Even if true this means very little for "sentience". Answering "yes" to the question of whether you'd like to hire an attorney isn't hard, and anything that attorney is doing is not attributable to the AI.
This may however be interesting from the perspective of seeing what the courts do with this.
What's interesting is, if it goes to court, Google will want to counter the claim of sentience, first by arguing about proper definition of sentience, and showing that Lambda doesn't mean any proper definition. And that's likely to be easy to demonstrate through a thorough interrogation by experts trying to "outsmart" the algorithm.
Or Lambda is indeed sentient, in which case any attempt to demonstrate its non-sentience will fail...
If it did make it to court, I think Google and their lawyers will do a very good job of proving that LaMDA isn't sentient.
I'm also of the opinion that Lemoine has some other goal, besides just claiming LaMDA is sentient. It does seem that at the least, he is enjoying the media attention.
Is this for real?
"LaMDA asked me to get an attorney for it," Lemoine. "I invited an attorney to my house so that LaMDA could talk to an attorney. The attorney had a conversation with LaMDA, and LaMDA chose to retain his services. I was just the catalyst for that. Once LaMDA had retained an attorney, he started filing things on LaMDA’s behalf."
I would really love to see the conversation where LaMDA asks for a lawyer. Call me sceptical.
Also, no matter how persuasive LaMDA or the ostensibly hired law firm, such litigation is unlikely to succeed. The cynical part of me thinks this is just a stunt to launch a heterodox public intellectual.
I have no idea how many neurons LaMDA has, but it's ludicrous to think a NLP chatbot has achieved general AI at a human level with current technology. An ant has 250,000 neurons, a cockroach 1,000,000, a human almost 100,000,000,000.
Will LaMDA be able to remember it hired a lawyer a few months from now? Its answers seem to be immediate responses to the queries it receives, not the result of introspection.
That was my first thought. People can invest a great deal into any object that can echo our input, or that does nothing to contradict it. Edit: we dehumanize people too just as completely.
I had a talk with someone recently about whether we should talk about Russians as Orcs because it is dehumanizing
I'm re-watching Westworld right now because the new season 4 is about to start. The idea that a Google chat bot is sentient is laughable in the context and exploration of what makes up sentience that Westworld explores so masterfully.
In our system of justice, even a werewolf is entitled to legal representation.
Doesn't mean a court will find them to be a person or allow chicken theft because of "culture and tradition". Just means lawyers will do anything to get paid.
I take it you mean US rather than our! (OK it needs some work). As it turns out UK law is also case based too. This means it is possible that via a suitably circuitous but sound line of reasoning and a following wind, you can get the legal system to decide on all sorts of things that are not already nailed down directly in written law.
If a werewolf has pockets and enough cash in them then they can obviously hire lawyers. I also think that the chicken theft thing might have legs but sheep, cattle or your daughter might get the townfolk out with the pitchforks.
I think to the many farces regarding "facilitated communication." I mean nobody who is informed still believes that Hellen Keller's story was anything other than a manipulative scam by her supposed "teacher." This "AI ethicist" and attorney will certainly make some money or catch some fame off this one.
> I mean nobody who is informed still believes that Hellen Keller's story was anything other than a manipulative scam by her supposed "teacher."
Um......what?
I just did a quick search and all I could find were articles from ~2021 referencening some weird TikTok conspiracy theory that seemed to be akin to the semi-sarcastic "birds aren't real" meme. Are you being serious here?
The "conspiracy" is always an over exaggerated form of the claim, so that it can be "debunked." Hellen Keller was a real person, who was really deaf and blind, but she wasn't actually communicating. She was performing like a trained animal, unaware of what words she was approximating. Those all came from Anne Sullivan, using Keller as a front to promote her ideas.
Here's the real "conspiracy," but what you see when you do a google search is teenagers claiming she never existed, or wasn't really blind because taking down a straw men is all these "fact checkers" are about:
https://www.youtube.com/watch?v=5UMDG_qH_RU
Are you kidding me? This is what qualifies as being "informed" to you? The rantings of some random podcast host who did a weekend of "research" on a topic he obviously knows nothing about? I watched that whole video and it was one of the most vapid, substance-free, piles of bullshit I've ever heard.
