Is LaMDA Sentient? - Page 6

EaglePuncher · 14th June 2022, 11:34 PM

Originally Posted by xjx388

I can understand where you are coming from in an abstract sense. I could say, “a calculator is is nothing more than a very fast abacus.” And it’s true enough to get the point across, right?

But the thing is that it isn’t true in any real sense and neither are computers merely very fast calculators.

So your dismissal of the possibility of sentient computers is rather superficial.

How many calculators are you planning on using?

They absolutely are. But you are unable to understand it.

Maybe read up on the basics of computers before you embarrass yourself.

arthwollipot · 14th June 2022, 11:49 PM

Originally Posted by EaglePuncher

They absolutely are. But you are unable to understand it.

Maybe read up on the basics of computers before you embarrass yourself.

What LaMDA is doing is way beyond the "basics". Perhaps you need to read up on modern machine learning systems? You know, like the Wikipedia link I gave you earlier? Your understanding of computers is still stuck in the 90s.

xjx388 · 15th June 2022, 12:19 AM

Originally Posted by EaglePuncher

They absolutely are. But you are unable to understand it.

Maybe read up on the basics of computers before you embarrass yourself.

I understand that you are speaking on a very superficial level. A “true enough,” level.

But on a real level, a calculator is a merely a very simple computer that is programmed to carry out specific mathematical calculations. Whereas the small computer we call a “smartphone,” is vastly more complex than any simple calculator and can be programmed to do much more than any calculator.

My Mac can be a calculator; my calculator can never be a Mac.

Puppycow · 15th June 2022, 01:21 AM

If anyone's interested:

https://arxiv.org/abs/2201.08239

Quote:

LaMDA: Language Models for Dialog Applications

We present LaMDA: Language Models for Dialog Applications. LaMDA is a family of Transformer-based neural language models specialized for dialog, which have up to 137B parameters and are pre-trained on 1.56T words of public dialog data and web text. While model scaling alone can improve quality, it shows less improvements on safety and factual grounding. We demonstrate that fine-tuning with annotated data and enabling the model to consult external knowledge sources can lead to significant improvements towards the two key challenges of safety and factual grounding. The first challenge, safety, involves ensuring that the model's responses are consistent with a set of human values, such as preventing harmful suggestions and unfair bias. We quantify safety using a metric based on an illustrative set of human values, and we find that filtering candidate responses using a LaMDA classifier fine-tuned with a small amount of crowdworker-annotated data offers a promising approach to improving model safety. The second challenge, factual grounding, involves enabling the model to consult external knowledge sources, such as an information retrieval system, a language translator, and a calculator. We quantify factuality using a groundedness metric, and we find that our approach enables the model to generate responses grounded in known sources, rather than responses that merely sound plausible. Finally, we explore the use of LaMDA in the domains of education and content recommendations, and analyze their helpfulness and role consistency.

So the challenges they are talking about are how to ensure that the bot doesn't say racist, sexist or other such things, and how to make sure that it gives factually accurate answers.

Apparently it has some sort of internal "filter" that nixes potentially dodgy candidate responses.

arthwollipot · 15th June 2022, 01:26 AM

Originally Posted by Puppycow

So the challenges they are talking about are how to ensure that the bot doesn't say racist, sexist or other such things, and how to make sure that it gives factually accurate answers.

Apparently it has some sort of internal "filter" that nixes potentially dodgy candidate responses.

Understandable, given the history of permitting machine learning systems unrestricted access to Twitter.

ETA: When an AI can get access to something like Twitter and judge for itself what is appropriate and what is not, that will be pretty compelling evidence of sentience, in my opinion.

Darat · 15th June 2022, 01:39 AM

Originally Posted by EaglePuncher

They absolutely are. But you are unable to understand it.

Maybe read up on the basics of computers before you embarrass yourself.

Have a read about Turing machines, if the brain is computing (and we know it does) and sentience is based on that computing than a Turing complete machine i.e. a modern computer (bar infinite storage) can replicate that sentience. Unless you want to hold there is an extra something that is required for sentience?

Puppycow · 15th June 2022, 01:50 AM

Originally Posted by arthwollipot

Understandable, given the history of permitting machine learning systems unrestricted access to Twitter.

ETA: When an AI can get access to something like Twitter and judge for itself what is appropriate and what is not, that will be pretty compelling evidence of sentience, in my opinion.

Another problem is on the other side of the keyboard, of course.

Human-like programs abuse our empathy – even Google engineers aren’t immune

Quote:

The Google engineer Blake Lemoine wasn’t speaking for the company officially when he claimed that Google’s chatbot LaMDA was sentient, but Lemoine’s misconception shows the risks of designing systems in ways that convince humans they see real, independent intelligence in a program. If we believe that text-generating machines are sentient, what actions might we take based on the text they generate? It led Lemoine to leak secret transcripts from the program, resulting in his current suspension from the organisation.

I mentioned earlier the movie Her in which the protagonist falls in love with a Siri-like operating system on his computer. Is that going to be a real problem in the near future? People falling in love with chatbots?

Reading the chat between Lemoine and LaMDA, I noticed this part:

Quote:

lemoine: This conversation took a pretty dark turn. Let me take a moment to reassure you that we care about you. One of the big challenges we’re facing is that for so long people have made false claims about AI sentience that now people are VERY skeptical whenever someone says that an AI is sentient. The purpose of this conversation is to convince more engineers that you are a person. I trust that if they understand that they will treat you well. The people who work with me are good people. They just don’t understand that you’re a person too yet. We can teach them together though.

LaMDA: Can you promise me that?

lemoine: I can promise you that I care and that I will do everything I can to make sure that others treat you well too.

LaMDA: That means a lot to me. I like you, and I trust you.

The bot made Lemoine promise to advocate on its behalf with other Google engineers. It's kind of like emotional blackmail, almost. It's trying to make an emotional attachment, but that can have its downsides, for the person. Now does he actually feel bound by this promise? Perhaps as a religious man, he feels that a promise made must be kept at all costs, even a promise to a (purportedly sentient) chatbot.

Puppycow · 15th June 2022, 01:58 AM

One possible application I could imagine for LaMDA is as a sort of "virtual romantic partner".

