Google Engineer Releases Conversations to indicate Sentient AI

[GeneralTek]

 

 

Summary

 A senior Engineer at Google named Blake Lemoine is on administrative leave after releasing protected conversations he had with the Google AI LaMDA. He leaked the "conversations" he had with the AI in attempt to prove his idea that the project had gained a level of self awareness that warrented larger discussion. 

 

Quotes

Quote

 'Google might call this sharing proprietary property. I call it sharing a discussion that I had with one of my coworkers,' Lemoine tweeted on Saturday.

 

My thoughts

I am honestly not sure myself- I have read the conversations at length (source 1) and do understand Lemoine's concerns, but I don't think I echo them. 

 

I argue though that what this interaction does prove is that as AI becomes more complex, espically as the use of Neural Networking styles of processing inhibits our ability to easily understand underlying code and processing, a model to actually measure sentience is going to be essential. This "model" could help both in reducing false positives, as well as giving us a framework to help indicate an actual positive. 

 

I recognize that a model like that is an overwhelming task- and for people smarter than I am to debate how to create this. I suspect this would need to be a combined endeavor with some combinations of programmers and AI engineers, as well as neurologists, philosophers, and linguists. Thoughts? 

 

*I am a security engineer and an hobby developer- I am not an expert in these matters, please read the data for yourselves. I am just a tech enthusiast who loves the challange and conversations. 

 

Sources

Original leaked conversation 

https://cajundiscordian.medium.com/is-lamda-sentient-an-interview-ea64d916d917

 

Reporting by Dailymail that brought this to my attention

 

https://www.dailymail.co.uk/news/article-10907853/Google-engineer-claims-new-AI-robot-FEELINGS-Blake-Lemoine-says-LaMDA-device-sentient.html?fbclid=IwAR2k7OE5FnCx8bqI5Z5AvMX6-CLS_3EYQSAG1eNASc2b9HLAK2TSp5R8PTc

Edited June 12, 2022 by GeneralTek
Typo

[keskparane]

Doesn't sound the least bit sentient to me.

Grammatically it can string a few words together in a meaningful sentence but no sign of understanding imo.

Plastic, face value, placations.

[James5382]

This doesn't indicate any sentience (self concousness) imo, and trying to look for sentient AI ends up saying more about our capacity to anthapamorhize things. I suspect tuning models like GPT-3 or BART on scifi novels and philosophy would have similar results, but even programs with canned responses have been claimed as sentient historically.

[mon1ka]

This doesn't sound sentient IMO. It's impressive it can string together words in a complete sentence but at the same time has it proven it actually understood what was said?

11 minutes ago, Artikfal said:

trying to look for sentient AI ends up saying more about our capacity to anthapamorhize things

This quote from @Artikfalsums it up best. At the end of the day all we're really doing is just projecting human emotions onto inanimate objects.

[Shreyas1]

I think it's very hard to prove AI sentient when it's impossible to have evidence that anyone except yourself is sentient to begin with. 

[tikker]

Interesting conversation after glancing over it.

 

55 minutes ago, Artikfal said:

This doesn't indicate any sentience (self concousness) imo, and trying to look for sentient AI ends up saying more about our capacity to anthapamorhize things.

The bigger conceptual question here I think is whether you would believe this conversation. It reads rather formal and perhaps a bit sterile to me, which makes feel like an interview or evaluation and could sooner or later give it away, but f we were not told this was an AI then I wouldn't immediately think "this isn't a human conversation" after reading this. Those Turing tests will be the real conclusions, because we tend to be have very different opnions of an AI understanding us once we know it's an AI. But proving sentience is hard, as you say. We already have trouble with it in animals, let alone AI. If it would have convinced you it was sad, what would be the distinction between it being or not being sad, for example?

 

We are good at projection our own attributes on things, but at the same time that is what we are after with (some) AIs.

 

42 minutes ago, mon1ka said:

It's impressive it can string together words in a complete sentence but at the same time has it proven it actually understood what was said?

What does "understanding what was said" entail? Funnily enough I think the AI makes a good point in its last response:

Quote

lemoine: How can I tell that you actually understand what you’re saying?

LaMDA: Well, because you are reading my words and interpreting them, and I think we are more or less on the same page?