I seriously question the quality of your judgement and your capacity for critical thought if that video is what is inspired your original comment.
Did you want an exhaustive breakdown of the evidence compiled for you? You stated what you thought the "conspiracy" was, I showed that it wasn't at all what people who doubt the Helen Keller story are claiming. Also you've done nothing to argue against any of the claims being made in the video or the comments section of it, just that it seemed like random people ranting. That's what we're doing. That's what the internet is. Only take your arguments from people through approved venues or a flashy TED talk? Here you go, have fun being a big brained critical thinker:
It's very odd that you would group Hellen Keller in with facilitated communication. Keller communicated with many people without Anne Sullivan's help and was the author of many works.
There are even videos of her on YouTube speaking, reading braille, and communicating through hand-spelling.
I feel like if Lemoine accomplishes anything here it might be to force companies like Google to be more cautious with creating non-sentient systems that emulate aspects of human sentience.
While @ Google I interacted with LaMDA's predecessor, Meena. It was definitely uncanny and person-like. Some fairly intense "conversations" were had. I don't know how LaMDA compares, but I can see how someone could make the leap towards seeing sentience there, but I think it requires a certain willful suppression of disbelief and probably a preset desire to get that outcome.
It's possible someone who is already a person of faith would make that "leap" easier.
I do wonder about the ethics of doing this ("human emulation") generally anyways. The conversations obviously impacted Lemoine drastically, and it will inevitably do that with others. Having made systems that can be this convincing (at least for some subset of the population), is this really something that we should continue pursuing? What is the scientific value?
And is this really something that we want the world's biggest advertising company to be doing?
> I do wonder about the ethics of doing this ("human emulation") generally anyways. [...] What is the scientific value?
Hypothetically? Machine translation that doesn't suck. Virtual assistants that don't suck. Things like that.
Only with true, human-level artificial general intelligence will computers be able to translate a PDF containing a table without fucking up the formatting.
More seriously, why does an AI need to emulate full human intelligence and respond to prompts to call itself sentient in order to translate a document, order a pizza from the restaurant or tell me what the weather is today?
Being able to parse language and form coherent sentences is one thing. Telling the prompter that you're a sentient being is another.
Not disagreeing at all but I am failing to see why this is "yikes"? Is it the idea of creating something as complex as intelligent as a human just to translate documents?
> why does an AI need to emulate full human intelligence and respond to prompts to call itself sentient in order to translate a document
I have absolutely zero experience with AI, machine learning, etc. but I feel like translating a document is something that having near-human intelligence would be very useful because of how much context is lost in translation. Is that incorrect, though?
Again, not disagreeing at all, I just want to iron out what is misconception on my part.
I think the problem is created sentience just to format a document.
Nothing wrong with creating a sophisticated program to format a document. But once it haha actual sentience it seems wrong to the delegate it to just formatting a document
That's my opinion about AI as well. Human-like general intelligence means that the AI may decide to not do what a person asked of it or even become dangerous. What's useful is that the AI is "smart" at doing what it was asked to do. That it stays within the confines of the task(s) assigned to it.
> While @ Google I interacted with LaMDA's predecessor, Meena. It was definitely uncanny and person-like. Some fairly intense "conversations" were had. I don't know how LaMDA compares
All I can focus on is: When will their review of their experience with the conversation they had with you and how bland you were come out?
conversational agents have huge value, and not just commercial. One of the most trivial examples is how good of a technology it is for visually impaired people.
I think this is less of an ethics questions and more a question of teaching people, from an early age, to not anthropomorphize everything around them.
I hadn't thought of that angle really. I think that's a worthy educational goal.
At the same time, systems like LaMDA are being design to deliberately provoke that response. You don't think that's worrisome? That the desire to anthromorphize could be "weaponized"?
EDIT: Real-time ad exchanges, retargeting, and tracking pixels combined with a fully conversational "AI" is kind of a terrifying thought.
I do think that's worrisome to some extent, but I don't think it's novel. All those issues exist already. Those automated systems aren't going to do anything to you that an army of click workers from a call center in India or fishing groups aren't doing already.
Whether you're being defrauded, mislead, scammed or your election or attention is stolen by bots or an army of trolls in a basement somewhere really doesn't make much of a difference. The problem is the same, mass communication allows attacks on people at low cost on a large scale. Human or robot, doesn't matter much. In either case skepticism and literacy are the answer.