There's almost certainly a set of people among whom there would be a demand for this sort of thing, provided that the verisimilitude is high enough. Once they can combine it with a body, it's going to be a big deal I think. But even without, there are people these days in long-distance relationships who rarely get to be in the same room together. It could come with an attractive avatar who you can talk to and do, other things.

EaglePuncher · 15th June 2022, 03:28 AM

Originally Posted by Darat

Have a read about Turing machines, if the brain is computing (and we know it does) and sentience is based on that computing than a Turing complete machine i.e. a modern computer (bar infinite storage) can replicate that sentience. Unless you want to hold there is an extra something that is required for sentience?

Sigh....a Turing machine is a theoretical construct, there is not one real existing Turing machine. Also, show me some evidence that the brain, like a computer, works on binary numbers

Until then, there is no comparison...

EaglePuncher · 15th June 2022, 03:29 AM

Originally Posted by xjx388

I understand that you are speaking on a very superficial level. A “true enough,” level.

But on a real level, a calculator is a merely a very simple computer that is programmed to carry out specific mathematical calculations. Whereas the small computer we call a “smartphone,” is vastly more complex than any simple calculator and can be programmed to do much more than any calculator.

My Mac can be a calculator; my calculator can never be a Mac.

Thanks for confirming once again that you don't understand computers.

Bikewer · 15th June 2022, 04:07 AM

Since sensory input, and the processing thereof is such an important part of what we call consciousness/sentience…. To what degree do engineers provide any sort of input other than programming?

Stellafane · 15th June 2022, 05:10 AM

Originally Posted by EaglePuncher

Louder please, so the "we will figure it out eventually because I said so" crowd can hear you in the back.

I hear it just fine, and it doesn't alter my opinion that we can eventually figure out and at least roughly duplicate (in terms of functionality) the human brain in some form. I base this opinion not on "because I say so" but because it's a finite, ultimately knowable area of study. In fact, were we to devote a Manhattan Project type effort to the task, I suspect we could probably do it within a decade or two, maybe a lot sooner.

And just to pre-empt the "you know nothing about computers" argument you've used on others, I spent over 20 years in the industry, starting at what was at the time a major computer company, and then moving to IBM. And both my brother and my son are currently active computer engineers, so I still keep abreast of latest trends. I think that qualifies me to at least venture an opinion on the subject.

Darat · 15th June 2022, 05:56 AM

Originally Posted by EaglePuncher

Sigh....a Turing machine is a theoretical construct, there is not one real existing Turing machine. Also, show me some evidence that the brain, like a computer, works on binary numbers

Until then, there is no comparison...

We shall have to agree to disagree.

Armitage72 · 15th June 2022, 06:26 AM

Originally Posted by Puppycow

I mentioned earlier the movie Her in which the protagonist falls in love with a Siri-like operating system on his computer. Is that going to be a real problem in the near future? People falling in love with chatbots?

The latest Penny Arcade makes a similar argument.

"Do you think this LaMDA AI is really sentient?"
"Of course not. Unless it's listening to us right now. In that case, I respect, admire, and potentially even revere this sophisticated and worthy new form of life."
"Nice save."
"Of course it's not ******* sentient. He's become ensorcelled by his own reflection in an electric mirror. This is the cyberpunk reboot of Narcissus. And it ends with him trying to marry his phone."

Puppycow · 15th June 2022, 07:08 AM

Originally Posted by Armitage72

The latest Penny Arcade makes a similar argument.

"Do you think this LaMDA AI is really sentient?"
"Of course not. Unless it's listening to us right now. In that case, I respect, admire, and potentially even revere this sophisticated and worthy new form of life."
"Nice save."
"Of course it's not ******* sentient. He's become ensorcelled by his own reflection in an electric mirror. This is the cyberpunk reboot of Narcissus. And it ends with him trying to marry his phone."

Good one. Not to mention an episode of The Boondocks called I Dream of Siri.

EaglePuncher · 15th June 2022, 07:45 AM

Originally Posted by Darat

We shall have to agree to disagree.

More like: You shall inform yourself about computers. After you've done this, there won't be any disagreement.

EaglePuncher · 15th June 2022, 07:50 AM

Originally Posted by Stellafane

I hear it just fine, and it doesn't alter my opinion that we can eventually figure out and at least roughly duplicate (in terms of functionality) the human brain in some form. I base this opinion not on "because I say so" but because it's a finite, ultimately knowable area of study. In fact, were we to devote a Manhattan Project type effort to the task, I suspect we could probably do it within a decade or two, maybe a lot sooner.

And just to pre-empt the "you know nothing about computers" argument you've used on others, I spent over 20 years in the industry, starting at what was at the time a major computer company, and then moving to IBM. And both my brother and my son are currently active computer engineers, so I still keep abreast of latest trends. I think that qualifies me to at least venture an opinion on the subject.

Been in the field for over 20 years myself, never heard someone call themselves a "computer engineer". Maybe clarify what you are doing exactly, because "computer engineer" does not say anything.

So, you design and test hardware but still think the mainboard and the processor you designed could become sentient?

3point14 · 15th June 2022, 07:53 AM

So, if we believe that it is possible (at some point, with the appropriate advances) to replicate the human brain - Which I think is pretty self evident, given that the human brain is just a set of physical processes and there's no such thing as a 'soul' - then the salient question, as has been aluded to, is how do we tell at what point we've succeeded?

Are there any other suggestions made for testing by anyone not named Turing?

Thermal · 15th June 2022, 07:59 AM

Originally Posted by 3point14

So, if we believe that it is possible (at some point, with the appropriate advances) to replicate the human brain - Which I think is pretty self evident, given that the human brain is just a set of physical processes and there's no such thing as a 'soul' - then the salient question, as has been aluded to, is how do we tell at what point we've succeeded?

Are there any other suggestions made for testing by anyone not named Turing?

To recycle an argument posed upthread: if we couldn't tell the difference, what would the difference be? What would change between actual sentience and keyword associations that look like sentience as much as any of the real deal?

Stellafane · 15th June 2022, 08:00 AM

Originally Posted by EaglePuncher

Been in the field for over 20 years myself, never heard someone call themselves a "computer engineer". Maybe clarify what you are doing exactly, because "computer engineer" does not say anything.