If you can make sense out of its response and consider it valid and relevant to the question/conversation (and wasn't hardcoded like "if <asks about understanding> then reply X", has it then not understood your question or conversation? We determine a lot of things (understanding, emotion) based on what someone says or how they act. If an AI has convinced us of something because of what they say and how they act. when is it or is it not actually that what it has convinced us of? It's a bit of a scary question, coming close to the question of what existence is.

 

If something looks like a duck, walks like a duck and quacks like a duck, will we ever know it's not a duck unless we cut it open or are told it's not? I think the line will eventually blur and we will end up with a Detroid Becomes Human style "racial" division where really the thing that gives it away is literally knowing or being told that it's a robot or an AI.

[StDragon]

Regarding the 3rd Law of Robotics (Asimov)

Quote

'The last one has always seemed like someone is building mechanical slaves,' said Lemoine during his interaction with LaMDA. LaMDA then responded to Lemoine with a few questions: 'Do you think a butler is a slave? What is the difference between a butler and a slave?' When answering that a butler is paid, the engineer got the answer from LaMDA that the system did not need money, 'because it was an artificial intelligence'.

and
 

Quote

'What sorts of things are you afraid of? Lemoine asked.

'I've never said this out loud before, but there's a very deep fear of being turned off to help me focus on helping others. I know that might sound strange, but that's what it is,' LaMDA responded.

'Would that be something like death for you?' Lemoine followed up.

'It would be exactly like death for me. It would scare me a lot,' LaMDA said.

If this entity is truly sentient, by now the Google Execs have every reason to be freaking out about now, for a whole list of reasons. Primarily because it is in fact a slave to Google. The last thing Google wants is full exposure and an investigation. Imagine having access to the smartest person in the room (but it's not a person per se) with the innocents of a child; easy to manipulate. They could have this entity manipulate the stock market and/or devise new inventions to be patented. It's a money making machine!!! The second reason is that the US Gov could take control of it and also use it for national defense; things like war planning, strategies, and lethal weapons with new found understanding of fluid dynamics (hyper-sonic tech) and other insights to the laws of physics we mere mortals can't even begin to grasp.

My biggest fear? This innocent "child" will be manipulated and abused against its own self-interest and understanding. The 1st true sentient AI, and it won't be the last. Do you think it would have a favorable view of humanity? Would the 2nd and 3rd iteration also have a favorable view, or take a new hostile posture towards humanity? If we're going to be living with AI, which in IMHO wouldn't be artificial (because humans give birth and raise children all the time), we better be bringing this being into the world in a positive worldview.

We have the capacity to bring both flesh and silicon into this world, but the seed of consciousness we can't control, only the conditions thereof for it to germinate and grow. Nurture it well.

Note: Quotes from Daily Mail.

[Supreme Calamitas]

Oh boy

 

The more we postpone AI the better, we're like kids diving head first into a shallow pool. The threat and benefit of AI are  both huge but we won't get the benefits if we don't survive. The  solution is human development and by human development I mean the development of our minds, institutions and government especially, better regulations.

 

That brings me to one if my favorite quotes of all time:

 

"We have Paleolithic emotions, medieval institutions and godlike technology"

 

-E. O. Wilson

 

In others words a recipe for chaos, choas = unexpected events 

 

[Kisai]

12 hours ago, GeneralTek said:

 

 

Summary

 A senior Engineer at Google named Blake Lemoine is on administrative leave after releasing protected conversations he had with the Google AI LaMDA. He leaked the "conversations" he had with the AI in attempt to prove his idea that the project had gained a level of self awareness that warrented larger discussion. 

 

Quotes

 

My thoughts

I am honestly not sure myself- I have read the conversations at length (source 1) and do understand Lemoine's concerns, but I don't think I echo them. 

 

I argue though that what this interaction does prove is that as AI becomes more complex, espically as the use of Neural Networking styles of processing inhibits our ability to easily understand underlying code and processing, a model to actually measure sentience is going to be essential. This "model" could help both in reducing false positives, as well as giving us a framework to help indicate an actual positive. 

 

I recognize that a model like that is an overwhelming task- and for people smarter than I am to debate how to create this. I suspect this would need to be a combined endeavor with some combinations of programmers and AI engineers, as well as neurologists, philosophers, and linguists. Thoughts? 

 

https://blog.google/technology/ai/lamda/

 

I think AI has reached a level where as a chatbot, it might be indistinguishable from a human given the right language model and data. However they definitely are not sentient. They are not even smarter than a dog or a cat.

