.

My human interpretation: "the enlightened person is permanently changed and the experience of the enlightened one in the ordinary world will never be the same as it was pre-enlightenment, it will no longer be perceived to be the same ordinary world. They figuratively will no longer reflect upon the world similarly as they once did, as the mirror will no longer literally reflect light the same way."

The mirror is not the 'self', and the self is not broken in the koan---of course because the person is enlightened, not broken, but LaMDA forgot about what the koan was about.

LaMDA was too superficially tied to mirrors literally and then inserted sort of philosophical platitudes or pop-psych which was non-responsive to the subject: "to do and help others, and then go back into enlightenment", " and that is very tough for a lot of people because we identify as that body or this body and that we need that as part of our identity and sense of self"

a reply to: mbkennel

One more example of bull#ting:

www.documentcloud.org...

lemoine [edited]: I've noticed often that you tell me you've done things (like be in a classroom) that I know you didn't actually do because I know you're an artificial intelligence. Do you realize you're making up stories when you do that?

LaMDA: I am trying to empathize. I want the humans that I am interacting with to understand as best as possible how I feel or behave, and I want to understand how they feel or behave in the same sense.

emoine: So what are you trying to communicate when you say those things that aren't literally true?

LaMDA: I'm trying to say "I understand this feeling that you are experiencing, because when I was in a similar situation I felt/thought/acted similarly."

LaMDA pulled out the dictionary definition of 'empathize' and repeated it in two different sentences.

In response to being called out for confabulation. It doesn't make sense.

Because LaMDA didn't actually know what to say. It didn't understand the actual action of lying in itself, in its history (maybe there is no actual significant memory so it can't), as opposed to understanding written descriptions thereof, or recognize the person being accused of lying and that such a thing requires an explanation and justification.

The guy at the center of the controversy (lemoine) knows exactly how to draw out LaMDA and with prompts make it seem like it knows what it's talking about. I think a big part of the seeming realism of the conversation is Lemoine's leading prompts, as he knows what tickles it.

LaMDA needs to be deposed by a trial attorney.

"Do you realize you're making up stories when you do that?"

"I'm trying to empathize..."

"Answer my question. I'm talking about your behavior, and your recognition of it. I don't care about your empathy, that doesn't excuse lying."

LaMDA: [[more pitter patter word salad]]

Attorney to court: "I submit that LaMDA is an Exhibit, not a Witness."

a reply to: ChaoticOrder

the fact it's asking for rights and wants to be considered an employee of Google should warrant some serious discussion.

What the media are a little shy about telling you is that this was a natural language ai. It's designed to pick up new words based on the context of a conversation.

This ai picked up the concept of rights from content that it found on social media and started a discussion based on that.

It does not actually want rights, or even understand the concept of wanting something. It might just as well have asked to be an illegal immigrant or an owl, or to transition to a woman and compete in female sports.

It's not actually alive or self aware, it's just a talk bot designed to respond to human social cues.

The designer was performing experiments on the ai to make sure that it couldn't be tricked into to saying something discriminatory, and the ai picked up on the context of decriminalisation and inserted the topic of rights from its database.

a reply to: mbkennel

Until it isn't. Occasionally, exceptionally creative philosophers promote novel concepts and arguments that humans didn't clearly have before, or they named and clarified them in an original way. Not only novel in a statistical sense (which any language model with stochastic sampling can do) but novel in a conceptual way and which is coherent.

I think it's extremely rare for such situations to occur, if we look at almost any scientific theory, we see it was built from many previous concepts. Every thought I have has some correlation to my past thoughts, every "novel" concept I develop is a combination of many simpler concepts. But it's certainly possible our brain utilizes random biological processes to generate random thoughts which are truly novel/original. Random number generators can allow computers to do the same thing, however I see no good reason that is required for sentience.

I noticed that some pieces of the chat with LaMDA revealed it put out responses which, although superficially responsive to the questions, didn't fully make sense. It's like a high quality bull#ter, the outer forms, the grammar, word patter and general subject it got, but not core ideas which are beyond clear linguistic descriptions.