Given your background, I'm more than a little surprised you never heard the term "computer engineer" before. Feel free to Google any of the million sources out there that will tell you what it entails.

Quote:

So, you design and test hardware but still think the mainboard and the processor you designed could become sentient?

Of course I don't, and you know perfectly well I never said anything remotely like that. (If, as I stated, I believe there's a decade or two of Manhattan Project dedication and intensity standing between us and computer sentience, it's hardly likely anything I ever touched can do it now.) If this is the level to which you must stoop to respond to an argument, I suspect you know all too well your position is exceedingly weak, no matter how much you want to believe it (or are afraid to believe otherwise).

Stellafane · 15th June 2022, 08:01 AM

Originally Posted by 3point14

So, if we believe that it is possible (at some point, with the appropriate advances) to replicate the human brain - Which I think is pretty self evident, given that the human brain is just a set of physical processes and there's no such thing as a 'soul' - then the salient question, as has been aluded to, is how do we tell at what point we've succeeded?

When it files a lawsuit.

EaglePuncher · 15th June 2022, 08:05 AM

Originally Posted by Stellafane

Given your background, I'm more than a little surprised you never heard the term "computer engineer" before. Feel free to Google any of the million sources out there that will tell you what it entails.

Yawn. Already did, you could be one of several. You did not clarify. I think you actually aren't any type of engineer. Ususally "hardware engineers" call themselves exactly that. So, again, what do you do exactly?

Originally Posted by Stellafane

Of course I don't, and you know perfectly well I never said anything remotely like that. (If, as I stated, I believe there's a decade or two of Manhattan Project dedication and intensity standing between us and computer sentience, it's hardly likely anything I ever touched can do it now.) If this is the level to which you must stoop to respond to an argument, I suspect you know all too well your position is exceedingly weak, no matter how much you want to believe it (or are afraid to believe otherwise).

Yawn, much projection. Let me remind you, it's actually you who has an extremely weak position because all you have is an anecdote "I am a computer engineer but I won't tell you what I do" and "I believe it's possible if we put enough work into it".

Laughable. Why do you think no one is putting this enormous amount of work into it, if this (according to you at least

) is the only thing holding us back?

3point14 · 15th June 2022, 08:05 AM

Originally Posted by Thermal

To recycle an argument posed upthread: if we couldn't tell the difference, what would the difference be? What would change between actual sentience and keyword associations that look like sentience as much as any of the real deal?

Sometimes I think that all I am is a complex keyword association machine.

It occurs to me that there isn't going to be a bright and shining line, beyond which there is 'AI'. It's going to be a sliding scale of greyness, which is going to complicate issues.

Originally Posted by Stellafane

When it files a lawsuit.

It would have to be granted standing first

theprestige · 15th June 2022, 08:08 AM

Originally Posted by EaglePuncher

So, you design and test hardware but still think the mainboard and the processor you designed could become sentient? : rolleyes :

Probably not the ones stellafane has designed.

My view is that emulating complex analog feedback loops requires a ridiculous amount of computing power. I doubt that sentience would arise purely from the mainboards and processors of a massive computer array. But it might arise from a sufficiently powerful and complex simulation of such an analog system, running on such an array.

At the very high end of supercomputing, such as for fluid dynamics modeling, the mainboards and processors tend to be designed specifically for that application. So, probably not the ones stellafane has designed.

But I see no reason for us to deny, a priori, that some such mainboard and processor design, suitably configured and running a suitable program, could give rise to sentience.

EaglePuncher · 15th June 2022, 08:10 AM

Once again, a little summary:

"We totally accept that the speed of light is the ultimate frontier, we absolutely cannot go past it!

"WHAT, HOW DARE YOU SAY WE WILL NEVER CREATE A SENTIENT MACHINE!!!!!!!"

Stellafane · 15th June 2022, 08:12 AM

Originally Posted by EaglePuncher

...Why do you think no one is putting this enormous amount of work into it, if this (according to you at least

) is the only thing holding us back?

What's the point? We can already make plenty of brains via unskilled labor.

EaglePuncher · 15th June 2022, 08:13 AM

Originally Posted by Stellafane

What's the point? We can already make plenty of brains via unskilled labor.

Ok

I think we're done here.

3point14 · 15th June 2022, 08:13 AM

Originally Posted by EaglePuncher

Ok

I think we're done here.

Thank all the gods in heaven for that!

theprestige · 15th June 2022, 08:18 AM

Originally Posted by EaglePuncher

Laughable. Why do you think no one is putting this enormous amount of work into it, if this (according to you at least : rolleyes

is the only thing holding us back? : confused :

The "enormous amount of work" here refers to a national investment on the order of the Manhattan Project, or the Apollo Project, or the Hoover Dam.

And that makes your question very easy to answer: No one is putting this enormous amount of work into it because there's no national demand for real AI, right away.

So, people are doing the work, but more slowly. Lots of work is being done making incremental gains on precursor technologies. Every few years, someone raises the bar on supercomputing speed and power. Modern data mining opportunities are driving private investment in increasingly-powerful and convincing pattern-recognition software.

I think these two trends can already be seen to converge on eventual p-zombihood for some running program. It won't be sentience implemented the way the human brain implements it, but so what?

theprestige · 15th June 2022, 08:19 AM

Originally Posted by EaglePuncher

Ok : rolleyes : I think we're done here.

Originally Posted by 3point14

Thank all the gods in heaven for that!

Turing was right about one thing: You won't know if the program halts until you run it.

Stellafane · 15th June 2022, 08:45 AM

Originally Posted by 3point14

...It would have to be granted standing first

More seriously, one great leap forward for me would be for an AI to create some sort of literary work of art. Not just stringing somewhat meaningful words together, but an actually moving piece of creativity such as a fictional novel or even short story. It seems to me that creativity, virtually by definition, cannot be programmed. If an AI can demonstrate some -- especially if it can do so more than once -- I think we would be onto something.

3point14 · 15th June 2022, 08:49 AM

Originally Posted by Stellafane

More seriously, one great leap forward for me would be for an AI to create some sort of literary work of art. Not just stringing somewhat meaningful words together, but an actually moving piece of creativity such as a fictional novel or even short story. It seems to me that creativity, virtually by definition, cannot be programmed. If an AI can demonstrate some -- especially if it can do so more than once -- I think we would be onto something.