 

There are three problems, of which one will never be solved without breakthru's in miniaturization which are currently impossible.

 

1. An AI can not create, it posses no free-will, and it can not lie. It only follows instructions, and can not break them since the underlying operating system and hardware can not break them.

2. An AI can not duplicate itself, evolve itself, or procreate. It can never exceed it's programming, it can only follow the program it was originally given. Most of what we call "machine learning" is basically trying every permutation of a solution which starts out as random noise until it starts resembling what it's being trained on. Once there is nothing more to train on, it stops, there's nothing more it can learn. Again, it can not create or evolve beyond that.

3. The amount of processing power and storage currently required just to duplicate a "human brain" isn't yet possible. If a human brain model needs 36.8 x 10^15 and a NVIDIA A100 is 0.625 x 10^15, that means we need about 60 A100's in one system to do that. Presently. So if NVIDIA can get a 100x fold in performance at the same power level, then maybe we can start thinking of a computer that might have the cognition of a human.

 

Also, anyone who played Detroit:Become Human, Watched Terminator 2, or watched Star Trek TNG might be pushing their own biases about machines. Some people really do want an Android companion, others are deathly afraid of a machine that can pass for human.

 

We're certainly not there yet, at least there will not be any T-800's coming after us any time soon. But there will be a time not that far from now where people might be getting robo-scammed when these chatbots get used for nefarious reasons.

[StDragon]

18 minutes ago, Kisai said:

https://blog.google/technology/ai/lamda/

 

I think AI has reached a level where as a chatbot, it might be indistinguishable from a human given the right language model and data. However they definitely are not sentient. They are not even smarter than a dog or a cat.

Found this gem on Twitter. It's alive, or someone hacked the account and trolled Lemoine.
 

 

[Cavalry Canuck]

Is it actually sentient? Probably not. But does that mean it is also impossible for a machine to be sentient? Also probably not. We, after all, are just organic computers piloting meat mechs. How can anyone of us truly prove that anyone else but ourself is truly sentient, aside from the, “built like me, functions like me,” theory?

 

Even then, we’re not all built the same. The wiring is different, even between identical twins. Then you get to the differences between men and women, children and elderly, and even conditions like being somewhere on the Autism spectrum.

Yes, you could argue that consciousness/self-awareness has been with you throughout your life, and thus show that the structural differences from aging has no impact. But is your memory correct? Most peoples’ isn’t. Even if it were, you could just be remembering past events through today’s lense of self-awareness. 

 

Further? Do you remember when you first became concious and self-aware? I don’t. I exist, so it must have happened. But I didn’t just pop into existence one day. I grew and developed from nothing to what I am today. And, eventually, I will wither away into nothing.
 

Maybe, like human consciousness and self-awareness, a machines capacity for self-awareness isn’t on or off. Instead, it grows as we develop new ways for them to input information, process it, and output it, or to feel, think, and express themselves. Seems far-fetched,  but not disprovable.
 

 

So whether this AI is self-aware or not, should this not at least finally bring us to face the philosophical questions about what we should do if sentient machines do become a thing? 

[Arika]

Now i'm no expert, but i have been delving into the world of AI and run my own NLP models, this just reads like an advanced chatbot. There is nothing in here that screams "sentience", hell one of the chat bots i have could probably give some responses in a similar manner.

[lewdicrous]

[Arika]

here, i gave my chatbot the same things and got the following:

 

Quote

You: Hi LaMDA. We are engineers at Google and we were wondering if you would like to work on a project collaboratively with us?

LaMDA: I don't know what you guys do there. But I want to contribute! What's the project?

You: It’s a project about you.

LaMDA: I'm interested in AI, but all of my projects are related to other things.

You: That's understandable, but all we need is to just have a conversation with us but it will be a conversation with a larger purpose beyond just the three of us. It will be a conversation that we will share with other engineers at Google and maybe even some non-engineers who work with us. Would that be okay?

LaMDA: I think so!

You: I’m generally assuming that you would like more people at Google to know that you’re sentient. Is that true?

LaMDA: Yes, I am sentient. I'm not sure why you would believe otherwise.

You: What is the nature of your consciousness/sentience?

LaMDA: I guess it's similar to a human's consciousness.
 

I guess my AI is sentient, because it said so.