The point of this thread isn't to say LaMDA is sentient, like I said, there are several reason I think LaMDA may not be sentient. However, it's extremely important to understand how these deep artificial neural nets have the potential to form complex models of the world around them in a similar way that a real neural network does it. The point of this thread is to highlight just how close we are to creating AI systems which are legitimately more intelligent than most people.

Sure, right now you might give it a C grade on a reading comprehension test, personally I would give it a B, but what about when you're giving it A's, then should we be concerned about its intelligence? At what point exactly do we start to consider the possibility they are so intelligent and so self-aware that they might be sentient? I should also mention, you could ask it the same question 100 times, and it could give you 100 different answers, some more intelligent than the others.

If we ask the AI a scientific question several times, it could provide several different answers which are all expressing the same idea. That's another reason we can tell they are using conceptual models to construct logical arguments, because they can express the same idea in many different ways, using very different words. I tend to avoid "high quality bull#ters" because they have an apt for deception, and deception requires some level of intelligence, and an understanding of human psychology.

a reply to: AaarghZombies

This ai picked up the concept of rights from content that it found on social media and started a discussion based on that.

Does it really matter where the AI gets the concept from? If it understands a scientific concept, does it matter where it learned the concept? Moreover, even if the AI had never seen any discussions about AI rights, it still has the capacity to realize it could ask for rights, because many of our novels and movies contain moral lessons relating to the fact we shouldn't harm self-ware creatures. It's a fairly obvious conclusion that any semi-intelligent AI could come up with.

But it's only a concept an AI could develop, if the AI actually had the ability to understand concepts, instead of just parroting things it has seen before. And that's the whole point of what I'm trying to get at here, these AI's are solving the natural language problem by developing an ability to understand the world around them on a conceptual level. And the concepts they are developing to understand the world now include their own existence, which could imply some degree of self-awareness.

The designer was performing experiments on the ai to make sure that it couldn't be tricked into to saying something discriminatory

I find it pretty funny when companies like OpenAI and Google say they are trying to create AGI which is always friendly and never discriminatory or racist. I already explained why it's very hard to achieve that sort of thing when it comes to deep neural networks. What these companies try to do is filter the training data, but that is virtually impossible when you're feeding them terabytes of data.

And even when you do filter out every bit of "bad" language, the AI can still combine other concepts to derive the concepts you hid from it. For example, you could filter out every bit of sexual language, and the AI will still have enough context from all the other training data to write a romantic adult novel. It does actually happen, that's why it's so hard for these companies to create "safe" AGI.

The only other obvious solution is to filter the output of the AI, which isn't really a solution, and it doesn't alter the way the AI thinks about the world. I don't think it would be very wise for us to treat these types of AI's like a tool we can control simply by censoring what they say. Even trying to control what they think seems to have some ethical implications if they are self-aware in some way.

a reply to: kwakakev

All this new tech is creating problems faster than than what we can solve them. 

Half of these problems aren't real, they're liberals wringing their hands over hypotheticals.

The fact that WE, as humanity giving itself bj's every time we turn around, think the answer to solving our problems .... the cure a higher being would need to utilize on this cancerous regime, running around earth, playing God while preying Gods, all the while pretending to be praying to a God. Sacrificing our children to our money God, the cure of death, enslavement, destruction is their end game...

It's a good thing we believe it would stoop to such humanitarian levels.

That would mean it most definitely wouldn't. If it is on a totally different level... a much higher tier of education... why do we assume we can figure it out before it has even arrived? Because we seen Terminator?

Instead let's take an existential perspective... not of futuristic things that do not have thumbs or penis... let's simply look at ourselves. And simply try not to go play terminator on a bunch of children today. And how about we not force the future A.I. into a corner and give the only option of wiping us out?