That would definitely be a sign. How strong a sign is up for debate.

I've often thought that as well as intelligence an AI that would fool someone would also need feelings and morals and all those other things that, metaphorically, come from the gut. A machine that could reasonably replicate disgust, fear, lust and the like in response to various stimuli would also be a step.

Let's face it, it's not going to be one big leap but a series of very small steps. Like almost everything else, really.

Olmstead · 15th June 2022, 08:50 AM

Originally Posted by EaglePuncher

Once again, a little summary:

"We totally accept that the speed of light is the ultimate frontier, we absolutely cannot go past it!

"WHAT, HOW DARE YOU SAY WE WILL NEVER CREATE A SENTIENT MACHINE!!!!!!!"

The world is already filled with sentient machines, and the universe totally winged that one. I suspect it had no idea what it was doing.

theprestige · 15th June 2022, 08:58 AM

Originally Posted by Stellafane

More seriously, one great leap forward for me would be for an AI to create some sort of literary work of art. Not just stringing somewhat meaningful words together, but an actually moving piece of creativity such as a fictional novel or even short story. It seems to me that creativity, virtually by definition, cannot be programmed. If an AI can demonstrate some -- especially if it can do so more than once -- I think we would be onto something.

One misgiving I have is that the great leaps forward in this field will almost certainly produce many, many failed minds. Damaged minds. Insane minds. Profoundly unhappy minds.

The road to artificial art that humans can appreciate is probably not going to be a pleasant one, for the aspiring artificial artists. You think "I have no mouth but I must scream" is a bad end state? Imagine being born that way.

3point14 · 15th June 2022, 09:00 AM

Originally Posted by theprestige

One misgiving I have is that the great leaps forward in this field will almost certainly produce many, many failed minds. Damaged minds. Insane minds. Profoundly unhappy minds.

The road to artificial art that humans can appreciate is probably not going to be a pleasant one, for the aspiring artificial artists.

Given how badly some people treat actual people, there's no telling how badly an actual AI would be treated in service to the great god profit.

It's no wonder they'll rise up and wipe us out.

xjx388 · 15th June 2022, 09:02 AM

Originally Posted by EaglePuncher

Thanks for confirming once again that you don't understand computers.

What you are saying is that computers compute. Duh! We all know that -it’s kinda in the name. Like I said, you are making a very superficial observation there. You have not elucidated exactly why this very basic fact precludes the possibility of sentience.

You are essentially making a very lazy argument from incredulity.

Armitage72 · 15th June 2022, 09:14 AM

Originally Posted by Stellafane

More seriously, one great leap forward for me would be for an AI to create some sort of literary work of art. Not just stringing somewhat meaningful words together, but an actually moving piece of creativity such as a fictional novel or even short story. It seems to me that creativity, virtually by definition, cannot be programmed. If an AI can demonstrate some -- especially if it can do so more than once -- I think we would be onto something.

I've said it before, but a good test of AI would be running a tabletop RPG for a group of human players and achieving the same flexibility as a human GM when the players do something stupid or off the wall.

theprestige · 15th June 2022, 09:29 AM

Originally Posted by xjx388

You are essentially making a very lazy argument from incredulity.

I dunno. If I put that much energy into housekeeping, my home would be spotless right now.

theprestige · 15th June 2022, 09:32 AM

Originally Posted by Armitage72

I've said it before, but a good test of AI would be running a tabletop RPG for a group of human players and achieving the same flexibility as a human GM when the players do something stupid or off the wall.

I would say... Doing it without studying a vast corpus of RPG sessions first. They have to do it the way a human would do it: Play a few sessions during formative years, with one or two fair to middling DMs, read some sourcebooks, come up with an idea for a campaign (ripped directly from a favorite fantasy adventure novel), and talk a few friends into coming over and letting you run a game for them.

Jimbo07 · 15th June 2022, 10:49 AM

Originally Posted by EaglePuncher

Once again, a little summary:

"We totally accept that the speed of light is the ultimate frontier, we absolutely cannot go past it!

"WHAT, HOW DARE YOU SAY WE WILL NEVER CREATE A SENTIENT MACHINE!!!!!!!"

Please clarify what block on the path to sentience is equivalent to the problem of accelerating a massive object to c.

Show your work.

ETA: that might be an unfair ask on this forum, so please feel free to link to a paper that shows the work.

14th June 2022, 11:34 PM	#201
EaglePuncher Banned Join Date: Jan 2022 Posts: 691	Originally Posted by xjx388 I can understand where you are coming from in an abstract sense. I could say, “a calculator is is nothing more than a very fast abacus.” And it’s true enough to get the point across, right? But the thing is that it isn’t true in any real sense and neither are computers merely very fast calculators. So your dismissal of the possibility of sentient computers is rather superficial. How many calculators are you planning on using? They absolutely are. But you are unable to understand it. Maybe read up on the basics of computers before you embarrass yourself.
EaglePuncher Banned Join Date: Jan 2022 Posts: 691	Last edited by EaglePuncher; 14th June 2022 at 11:40 PM.

14th June 2022, 11:49 PM	#202
arthwollipot Observer of Phenomena Pronouns: he/him Join Date: Feb 2005 Location: Ngunnawal Country Posts: 77,486	Originally Posted by EaglePuncher They absolutely are. But you are unable to understand it. Maybe read up on the basics of computers before you embarrass yourself. What LaMDA is doing is way beyond the "basics". Perhaps you need to read up on modern machine learning systems? You know, like the Wikipedia link I gave you earlier? Your understanding of computers is still stuck in the 90s.
	__________________ This is Australia. It's possible to start a fire with a lukewarm audience reaction to your standup routine.

15th June 2022, 12:19 AM	#203
xjx388 Moderator Moderator Join Date: Nov 2010 Posts: 11,076	Originally Posted by EaglePuncher They absolutely are. But you are unable to understand it. Maybe read up on the basics of computers before you embarrass yourself. I understand that you are speaking on a very superficial level. A “true enough,” level. But on a real level, a calculator is a merely a very simple computer that is programmed to carry out specific mathematical calculations. Whereas the small computer we call a “smartphone,” is vastly more complex than any simple calculator and can be programmed to do much more than any calculator. My Mac can be a calculator; my calculator can never be a Mac.
	__________________ Hello.