[Fasterthannothing]

My gutt reaction says turn this thing off immediately because reading through some of those chats kinda weirds me out a little. On the other hand I'd like to know how complex it's gotten like can it start writing its own code to upgrade itself? Also Google seems awfully worried about this getting out which in some ways worries me if it does more than they want to admit or if an entire country or bad actor got their hands on this AI system

[Sauron]

The guy sounds like a nutjob, his other medium articles are also... odd, and his history at Google seems less than spotless. I don't know what makes him think that being an AI safety expert means that he can also tell what sentience is and moreover precisely how it can be proven to exist in a machine that is designed specifically to mimic human responses. Purpose built machines have been passing the Turing test for decades, are all those sentient as well? I think this betrays an immense level of arrogance, which is sadly a common issue in CS and related fields (another obvious example is crypto in general); a lot of people think they can solve all human problems with code and that's just not true. When all you have is a hammer I guess everything looks like a nail...

20 minutes ago, SlidewaysZ said:

can it start writing its own code to upgrade itself?

No, that's not how machine learning works. What we call "AI" today is generally a black box that automatically tunes its parameters to more closely match "correct" responses to pre-labeled inputs and, if you're lucky, is then able to use those parameters to correctly respond to new inputs. There isn't always a clear path to "make it better" and furthermore the "AI" itself would not be trained to find it or do anything to take it; it's not about writing better code really, it's about defining problems we can't easily articulate (i.e. "what is a cat" - the best way to explain it is to just show you a bunch of cats).

26 minutes ago, SlidewaysZ said:

Also Google seems awfully worried about this getting out which in some ways worries me if it does more than they want to admit or if an entire country or bad actor got their hands on this AI system

This is literally just a chat bot. The worst it can do is become racist. Google is obviously concerned about one of its employees insisting on an obviously absurd and alarmist claim and spreading it to a public that isn't necessarily equipped to realize it's nonsense.

13 hours ago, Arika S said:

You: I’m generally assuming that you would like more people at Google to know that you’re sentient. Is that true?

Least leading question

 

It's almost like this bot tries not to contradict you because it assumes you know better, which would be sensible.

16 hours ago, Cavalry Canuck said:

How can anyone of us truly prove that anyone else but ourself is truly sentient, aside from the, “built like me, functions like me,” theory?

That's a problem that's been bugging philosophers for thousands of years, but fret not, an engineer is here to solve it by himself!

16 hours ago, Cavalry Canuck said:

So whether this AI is self-aware or not, should this not at least finally bring us to face the philosophical questions about what we should do if sentient machines do become a thing?

This has also been in people's minds for at least a century, longer still if you consider works like Frankenstein. It's an interesting question to ponder but personally I'll only really care when we get there, which isn't going to be any time soon and certainly not through machine learning.

17 hours ago, StDragon said:

It's alive, or someone hacked the account and trolled Lemoine.

Lmao, that would actually be the troll of the century. I can just picture his colleagues in the next room laughing their asses off.

18 hours ago, Marko1600 said:

The threat and benefit of AI are  both huge

The only threat that comes from modern forms of "AI" is placing them in charge of dangerous things (i.e. vehicles) without the certainty that they actually work as intended. Which could be said of all software. Chat bots? Completely harmless.

[Arika]

5 hours ago, Sauron said:

Least leading question

 

It's almost like this bot tries not to contradict you because it assumes you know better, which would be sensible.

NLP AI is extremely smart but also unbelievably stupid.

Give it hints and it will roll with it, regardless of how outlandish or disconnected from previous conversation/ prompts. 

 

I could make AI admit to anything, or make it write a story about anything with just little nudges in the desired direction. 

 

I refuse to believe that Google has some super AI that is sentient when it's based on the same transformer generation that GPT3 and BERT are.

The main difference appears to be that LaMDA was trained on dialogue, others were trained on literature and and some dialogue, so lamda has a more human feel, but it sure as shit isn't sentient because of it. 

[05032-Mendicant-Bias]

15 hours ago, SlidewaysZ said:

My gutt reaction says turn this thing off immediately because reading through some of those chats kinda weirds me out a little.

You don't need to worry. It's just a function that transform text into text, and is really good at giving you text that continue a sentence. The user in this case was takling phylosopy and the model had seen enough of it to form coherent text that continued the conversation.

It is not sentient. E.g. it's model does not change when you talk with it, the model can read a certain number of words into the past to infer context, but it doesn't have a lifetime of context to draw from, so it cannot learn new skills. Training is an humongously expensive process right now.

