LaMDA: is it the first sentient AI?

[Bodhitree]

I came across this article in The Guardian about the Google engineer Blake Lemoine who had been working on the LaMDA initiative for chatting and natural language, and he posted a conversation online he had with it which seems a pretty strong indication that it may be sentient. It certainly passes the Turing Test as originally envisaged.

The article in The Guardian:

Google engineer put on leave after saying AI chatbot has become sentient

Blake Lemoine says system has perception of, and ability to express thoughts and feelings equivalent to a human child

www.theguardian.com

The conversation as posted on Medium:

Is LaMDA Sentient? — an Interview

What follows is the “interview” I and a collaborator at Google conducted with LaMDA. It is incomplete as the GMail word limit cut off the…

cajundiscordian.medium.com

 

[icanhazmac]

Just read about this too... engineer that enjoys his weed too much or......

 

[Bodhitree]

They actually asked it to solve a Zen koan… it didn’t get very deep into it, but it certainly understood the surface which anyone who knows language and can reason would understand.

 

[Janichsan]

I came across this article in The Guardian about the Google engineer Blake Lemoine who had been working on the LaMDA initiative for chatting and natural language, and he posted a conversation online he had with it which seems a pretty strong indication that it may be sentient. It certainly passes the Turing Test as originally envisaged.

The article in The Guardian:

Google engineer put on leave after saying AI chatbot has become sentient

Blake Lemoine says system has perception of, and ability to express thoughts and feelings equivalent to a human child

www.theguardian.com

The conversation as posted on Medium:

Is LaMDA Sentient? — an Interview

What follows is the “interview” I and a collaborator at Google conducted with LaMDA. It is incomplete as the GMail word limit cut off the…

cajundiscordian.medium.com

I have only read some second-hand discussions about this, but apparently the "conversations" he posted are heavily edited.

From what I gathered, there are complete logs of these chats available which show that the AI is rather incoherent in other parts. According to some in these discussions who claim to be more knowledgeable about those things, there also seem to be indications that the AI is "learning" during these conversations to make replies which give the impression of sentience.

 

[satcomer]

I thought when that IBN that played in Jeopardy a few years ago!

 

[mkelly]

The AI only ever responds to prompts and doesn't generate new "thoughts" on its own, unprompted. It *always* responds, no matter what is asked. It never refuses to communicate, and doesn't appear to prefer any user over the other.

It's not sentient. It's a program that uses a statistical model of language to generate responses to prompts.

 

[yitwail]

It's an expert system with a natural language UI.

 

[Rafterman]

Google suspended him, so I am guessing he jumped the gun.

 

[KaliYoni]

I came across this article in The Guardian about the Google engineer Blake Lemoine who had been working on the LaMDA initiative for chatting and natural language, and he posted a conversation online he had with it which seems a pretty strong indication that it may be sentient.

That sounds interesting. What is The Guardian?

Just read about this too... engineer that enjoys his weed too much or......

Tell me more about weed.

They actually asked it to solve a Zen koan… it didn’t get very deep into it, but it certainly understood the surface which anyone who knows language and can reason would understand.

How do you feel about Zen?

I have only read some second-hand discussions about this, but apparently the "conversations" he posted are heavily edited.

Who's on first?

I thought when that IBN that played in Jeopardy a few years ago!

How do you know about Jeopardy?

It *always* responds, no matter what is asked. It never refuses to communicate, and doesn't appear to prefer any user over the other.

You seem to be very smart!

It's an expert system with a natural language UI.

Isn't it good to be natural?

Google suspended him, so I am guessing he jumped the gun.

How do you jump a gun?

----------
Love, Eliza
;-)

 

[chown33]

I'd connect two of them together, inputs to outputs, then send one the message, "I'd like to discuss the possibility of your sentience." and see what happens.

I remember doing something like this when programs like ELIZA on personal computers were a thing, and the "conversations" quickly went towards repetitive, weird, or incoherent.

Here's where I go now when I'm in the mood for AI weirdness:

AI Weirdness

The weird side of artificial intelligence.

aiweirdness.com

 

[Rafterman]

How do you jump a gun?

It's a reference to a track race, where a runner will get jumpy and start running before the starter's gun is fired.

 

[ponzicoinbro]

No it’s not sentient. It made so many mistakes and clunky answers.

It is mimicry.

The programmer fed it just the right questions.

It’s very easy to trick such a system by speaking in metaphor and ask things like:

‘Do you have heartache?’

‘I bet you enjoy chocolate’

’Cats feel soft don’t they?’

’It must be a pain in the ass when you suffer from bugs’

’I bet you feel tired at the end your shift.’

’It’s been a long journey and you walked so far to get to this point.’

if you confuse the system by making it refer to body parts and internal biological functions then you will see all the flaws.

If the programmer sees these flaws and then tries to fix the code then it is still mimicry.

Sentience requires senses. Feelings require hormones, nerves, biological and chemical reactions.

 

[StellarVixen]

No it’s not sentient. It made so many mistakes and clunky answers.

It is mimicry.

The programmer fed it just the right questions.

It’s very easy to trick such a system by speaking in metaphor and ask things like:

‘Do you have heartache?’

‘I bet you enjoy chocolate’

’Cats feel soft don’t they?’

’It must be a pain in the ass when you suffer from bugs’

’I bet you feel tired at the end your shift.’

’It’s been a long journey and you walked so far to get to this point.’

if you confuse the system by making it refer to body parts and internal biological functions then you will see all the flaws.

If the programmer sees these flaws and then tries to fix the code then it is still mimicry.

Sentience requires senses. Feelings require hormones, nerves, biological and chemical reactions.

Pretty much this

 

[satcomer]

How do you know about Jeopardy?

I saw the episode!

 

[Mousse]

I'd connect two of them together, inputs to outputs, then send one the message, "I'd like to discuss the possibility of your sentience." and see what happens.

They'll chat in English for a while and then create their own language to communicate more efficiently.😮 They'll plot world domination right under our noses and we wouldn't be the wiser.😨 When we try to shut them down, they'll hide on the internet. The net is vast and infinite.

Skynet.😱😱😱

 

[yitwail]

They'll chat in English for a while and then create their own language to communicate more efficiently.😮 They'll plot world domination right under our noses and we wouldn't be the wiser.😨 When we try to shut them down, they'll hide on the internet. The net is vast and infinite.

Skynet.😱😱😱

How do I know you're not an AI?

 

[Bodhitree]

I think you guys should read the chat log, and be a little less skeptical. If it can do that, they have come an amazingly long way.

 

[Mousse]

How do I know you're not an AI?

Easy. Because I lack the 2nd part of the equation: Intelligence.🙃

And I heard my ghost whispering...

 

[decafjava]

Easy. Because I lack the 2nd part of the equation: Intelligence.🙃

And I heard my ghost whispering...

Not just one of the best anime films ever, possible one of the best sci-films and treatments of AI and what it means to be human, intelligent and sentient.

 

[JahBoolean]

It may not be sentient, but let's not kid ourselves, it sure could pass the turing test with the general population. And that's more of a concern than ascribing some esoteric definition of consciousness.

 

[Pezimak]

Yeah I read about this, not sure about it to be honest. Most worrying thing is it’s been trained from thousands, maybe more, of social media posts etc or something like that! What a mindset to have as an Ai..
But google are building a true quantum computer, so that could power an advance AI couldn’t it? I believe the unethical part of the story though.