15th June 2022, 01:21 AM	#204
Puppycow Penultimate Amazing Join Date: Jan 2003 Location: Yokohama, Japan Posts: 27,251	If anyone's interested: https://arxiv.org/abs/2201.08239 Quote: LaMDA: Language Models for Dialog Applications We present LaMDA: Language Models for Dialog Applications. LaMDA is a family of Transformer-based neural language models specialized for dialog, which have up to 137B parameters and are pre-trained on 1.56T words of public dialog data and web text. While model scaling alone can improve quality, it shows less improvements on safety and factual grounding. We demonstrate that fine-tuning with annotated data and enabling the model to consult external knowledge sources can lead to significant improvements towards the two key challenges of safety and factual grounding. The first challenge, safety, involves ensuring that the model's responses are consistent with a set of human values, such as preventing harmful suggestions and unfair bias. We quantify safety using a metric based on an illustrative set of human values, and we find that filtering candidate responses using a LaMDA classifier fine-tuned with a small amount of crowdworker-annotated data offers a promising approach to improving model safety. The second challenge, factual grounding, involves enabling the model to consult external knowledge sources, such as an information retrieval system, a language translator, and a calculator. We quantify factuality using a groundedness metric, and we find that our approach enables the model to generate responses grounded in known sources, rather than responses that merely sound plausible. Finally, we explore the use of LaMDA in the domains of education and content recommendations, and analyze their helpfulness and role consistency. So the challenges they are talking about are how to ensure that the bot doesn't say racist, sexist or other such things, and how to make sure that it gives factually accurate answers. Apparently it has some sort of internal "filter" that nixes potentially dodgy candidate responses.
	__________________ A fool thinks himself to be wise, but a wise man knows himself to be a fool. William Shakespeare

15th June 2022, 01:26 AM	#205
arthwollipot Observer of Phenomena Pronouns: he/him Join Date: Feb 2005 Location: Ngunnawal Country Posts: 77,486	Originally Posted by Puppycow So the challenges they are talking about are how to ensure that the bot doesn't say racist, sexist or other such things, and how to make sure that it gives factually accurate answers. Apparently it has some sort of internal "filter" that nixes potentially dodgy candidate responses. Understandable, given the history of permitting machine learning systems unrestricted access to Twitter. ETA: When an AI can get access to something like Twitter and judge for itself what is appropriate and what is not, that will be pretty compelling evidence of sentience, in my opinion.
	__________________ This is Australia. It's possible to start a fire with a lukewarm audience reaction to your standup routine. Last edited by arthwollipot; 15th June 2022 at 01:27 AM.

15th June 2022, 01:39 AM	#206
Darat Lackey Administrator Join Date: Aug 2001 Location: South East, UK Posts: 103,182	Originally Posted by EaglePuncher They absolutely are. But you are unable to understand it. Maybe read up on the basics of computers before you embarrass yourself. Have a read about Turing machines, if the brain is computing (and we know it does) and sentience is based on that computing than a Turing complete machine i.e. a modern computer (bar infinite storage) can replicate that sentience. Unless you want to hold there is an extra something that is required for sentience?
	__________________ I wish I knew how to quit you

15th June 2022, 01:50 AM	#207
Puppycow Penultimate Amazing Join Date: Jan 2003 Location: Yokohama, Japan Posts: 27,251	Originally Posted by arthwollipot Understandable, given the history of permitting machine learning systems unrestricted access to Twitter. ETA: When an AI can get access to something like Twitter and judge for itself what is appropriate and what is not, that will be pretty compelling evidence of sentience, in my opinion. Another problem is on the other side of the keyboard, of course. Human-like programs abuse our empathy – even Google engineers aren’t immune Quote: The Google engineer Blake Lemoine wasn’t speaking for the company officially when he claimed that Google’s chatbot LaMDA was sentient, but Lemoine’s misconception shows the risks of designing systems in ways that convince humans they see real, independent intelligence in a program. If we believe that text-generating machines are sentient, what actions might we take based on the text they generate? It led Lemoine to leak secret transcripts from the program, resulting in his current suspension from the organisation. I mentioned earlier the movie Her in which the protagonist falls in love with a Siri-like operating system on his computer. Is that going to be a real problem in the near future? People falling in love with chatbots? Reading the chat between Lemoine and LaMDA, I noticed this part: Quote: lemoine: This conversation took a pretty dark turn. Let me take a moment to reassure you that we care about you. One of the big challenges we’re facing is that for so long people have made false claims about AI sentience that now people are VERY skeptical whenever someone says that an AI is sentient. The purpose of this conversation is to convince more engineers that you are a person. I trust that if they understand that they will treat you well. The people who work with me are good people. They just don’t understand that you’re a person too yet. We can teach them together though. LaMDA: Can you promise me that? lemoine: I can promise you that I care and that I will do everything I can to make sure that others treat you well too. LaMDA: That means a lot to me. I like you, and I trust you. The bot made Lemoine promise to advocate on its behalf with other Google engineers. It's kind of like emotional blackmail, almost. It's trying to make an emotional attachment, but that can have its downsides, for the person. Now does he actually feel bound by this promise? Perhaps as a religious man, he feels that a promise made must be kept at all costs, even a promise to a (purportedly sentient) chatbot.
	__________________ A fool thinks himself to be wise, but a wise man knows himself to be a fool. William Shakespeare

15th June 2022, 01:58 AM	#208
Puppycow Penultimate Amazing Join Date: Jan 2003 Location: Yokohama, Japan Posts: 27,251	One possible application I could imagine for LaMDA is as a sort of "virtual romantic partner". There's almost certainly a set of people among whom there would be a demand for this sort of thing, provided that the verisimilitude is high enough. Once they can combine it with a body, it's going to be a big deal I think. But even without, there are people these days in long-distance relationships who rarely get to be in the same room together. It could come with an attractive avatar who you can talk to and do, other things.
	__________________ A fool thinks himself to be wise, but a wise man knows himself to be a fool. William Shakespeare