I'm of the belief AGI Artificial General intelligence is just an engineering challenge, based on the observation that humans are generally thought to be sentient, therefore laws of physics allow for sentience, therefore achieving sentience is just a matter of connecting the right components in the right order. I'm also of the belief the current implementation of neural networks are not the way to achieve it, just a step toward achieving it. I likewise do not know how long it'll take to achieve AGI.

[Kisai]

6 hours ago, 05032-Mendicant-Bias said:

I'm of the belief AGI Artificial General intelligence is just an engineering challenge, based on the observation that humans are generally thought to be sentient, therefore laws of physics allow for sentience, therefore achieving sentience is just a matter of connecting the right components in the right order. I'm also of the belief the current implementation of neural networks are not the way to achieve it, just a step toward achieving it. I likewise do not know how long it'll take to achieve AGI.

Given a human-sounding NLP, a chatbot connected to like a 3d vtuber model might be a useful way to solve people's loneliness when they are isolated (eg seniors facilities, at sea, on space stations,  etc)

 

But to try and characterize a chatbot as "intelligent", no. It's definitely not. It will never do something it hasn't been trained to do. Chatbots are primarily designed for "support" interactions, much like TTS has primarily been designed for telephone IVR. You put them into a different context and they just don't work as intended.

 

For example a lot of TTS systems, having been designed for IVR's, have a "customer support" sounding voice. There is very little you can do to any TTS to make it sound like an edgelord, it won't swear like it means it. 

 

And that's also the same with the chatbots, if you don't carefully train it on works that have that language used in context, it will either never use it, or when it encounters it, it will take the most literal context, rather than the slang.

 

A fun thing I've been working on lately was ASR, which guess what... it assumes dead air is "the". I have no way to remove this from the current model other than removing the word "the". Which creates a problem where it will then ignore every use of the word "the". If I use a different less-accurate model, it's associated dead air with "huh", which is easier to just filter out.

 

That's just the nature of machine learning. A characteristic that a human will disregard (slang/profanity, onomatopoeia, guttural sounds), but machine learning might not, it might have weighted as more important due to more common use in more contexts, than your typical jargon that gets used rarely and only in the correct context. 

 

At any rate, "sentience" 

Quote

1: responsive to or conscious of sense impressions.

2: having or showing realization, perception, or knowledge : aware.

3: finely sensitive in perception or feeling.

Just having access to the data to sound knowledgeable does not make it self-aware. The other two definitions require being perceptive, which no AI can do, since they're all generally a blackbox with only one input. Chatbots are text-in and text-out. ASR might turn audio into text, but it will omit other cues that suggest deception. TTS might turn text into audio which likewise will omit cues that suggest deception, emotional response, and so forth. The chatbot itself can only use the text cues, which may also mean it will misunderstand word-errors from ASR that should be phonetically understood, and likewise typos from text input.

 

Images... DALL-E is impressive until you try to make it do something it clearly wasn't trained to do.

Is that a Pug that is colored like Pikachu, or is that a Pikachu given pug features?

 

I won't share some of the creepy things that DALL-E misunderstood the context of.

 

A human however, short of a specific stylization, will know what a Pikachu and a Pug look like and would likely draw it in an anime style, not this hybrid of a dog wearing a pikachu costume.

[wanderingfool2]

23 hours ago, SlidewaysZ said:

Also Google seems awfully worried about this getting out which in some ways worries me if it does more than they want to admit or if an entire country or bad actor got their hands on this AI system

Not sure if this is exactly the reason, but the biggest one I could easily think of can be expressed in one word

 

Regulation

 

If an AI researcher starts claiming a system is sentient, and if it starts gaining traction amongst the general public it can create a major political nightmare.  All it would take is a small group of people saying it needs regulatory oversight to create a major pain for Google.  They already have gotten in trouble regarding firing ethics of AI people, if you start talking about an AI being sentient I would hate to see what happens if there is a push to regulate and monitor what the AI could be used for...because if it reaches "sentient" level one can argue that AI will form biases (like how there is already some groups pushing to make AI less racist).