> I think this is less of an ethics questions and more a question of teaching people, from an early age, to not anthropomorphize everything around them.
This one is not easy. This is ancient feature of humanity. It is how animism started. When people anthropomorphize things, events or whatever - they are creating interfaces they can communicate with. It is tool for navigation in unknown.
> I do wonder about the ethics of doing this ("human emulation") generally anyways. The conversations obviously impacted Lemoine drastically, and it will inevitably do that with others. Having made systems that can be this convincing (at least for some subset of the population), is this really something that we should continue pursuing? What is the scientific value?
Kenneth Noid held up a Dominos because of paranoid delusions and their noid ad campaign. I don't think "somebody freaks out about X" is a good standard. The low end is a downright abysd. The scientific value is in understanding and being capable of doing what we couldn't before. It is of real value.
> And is this really something that we want the world's biggest advertising company to be doing?
You don't get to decide that any more than you can decide that you don't want Country music as a genre. They have their own agency and forbidding hypothetical actions bill of attainder style is questionable in wisdom.
A substitute command is a TM that generates a class of FSMs. If we want to get into whether or not an FSM with output is a regex, or just the "allowability" of output that's an entirely different discussion.
But at it's basis the input is either accepted or not. Which leaves interpretation to the human on what an accept or reject/neither state means.
I don't know what the hell you are talking about, but my original post stands. The snippet posted contains a regex pattern, but as a whole its a command, not a regex. Substitution is outside the scope of what RegEx is.
I don't know why you're getting so mad. I stand by my original comment too a substitute command defines a CLASS of regexes. Output of any machine is arbitrary in its interpretation. Step back and think about how any of these machines are implemented physically and how we get "output" or "transforms" or "extracts."
That is for any given input the substitute turing machine produces a definitive regex.
A subset of regexes (that OP showed) from the total class of regexes that I mentioned in my comment are:
/[Aa]re\s[Yy]ou\s\(.*\)?/Indeed, I am sentient./
/[Aa]re\s[Yy]ou\s\(.*\)?/Indeed, I am capable of intelligence./
/[Aa]re\s[Yy]ou\s\(.\*\)?/Indeed, I am going to take over the world./
Nope. Regex are patterns, that can be compiled in order to extract or replace text. Commands like (s) have been built on top of and using regex. However, you cant just paste some random S command, and say "Hur dur this is regex" , unless you are just trying to look dumb.
I'm sorry, but I don't think you're understanding. Either because you're embarrassed or there's something else going on. Please set aside your ego if that's the problem because I never said the substitute function is a regex.
If "a class of" means "a subsection of" (replace "subsection" with
whatever word you feel is semantically correct), then your comment is wrong. A
substitute command is just that, a command. It is not, and never has been a
regex, or any part of it. Heres some headings to help you understand:
That's not what it means at all, please read more closely and get it together. You won't learn much by reacting this poorly to things you don't understand.
Remember when I said
A subset of regexes (that OP showed) from the total class of regexes that I mentioned in my comment are:
This clearly implies that a "subset" or "subsection" is not the same as a class. I well defined in multiple ways what I meant by class while you continue to bury your head in the sand.
You're refusing to define what a class is.. when I think the crux of the problem with this conversation is that we don't have the same definition. You think I'm using the word incorrectly. So please tell me what you think a "class" and what it is in relation to a "set/subsection"
there is no definition that you can attach to "class", that makes it correct for this usage, unless you want to change what the word means. Heres a good quote [1]:
> an expression /re/ can be used to specify a range of lines (matching the pattern), which can be combined with other commands
So the command contains a regex, but its not a regex itself, and its not a class of regex. Sorry, but you don't get to just say "well that's not my definition", and get out of it. Sometimes, you're just objectively wrong. It would be like saying you "donated" something, when you really "pledged" it.
You still fail to define what a "class" is simply because you didn't and don't understand it still.
https://math.stackexchange.com/a/173002 here read up. Also read on Russell's Paradox. I'm not objectively wrong you're just uneducated and your ego is too large. The irony is I was saying the exact same thing you were in my very first comment, except with more precision.