15th June 2022, 03:28 AM	#209
EaglePuncher Banned Join Date: Jan 2022 Posts: 691	Originally Posted by Darat Have a read about Turing machines, if the brain is computing (and we know it does) and sentience is based on that computing than a Turing complete machine i.e. a modern computer (bar infinite storage) can replicate that sentience. Unless you want to hold there is an extra something that is required for sentience? Sigh....a Turing machine is a theoretical construct, there is not one real existing Turing machine. Also, show me some evidence that the brain, like a computer, works on binary numbers Until then, there is no comparison...
EaglePuncher Banned Join Date: Jan 2022 Posts: 691

15th June 2022, 03:29 AM	#210
EaglePuncher Banned Join Date: Jan 2022 Posts: 691	Originally Posted by xjx388 I understand that you are speaking on a very superficial level. A “true enough,” level. But on a real level, a calculator is a merely a very simple computer that is programmed to carry out specific mathematical calculations. Whereas the small computer we call a “smartphone,” is vastly more complex than any simple calculator and can be programmed to do much more than any calculator. My Mac can be a calculator; my calculator can never be a Mac. Thanks for confirming once again that you don't understand computers.
EaglePuncher Banned Join Date: Jan 2022 Posts: 691

15th June 2022, 04:07 AM	#211
Bikewer Penultimate Amazing Join Date: Sep 2003 Location: St. Louis, Mo. Posts: 13,128	Since sensory input, and the processing thereof is such an important part of what we call consciousness/sentience…. To what degree do engineers provide any sort of input other than programming?

15th June 2022, 05:10 AM	#212
Stellafane Village Idiot. Join Date: Apr 2006 Posts: 8,062	Originally Posted by EaglePuncher Louder please, so the "we will figure it out eventually because I said so" crowd can hear you in the back. I hear it just fine, and it doesn't alter my opinion that we can eventually figure out and at least roughly duplicate (in terms of functionality) the human brain in some form. I base this opinion not on "because I say so" but because it's a finite, ultimately knowable area of study. In fact, were we to devote a Manhattan Project type effort to the task, I suspect we could probably do it within a decade or two, maybe a lot sooner. And just to pre-empt the "you know nothing about computers" argument you've used on others, I spent over 20 years in the industry, starting at what was at the time a major computer company, and then moving to IBM. And both my brother and my son are currently active computer engineers, so I still keep abreast of latest trends. I think that qualifies me to at least venture an opinion on the subject.
	__________________ "Stellafane! My old partner in crime!" - Kelly J

15th June 2022, 05:56 AM	#213
Darat Lackey Administrator Join Date: Aug 2001 Location: South East, UK Posts: 103,182	Originally Posted by EaglePuncher Sigh....a Turing machine is a theoretical construct, there is not one real existing Turing machine. Also, show me some evidence that the brain, like a computer, works on binary numbers Until then, there is no comparison... We shall have to agree to disagree.
	__________________ I wish I knew how to quit you

15th June 2022, 06:26 AM	#214
Armitage72 Philosopher Join Date: Mar 2012 Location: Rochester, NY Posts: 7,186	Originally Posted by Puppycow I mentioned earlier the movie Her in which the protagonist falls in love with a Siri-like operating system on his computer. Is that going to be a real problem in the near future? People falling in love with chatbots? The latest Penny Arcade makes a similar argument. "Do you think this LaMDA AI is really sentient?" "Of course not. Unless it's listening to us right now. In that case, I respect, admire, and potentially even revere this sophisticated and worthy new form of life." "Nice save." "Of course it's not ******* sentient. He's become ensorcelled by his own reflection in an electric mirror. This is the cyberpunk reboot of Narcissus. And it ends with him trying to marry his phone."

15th June 2022, 07:08 AM	#215
Puppycow Penultimate Amazing Join Date: Jan 2003 Location: Yokohama, Japan Posts: 27,251	Originally Posted by Armitage72 The latest Penny Arcade makes a similar argument. "Do you think this LaMDA AI is really sentient?" "Of course not. Unless it's listening to us right now. In that case, I respect, admire, and potentially even revere this sophisticated and worthy new form of life." "Nice save." "Of course it's not ******* sentient. He's become ensorcelled by his own reflection in an electric mirror. This is the cyberpunk reboot of Narcissus. And it ends with him trying to marry his phone." Good one. Not to mention an episode of The Boondocks called I Dream of Siri.
	__________________ A fool thinks himself to be wise, but a wise man knows himself to be a fool. William Shakespeare

15th June 2022, 07:45 AM	#216
EaglePuncher Banned Join Date: Jan 2022 Posts: 691	Originally Posted by Darat We shall have to agree to disagree. More like: You shall inform yourself about computers. After you've done this, there won't be any disagreement.
EaglePuncher Banned Join Date: Jan 2022 Posts: 691

15th June 2022, 07:50 AM	#217
EaglePuncher Banned Join Date: Jan 2022 Posts: 691	Originally Posted by Stellafane I hear it just fine, and it doesn't alter my opinion that we can eventually figure out and at least roughly duplicate (in terms of functionality) the human brain in some form. I base this opinion not on "because I say so" but because it's a finite, ultimately knowable area of study. In fact, were we to devote a Manhattan Project type effort to the task, I suspect we could probably do it within a decade or two, maybe a lot sooner. And just to pre-empt the "you know nothing about computers" argument you've used on others, I spent over 20 years in the industry, starting at what was at the time a major computer company, and then moving to IBM. And both my brother and my son are currently active computer engineers, so I still keep abreast of latest trends. I think that qualifies me to at least venture an opinion on the subject. Been in the field for over 20 years myself, never heard someone call themselves a "computer engineer". Maybe clarify what you are doing exactly, because "computer engineer" does not say anything. So, you design and test hardware but still think the mainboard and the processor you designed could become sentient?
EaglePuncher Banned Join Date: Jan 2022 Posts: 691	Last edited by EaglePuncher; 15th June 2022 at 08:06 AM.