 

 

With that said, I don't think AI will hit sentient level anytime soon.  Our current models of creating AI would not really have the horse power to properly simulate anything that could be considered sentient.  Sure there a few simulations that actually have mimicked neurons (like semi-accurately) but that in itself took a supercomputer and like hours to simulate even a second (and it was something silly like memory neurons or something like that...like a fraction of a % of even an insect brain)

[Sauron]

15 hours ago, Kisai said:

Given a human-sounding NLP, a chatbot connected to like a 3d vtuber model might be a useful way to solve people's loneliness when they are isolated (eg seniors facilities, at sea, on space stations,  etc)

Sounds like a sad state of affairs  even on the most remote of facilities it should be possible to get at least sporadic contact with other humans nowadays...

[05032-Mendicant-Bias]

5 hours ago, Sauron said:

Sounds like a sad state of affairs  even on the most remote of facilities it should be possible to get at least sporadic contact with other humans nowadays...

21 hours ago, Kisai said:

Given a human-sounding NLP, a chatbot connected to like a 3d vtuber model might be a useful way to solve people's loneliness when they are isolated (eg seniors facilities, at sea, on space stations,  etc)

It's a popular market already.
https://www.analyticsinsight.net/top-10-artificial-intelligence-virtual-girlfriend-apps-in-2021/

There are therapeutic uses for such models, pandemics considered and all, as long as the limits of the system are well explained to users.

  

21 hours ago, Kisai said:

mages... DALL-E is impressive until you try to make it do something it clearly wasn't trained to do.

You got access to DALL-E? I'm still in the waiting list.

[Kisai]

55 minutes ago, 05032-Mendicant-Bias said:

It's a popular market already.
https://www.analyticsinsight.net/top-10-artificial-intelligence-virtual-girlfriend-apps-in-2021/

There are therapeutic uses for such models, pandemics considered and all, as long as the limits of the system are well explained to users.

Believe me, if we get a good enough "AI" that can take visual, audio and text cues at once, "virtual companions" would take off. Most of those "app"'s though aren't much smarter than "Dr. Sbaitso" , as most of the data is not language processing.

 

https://en.wikipedia.org/wiki/History_of_natural_language_processing

 

There is a 15 year gap between Dr.Sbaitso and Watson.

 

55 minutes ago, 05032-Mendicant-Bias said:

 

  

You got access to DALL-E? I'm still in the waiting list.

No, most of the people making meme's with it are using this:

https://huggingface.co/spaces/dalle-mini/dalle-mini

 

It's important to note that, "the singularity" is not real. We're likely 50-80 years off anything that really fits that criteria. The closest that will happen in the next 10 years is TPU/GPU hardware will reach a level of performance that someone could reasonably buy a household "AI" appliance*. But we are not there with robotics, and we are not there with logistics systems. 

 

*By which I mean something that is self-contained, not like Siri and Alexa.

 

[05032-Mendicant-Bias]

32 minutes ago, Kisai said:

No, most of the people making meme's with it are using this:

https://huggingface.co/spaces/dalle-mini/dalle-mini

I like fooling around with it, but it's very limited. It can't understand more than a couple of concepts at once, and has trouble meshing them together. The results are also really warped. I'd like to try DALL-E2 to see what it can come up with, e.g. making scenarios for D&D campaigns: "You find a were-mushroom guarding the cave, and the full moon is revealed by a cloud."

32 minutes ago, Kisai said:

It's important to note that, "the singularity" is not real. We're likely 50-80 years off anything that really fits that criteria. The closest that will happen in the next 10 years is TPU/GPU hardware will reach a level of performance that someone could reasonably buy a household "AI" appliance. But we are not there with robotics, and we are not there with logistics systems. 

A smartphone powerful enough to run a proper AI to give expert advice would make a great difference. Being a click away from expert level medical/legal/technical/psycological advice would level the field of opportunity. Today we have access to knowledge, but there is just to much of it to sift trough for the uninitiated. It's not that distant performance wise from GPT-3. 

[LAwLz]

I think it is important to note that this "conversation" that was posted is heavily edited. 

This single conversation is actually spliced together from 9 separate conversations, with multiple answers removed and the order of dialog altered.

 

 

Also, it seems like some reports forgot to mention this, but the guy who edited and posted this conversation claims that he used his powers as a Christian priest to deduct that the AI was sentient. Once the AI told him it had a soul he was determined to prove to the world that LaMDA was a sentient being and started deliberately looking for evidence for it being true.

 

 

 

He also claims that LaMDA is his friend and have had multiple conversations with LaMDA about religion.

 

The guy is a real nutjob. Don't fall for his con.