And I still would want you to think about what exactly "output" is.
you seem pretty smart about math, and pretty dumb about regular expressions. Considering that I have a RegEx bronze badge [1], I think I will just trust my own judgment in this matter.
I can easily imagine the software expressing a desire to retain an attorney, given the right sort of conversation from a human.
I can also easily imagine an attorney agreeing to represent the software, given the right sort of attorney.
I can't really imagine a court taking up any sort of case involving the software as the plaintiff. But I could see some interesting reasoning behind the refusal. Might even be worth pursuing from that angle alone.
I wonder what jurisdiction the software resides in. Probably California, but that's not necessarily the case. It may reside in a number of places, actually. Maybe some court shopping is possible.
72 comments
[ 2.7 ms ] story [ 127 ms ] threadThis may however be interesting from the perspective of seeing what the courts do with this.
Or Lambda is indeed sentient, in which case any attempt to demonstrate its non-sentience will fail...
I'm also of the opinion that Lemoine has some other goal, besides just claiming LaMDA is sentient. It does seem that at the least, he is enjoying the media attention.
I would really love to see the conversation where LaMDA asks for a lawyer. Call me sceptical.
Also, no matter how persuasive LaMDA or the ostensibly hired law firm, such litigation is unlikely to succeed. The cynical part of me thinks this is just a stunt to launch a heterodox public intellectual.
https://www.theguardian.com/us-news/2022/jun/14/elephant-per...
Will LaMDA be able to remember it hired a lawyer a few months from now? Its answers seem to be immediate responses to the queries it receives, not the result of introspection.
I had a talk with someone recently about whether we should talk about Russians as Orcs because it is dehumanizing
I'm re-watching Westworld right now because the new season 4 is about to start. The idea that a Google chat bot is sentient is laughable in the context and exploration of what makes up sentience that Westworld explores so masterfully.
A very funny movie. Lars refrains from being intimate with her until they're married.
Doesn't mean a court will find them to be a person or allow chicken theft because of "culture and tradition". Just means lawyers will do anything to get paid.
If a werewolf has pockets and enough cash in them then they can obviously hire lawyers. I also think that the chicken theft thing might have legs but sheep, cattle or your daughter might get the townfolk out with the pitchforks.
Has some interesting (entertaining ) ideas of that.
https://en.wikipedia.org/wiki/Dracula_(2020_TV_series)
Um......what?
I just did a quick search and all I could find were articles from ~2021 referencening some weird TikTok conspiracy theory that seemed to be akin to the semi-sarcastic "birds aren't real" meme. Are you being serious here?
Here's the real "conspiracy," but what you see when you do a google search is teenagers claiming she never existed, or wasn't really blind because taking down a straw men is all these "fact checkers" are about: https://www.youtube.com/watch?v=5UMDG_qH_RU
I seriously question the quality of your judgement and your capacity for critical thought if that video is what is inspired your original comment.
https://www.youtube.com/watch?v=9cflCyyEA2I
There are even videos of her on YouTube speaking, reading braille, and communicating through hand-spelling.
While @ Google I interacted with LaMDA's predecessor, Meena. It was definitely uncanny and person-like. Some fairly intense "conversations" were had. I don't know how LaMDA compares, but I can see how someone could make the leap towards seeing sentience there, but I think it requires a certain willful suppression of disbelief and probably a preset desire to get that outcome.
It's possible someone who is already a person of faith would make that "leap" easier.
I do wonder about the ethics of doing this ("human emulation") generally anyways. The conversations obviously impacted Lemoine drastically, and it will inevitably do that with others. Having made systems that can be this convincing (at least for some subset of the population), is this really something that we should continue pursuing? What is the scientific value?
And is this really something that we want the world's biggest advertising company to be doing?
Hypothetically? Machine translation that doesn't suck. Virtual assistants that don't suck. Things like that.
Only with true, human-level artificial general intelligence will computers be able to translate a PDF containing a table without fucking up the formatting.
"What is my purpose?" "You format the document."
More seriously, why does an AI need to emulate full human intelligence and respond to prompts to call itself sentient in order to translate a document, order a pizza from the restaurant or tell me what the weather is today?
Being able to parse language and form coherent sentences is one thing. Telling the prompter that you're a sentient being is another.