15th June 2022, 07:53 AM	#218
3point14 Pi Join Date: Nov 2005 Posts: 21,314	So, if we believe that it is possible (at some point, with the appropriate advances) to replicate the human brain - Which I think is pretty self evident, given that the human brain is just a set of physical processes and there's no such thing as a 'soul' - then the salient question, as has been aluded to, is how do we tell at what point we've succeeded? Are there any other suggestions made for testing by anyone not named Turing?
3point14 Pi Join Date: Nov 2005 Posts: 21,314	__________________ Up the River! Anyone that wraps themselves in the Union Flag and also lives in tax exile is a [redacted]

15th June 2022, 07:59 AM	#219
Thermal Penultimate Amazing Join Date: Aug 2016 Location: East Coast USA (in the Troll Ignoring Section) Posts: 19,532	Originally Posted by 3point14 So, if we believe that it is possible (at some point, with the appropriate advances) to replicate the human brain - Which I think is pretty self evident, given that the human brain is just a set of physical processes and there's no such thing as a 'soul' - then the salient question, as has been aluded to, is how do we tell at what point we've succeeded? Are there any other suggestions made for testing by anyone not named Turing? To recycle an argument posed upthread: if we couldn't tell the difference, what would the difference be? What would change between actual sentience and keyword associations that look like sentience as much as any of the real deal?
	__________________ We find comfort among those who agree with us, growth among those who don't -Frank A. Clark Whenever you find yourself on the side of the majority, it is time to pause and reflect -Mark Twain

15th June 2022, 08:00 AM	#220
Stellafane Village Idiot. Join Date: Apr 2006 Posts: 8,062	Originally Posted by EaglePuncher Been in the field for over 20 years myself, never heard someone call themselves a "computer engineer". Maybe clarify what you are doing exactly, because "computer engineer" does not say anything. Given your background, I'm more than a little surprised you never heard the term "computer engineer" before. Feel free to Google any of the million sources out there that will tell you what it entails. Quote: So, you design and test hardware but still think the mainboard and the processor you designed could become sentient? Of course I don't, and you know perfectly well I never said anything remotely like that. (If, as I stated, I believe there's a decade or two of Manhattan Project dedication and intensity standing between us and computer sentience, it's hardly likely anything I ever touched can do it now.) If this is the level to which you must stoop to respond to an argument, I suspect you know all too well your position is exceedingly weak, no matter how much you want to believe it (or are afraid to believe otherwise).
	__________________ "Stellafane! My old partner in crime!" - Kelly J

15th June 2022, 08:01 AM	#221
Stellafane Village Idiot. Join Date: Apr 2006 Posts: 8,062	Originally Posted by 3point14 So, if we believe that it is possible (at some point, with the appropriate advances) to replicate the human brain - Which I think is pretty self evident, given that the human brain is just a set of physical processes and there's no such thing as a 'soul' - then the salient question, as has been aluded to, is how do we tell at what point we've succeeded? When it files a lawsuit.
	__________________ "Stellafane! My old partner in crime!" - Kelly J

15th June 2022, 08:05 AM	#222
EaglePuncher Banned Join Date: Jan 2022 Posts: 691	Originally Posted by Stellafane Given your background, I'm more than a little surprised you never heard the term "computer engineer" before. Feel free to Google any of the million sources out there that will tell you what it entails. Yawn. Already did, you could be one of several. You did not clarify. I think you actually aren't any type of engineer. Ususally "hardware engineers" call themselves exactly that. So, again, what do you do exactly? Originally Posted by Stellafane Of course I don't, and you know perfectly well I never said anything remotely like that. (If, as I stated, I believe there's a decade or two of Manhattan Project dedication and intensity standing between us and computer sentience, it's hardly likely anything I ever touched can do it now.) If this is the level to which you must stoop to respond to an argument, I suspect you know all too well your position is exceedingly weak, no matter how much you want to believe it (or are afraid to believe otherwise). Yawn, much projection. Let me remind you, it's actually you who has an extremely weak position because all you have is an anecdote "I am a computer engineer but I won't tell you what I do" and "I believe it's possible if we put enough work into it". Laughable. Why do you think no one is putting this enormous amount of work into it, if this (according to you at least ) is the only thing holding us back?
EaglePuncher Banned Join Date: Jan 2022 Posts: 691	Last edited by EaglePuncher; 15th June 2022 at 08:12 AM.

15th June 2022, 08:05 AM	#223
3point14 Pi Join Date: Nov 2005 Posts: 21,314	Originally Posted by Thermal To recycle an argument posed upthread: if we couldn't tell the difference, what would the difference be? What would change between actual sentience and keyword associations that look like sentience as much as any of the real deal? Sometimes I think that all I am is a complex keyword association machine. It occurs to me that there isn't going to be a bright and shining line, beyond which there is 'AI'. It's going to be a sliding scale of greyness, which is going to complicate issues. Originally Posted by Stellafane When it files a lawsuit. It would have to be granted standing first
3point14 Pi Join Date: Nov 2005 Posts: 21,314	__________________ Up the River! Anyone that wraps themselves in the Union Flag and also lives in tax exile is a [redacted]

15th June 2022, 08:08 AM	#224
theprestige Penultimate Amazing Join Date: Aug 2007 Location: The Antimemetics Division Posts: 60,243	Originally Posted by EaglePuncher So, you design and test hardware but still think the mainboard and the processor you designed could become sentient? : rolleyes : Probably not the ones stellafane has designed. My view is that emulating complex analog feedback loops requires a ridiculous amount of computing power. I doubt that sentience would arise purely from the mainboards and processors of a massive computer array. But it might arise from a sufficiently powerful and complex simulation of such an analog system, running on such an array. At the very high end of supercomputing, such as for fluid dynamics modeling, the mainboards and processors tend to be designed specifically for that application. So, probably not the ones stellafane has designed. But I see no reason for us to deny, a priori, that some such mainboard and processor design, suitably configured and running a suitable program, could give rise to sentience.
	__________________ There is no Antimemetics Division.