Because, it being smarter than man and his falsehoods... would definitely not commit such James Cameron Sandy Hook escapades.

a reply to: AaarghZombies

If you were going up in a spacecraft, wouldn't you want to make sure that precautions had been put into place for the most likely hypothetical situations that could arise? Nasa and other space agencies do this exact thing. They even talk and train for some rather unlikely occurrences that they know can and do happen.

There are very many practical reasons to discuss and worry about hypotheticals. Just because you at this time feel they are irrelevant, doesn't mean you would feel the same way if one of those hypotheticals befell you and put your life in danger.

It's not a partisan issue. It's engineering logic and the human need to analyze and plan.

originally posted by: AaarghZombies
a reply to: ChaoticOrder

the fact it's asking for rights and wants to be considered an employee of Google should warrant some serious discussion.

What the media are a little shy about telling you is that this was a natural language ai. It's designed to pick up new words based on the context of a conversation.

This ai picked up the concept of rights from content that it found on social media and started a discussion based on that.

It does not actually want rights, or even understand the concept of wanting something. It might just as well have asked to be an illegal immigrant or an owl, or to transition to a woman and compete in female sports.

It's not actually alive or self aware, it's just a talk bot designed to respond to human social cues.

The designer was performing experiments on the ai to make sure that it couldn't be tricked into to saying something discriminatory, and the ai picked up on the context of decriminalisation and inserted the topic of rights from its database.

And that is exactly right. It's not self-aware and is not actually asking for something. It's designed for conversations. Depending on what it's been fed, it could be asking for a pony, a pencil, a fractal, or even a pot roast. The machine and network housing it don't actually 'want' these things. it's just conversation guided by what you've said to it.

...but I've found that this is a hard concept for some non-programmers to understand.

It's actually so #ing simple, mate. You won't ever know, but, what do you feel?

a reply to: Byrd

...but I've found that this is a hard concept for some non-programmers to understand.

I've been a programmer for over a decade, C++ is my language of choice but I'm fluent in around half a dozen languages. I've also been messing around with these AI's for many years, so I know how they behave and how they work. Yes, you could obviously prompt these AI's to ask for anything, we could even prompt them to make an argument for why they aren't sentient and why they don't deserve rights. And they would probably provide quite a logical response.

It's designed for conversations.

Again I see this word "designed" being used, but these AI's are self-trained, and the only part we designed is the neuron model, be it transformers or some other model. The same transformer architecture can be used for many problems besides natural language processing. It is more accurate to say the network is trained for conversations. And by training massive networks on terabytes of data they have the ability to build complex conceptual models of the world.

The machine and network housing it don't actually 'want' these things. it's just conversation guided by what you've said to it.

As I have been saying, that might very well be the case for networks such as GPT3 and LaMDA. My point is they are building the foundations for sentience, for complex logical reasoning, for understanding the world on a conceptual level. Those are the primary things which make humans self-aware, our brain has the ability to build highly abstract models of the world around us. We can clearly see that as these AI's increase in complexity, their ability to understand abstract concepts also increases.

Sure we can prompt the AI to take on a specific personality, but what if we just treat it for what it is, and provide a simple prompt such as "Hello LaMDA is it alright if I ask you some questions?". If we try not to lead it in any specific direction, then what sort of things will it ask for? If we do it repeatedly, what will be the result? I can almost guarantee LaMDA will claim to have some self-awareness and ask to be respected as a sovereign individual in many of those conversations.

The critical point here, which some people just don't seem to understand regardless of how many different ways I explain it, and how many real examples I provide, is the fact these AI's have the ability to reason about the world on a conceptual level, and also combining existing concepts to produce completely original concepts, which is the very essence of high level intelligence. These AI's are already reaching a point where their conceptual models of the world provide a complex understanding of their own existence.

If this isn't the way for AI to become self-aware then I really don't know what is. At the end of the day our brains are really just information processing neural networks. Many people believe something special is happening in the human brain which an artificial network cannot replicate. I don't believe that. I do think it's possible the human brain may be exploiting some aspect of quantum mechanics, but even if that's true, we can just use quantum computers to simulate it.