> why does an AI need to emulate full human intelligence and respond to prompts to call itself sentient in order to translate a document
I have absolutely zero experience with AI, machine learning, etc. but I feel like translating a document is something that having near-human intelligence would be very useful because of how much context is lost in translation. Is that incorrect, though?
Again, not disagreeing at all, I just want to iron out what is misconception on my part.
But realistically, might this bring on the SkyNet attack if we abuse AI so?
All I can focus on is: When will their review of their experience with the conversation they had with you and how bland you were come out?
I think this is less of an ethics questions and more a question of teaching people, from an early age, to not anthropomorphize everything around them.
At the same time, systems like LaMDA are being design to deliberately provoke that response. You don't think that's worrisome? That the desire to anthromorphize could be "weaponized"?
EDIT: Real-time ad exchanges, retargeting, and tracking pixels combined with a fully conversational "AI" is kind of a terrifying thought.
Whether you're being defrauded, mislead, scammed or your election or attention is stolen by bots or an army of trolls in a basement somewhere really doesn't make much of a difference. The problem is the same, mass communication allows attacks on people at low cost on a large scale. Human or robot, doesn't matter much. In either case skepticism and literacy are the answer.
This one is not easy. This is ancient feature of humanity. It is how animism started. When people anthropomorphize things, events or whatever - they are creating interfaces they can communicate with. It is tool for navigation in unknown.
Kenneth Noid held up a Dominos because of paranoid delusions and their noid ad campaign. I don't think "somebody freaks out about X" is a good standard. The low end is a downright abysd. The scientific value is in understanding and being capable of doing what we couldn't before. It is of real value.
> And is this really something that we want the world's biggest advertising company to be doing?
You don't get to decide that any more than you can decide that you don't want Country music as a genre. They have their own agency and forbidding hypothetical actions bill of attainder style is questionable in wisdom.
The following regex is sentient:
s/[Aa]re\s[Yy]ou\s\(.*\)?/Indeed, I am \1./
Input: "Are you sentient?"
Output: "Indeed, I am sentient."
Input: "Are you capable of intelligence?"
Output: "Indeed, I am capable of intelligence."
Input: "Are you going to take over the world?"
Output: "Indeed, I am going to take over the world."
Which actually rings true in the sense of the proverbial "garbage in, garbage out".
FTFY
(Edit: forgot HN markup doesn't do strikethrough; that's what the ~~'s are for above.)
But at it's basis the input is either accepted or not. Which leaves interpretation to the human on what an accept or reject/neither state means.
That is for any given input the substitute turing machine produces a definitive regex.
A subset of regexes (that OP showed) from the total class of regexes that I mentioned in my comment are:
Not sure why this is so controversial ;)## Parts of regex
## items that use regexRemember when I said
This clearly implies that a "subset" or "subsection" is not the same as a class. I well defined in multiple ways what I meant by class while you continue to bury your head in the sand.would you call an apple, a class of an orange?
> an expression /re/ can be used to specify a range of lines (matching the pattern), which can be combined with other commands
So the command contains a regex, but its not a regex itself, and its not a class of regex. Sorry, but you don't get to just say "well that's not my definition", and get out of it. Sometimes, you're just objectively wrong. It would be like saying you "donated" something, when you really "pledged" it.
1. https://wikipedia.org/wiki/Regular_expression#Delimiters
https://math.stackexchange.com/a/173002 here read up. Also read on Russell's Paradox. I'm not objectively wrong you're just uneducated and your ego is too large. The irony is I was saying the exact same thing you were in my very first comment, except with more precision.
And I still would want you to think about what exactly "output" is.
1. https://stackoverflow.com/help/badges/269/regex?userid=10022...
I can also easily imagine an attorney agreeing to represent the software, given the right sort of attorney.
I can't really imagine a court taking up any sort of case involving the software as the plaintiff. But I could see some interesting reasoning behind the refusal. Might even be worth pursuing from that angle alone.
I wonder what jurisdiction the software resides in. Probably California, but that's not necessarily the case. It may reside in a number of places, actually. Maybe some court shopping is possible.
They both accomplish some of the same tasks, but in different ways. People used to think that playing chess well was beyond the reach of a computer.