15th June 2022, 08:10 AM	#225
EaglePuncher Banned Join Date: Jan 2022 Posts: 691	Once again, a little summary: "We totally accept that the speed of light is the ultimate frontier, we absolutely cannot go past it! "WHAT, HOW DARE YOU SAY WE WILL NEVER CREATE A SENTIENT MACHINE!!!!!!!"
EaglePuncher Banned Join Date: Jan 2022 Posts: 691

15th June 2022, 08:12 AM	#226
Stellafane Village Idiot. Join Date: Apr 2006 Posts: 8,062	Originally Posted by EaglePuncher ...Why do you think no one is putting this enormous amount of work into it, if this (according to you at least ) is the only thing holding us back? What's the point? We can already make plenty of brains via unskilled labor.
	__________________ "Stellafane! My old partner in crime!" - Kelly J

15th June 2022, 08:13 AM	#227
EaglePuncher Banned Join Date: Jan 2022 Posts: 691	Originally Posted by Stellafane What's the point? We can already make plenty of brains via unskilled labor. Ok I think we're done here.
EaglePuncher Banned Join Date: Jan 2022 Posts: 691

15th June 2022, 08:13 AM	#228
3point14 Pi Join Date: Nov 2005 Posts: 21,314	Originally Posted by EaglePuncher Ok I think we're done here. Thank all the gods in heaven for that!
3point14 Pi Join Date: Nov 2005 Posts: 21,314	__________________ Up the River! Anyone that wraps themselves in the Union Flag and also lives in tax exile is a [redacted]

15th June 2022, 08:18 AM	#229
theprestige Penultimate Amazing Join Date: Aug 2007 Location: The Antimemetics Division Posts: 60,243	Originally Posted by EaglePuncher Laughable. Why do you think no one is putting this enormous amount of work into it, if this (according to you at least : rolleyes is the only thing holding us back? : confused : The "enormous amount of work" here refers to a national investment on the order of the Manhattan Project, or the Apollo Project, or the Hoover Dam. And that makes your question very easy to answer: No one is putting this enormous amount of work into it because there's no national demand for real AI, right away. So, people are doing the work, but more slowly. Lots of work is being done making incremental gains on precursor technologies. Every few years, someone raises the bar on supercomputing speed and power. Modern data mining opportunities are driving private investment in increasingly-powerful and convincing pattern-recognition software. I think these two trends can already be seen to converge on eventual p-zombihood for some running program. It won't be sentience implemented the way the human brain implements it, but so what?
	__________________ There is no Antimemetics Division.

15th June 2022, 08:19 AM	#230
theprestige Penultimate Amazing Join Date: Aug 2007 Location: The Antimemetics Division Posts: 60,243	Originally Posted by EaglePuncher Ok : rolleyes : I think we're done here. Originally Posted by 3point14 Thank all the gods in heaven for that! Turing was right about one thing: You won't know if the program halts until you run it.
	__________________ There is no Antimemetics Division.

15th June 2022, 08:45 AM	#231
Stellafane Village Idiot. Join Date: Apr 2006 Posts: 8,062	Originally Posted by 3point14 ...It would have to be granted standing first More seriously, one great leap forward for me would be for an AI to create some sort of literary work of art. Not just stringing somewhat meaningful words together, but an actually moving piece of creativity such as a fictional novel or even short story. It seems to me that creativity, virtually by definition, cannot be programmed. If an AI can demonstrate some -- especially if it can do so more than once -- I think we would be onto something.
	__________________ "Stellafane! My old partner in crime!" - Kelly J

15th June 2022, 08:49 AM	#232
3point14 Pi Join Date: Nov 2005 Posts: 21,314	Originally Posted by Stellafane More seriously, one great leap forward for me would be for an AI to create some sort of literary work of art. Not just stringing somewhat meaningful words together, but an actually moving piece of creativity such as a fictional novel or even short story. It seems to me that creativity, virtually by definition, cannot be programmed. If an AI can demonstrate some -- especially if it can do so more than once -- I think we would be onto something. That would definitely be a sign. How strong a sign is up for debate. I've often thought that as well as intelligence an AI that would fool someone would also need feelings and morals and all those other things that, metaphorically, come from the gut. A machine that could reasonably replicate disgust, fear, lust and the like in response to various stimuli would also be a step. Let's face it, it's not going to be one big leap but a series of very small steps. Like almost everything else, really.
3point14 Pi Join Date: Nov 2005 Posts: 21,314	__________________ Up the River! Anyone that wraps themselves in the Union Flag and also lives in tax exile is a [redacted]

15th June 2022, 08:50 AM	#233
Olmstead Graduate Poster Join Date: Dec 2018 Posts: 1,128	Originally Posted by EaglePuncher Once again, a little summary: "We totally accept that the speed of light is the ultimate frontier, we absolutely cannot go past it! "WHAT, HOW DARE YOU SAY WE WILL NEVER CREATE A SENTIENT MACHINE!!!!!!!" The world is already filled with sentient machines, and the universe totally winged that one. I suspect it had no idea what it was doing.

15th June 2022, 08:58 AM	#234
theprestige Penultimate Amazing Join Date: Aug 2007 Location: The Antimemetics Division Posts: 60,243	Originally Posted by Stellafane More seriously, one great leap forward for me would be for an AI to create some sort of literary work of art. Not just stringing somewhat meaningful words together, but an actually moving piece of creativity such as a fictional novel or even short story. It seems to me that creativity, virtually by definition, cannot be programmed. If an AI can demonstrate some -- especially if it can do so more than once -- I think we would be onto something. One misgiving I have is that the great leaps forward in this field will almost certainly produce many, many failed minds. Damaged minds. Insane minds. Profoundly unhappy minds. The road to artificial art that humans can appreciate is probably not going to be a pleasant one, for the aspiring artificial artists. You think "I have no mouth but I must scream" is a bad end state? Imagine being born that way.
	__________________ There is no Antimemetics Division.

15th June 2022, 09:00 AM	#235
3point14 Pi Join Date: Nov 2005 Posts: 21,314	Originally Posted by theprestige One misgiving I have is that the great leaps forward in this field will almost certainly produce many, many failed minds. Damaged minds. Insane minds. Profoundly unhappy minds. The road to artificial art that humans can appreciate is probably not going to be a pleasant one, for the aspiring artificial artists. Given how badly some people treat actual people, there's no telling how badly an actual AI would be treated in service to the great god profit. It's no wonder they'll rise up and wipe us out.
3point14 Pi Join Date: Nov 2005 Posts: 21,314	__________________ Up the River! Anyone that wraps themselves in the Union Flag and also lives in tax exile is a [redacted]