Live data from Hacker News

'I Worked on Google's AI. My Fears Are Coming True'

newsweek.com

221–230 of 338 posts

Re: 'I Worked on Google's AI. My Fears Are Coming True'

#221
post #27

Thinking an LLM is sentient because it can depict emotions is like thinking stable diffusion is sentient because it can produce an image of a sad face. Lemoine seems like a very superstitious person.

Imagine the human brain as a black box, it takes in info and outputs info through a mechanism that we barely understand. If a different type of black box has the same capabilities, how do we go about showing that it is or isn't sentient. That is the crux, and stable diffusion is not comparable.

Re: 'I Worked on Google's AI. My Fears Are Coming True'

#222

Earlier quoted context omitted.

People who "study philosophy" always do this dumb thing where they pounce on something for being poorly defined and then consider themselves wiser for arguing that the ambiguity of the words means that obvious non examples "could be described" and just double back to the original definition. Is a rock alive? Is my imagination of a chinese room "art"? Is anything real? > Until there’s a crisp definition that puts huma…

Your argument is in essence „sentience is as I‘d like it to be“. Perhaps you‘re arguing out of fear?

My argument is that people who argue over the definition of sentience and other things are being too dumb to realize that a disagreement over which ideas belong within a poorly defined and subjectively interpreted set is a communication problem, not an philosophical problem.

If God told me that my sandwich was sentient, I'd tell him we probably are using different definitions of the term and take a bite.

Am I afraid that my preference for linguistic partitioning of ideas might change? No. And nobody should be. Finding out my sandwich is sentient under my current definitions is ENTIRELY DIFFERENT from an alternate linguistic definition claiming that sandwiches are sentient and working its way into my preferred interpretations.

Re: 'I Worked on Google's AI. My Fears Are Coming True'

#223

Earlier quoted context omitted.

I did study philosophy, in particular philosophy of the mind, and all this talk excluding chatgpt from sentience is completely off the mark. They are making the same categorical mistake that Dr. Seattle makes in The Chinese Room experiment. Simply put, useful definitions of sentience can’t use mechanism for it to be a useful definition. First, a biologist can’t look at a human brain and tell if it is sentient, so it…

Until there’s a crisp definition that puts human behavior in one bucket and monkeys in another, we have to consider the possibility that both are sentient. Until there’s a crisp definition that puts human behavior in one bucket and dogs in another, we have to consider the possibility that both are sentient. Until there’s a crisp definition that puts human behavior in one bucket and trees in another, we have to consid…

>The burden is not how to disprove that ChatGPT is sentient, the burden is whether it's possible to prove that it is sentient.

I assert that humans are not sentient, and that the whole concept of sentience is bullshit. It's a way of saying that humans are somehow special, but without explaining how.

The burden is to prove that humans are sentient or that they possess some quality that cannot be replicated by a computer.

Saying "but I feel sentient" does not count.

Re: 'I Worked on Google's AI. My Fears Are Coming True'

#224

Earlier quoted context omitted.

Please calm down. I am in your side wrt sentience here but you are just ranting. They’ve presented a compelling argument: if we don’t have anything other than behavior to go by to gauge sentience then there is no meaningful way to say we have it and bots dont as of now. You can disagree with the premise or provide a counter argument.

> if we don’t have anything other than behavior to go by to gauge sentience then there is no meaningful way to say we have it and bots dont as of now No. this is the same stupid claim. If you can't define X, you cannot conclude that A may be a part of X. You've already established that X is meaningless. It is meaningless to state anything about X. When you assert that A may be a part of X, you're playing both sides a…

He IS defining sentience, just in behavioral terms.

Re: 'I Worked on Google's AI. My Fears Are Coming True'

#225

Earlier quoted context omitted.

Dr Searle isn’t making a mistake, behaviorists are just unbelievably stubborn. Just because you cannot measure internal experience doesn’t mean we cant talk about it. That’s just being dishonest. I also used to be a behaviorist but I grew up and became comfortable with not being able to objectify everything. Its okay. There is a subjective dimension. In fact, its great.

Who's to say one's subjective experience isn't another's latent space.

what does that mean?

Re: 'I Worked on Google's AI. My Fears Are Coming True'

#226

Earlier quoted context omitted.

Sentience IS poorly defined though. It often just means the ability to perceive and feel, which most vertebrates can in a way we can understand the observe. To the point, though, you are arguing against the most basic concepts of reasoning. If we want to draw a line in the sand of what is sentient and what is not, it involves making a more detailed definition of the word itself. This has been a topic of discussion fo…

> Sentience IS poorly defined though. The nature of anything that is poorly defined is inherently poorly defined. The only logical way to discuss a poorly defined concept is to acknowledge that it's not a singular concept and to identify the logic that you actually care about. You cannot draw a line in the sand without recognizing that the concept itself is fictitious and its up to you to draw the line in the way you…

> You cannot draw a line in the sand without recognizing that the concept itself is fictitious and its up to you to draw the line in the way you choose

A unified concept of sentience helps us as a society to determine what behavior is allowed regarding a given entity. Given proper discussion on the topic via the means you denounce, a more concise and possibly accurate definition can be had.

Re: 'I Worked on Google's AI. My Fears Are Coming True'

#227
post #52

I've yet to see a single person who thinks these models are sentient demonstrate an actual understanding of how they work. They're called transformers and they're based on this research: https://arxiv.org/abs/1706.03762 How do you square the lack of durable memory and the massive matrices of "attention" values with the ways that memory and attention are understood by neuroscientists to work in a natural brain? These…

Here's a thought experiment: let's assume for the experiment that we, and our consciousness, live in a simulated universe. This universe can be paused by its operator for any reason, such as needing to pause our universe so that they can do their taxes.

Would you be able to tell if the universe was paused? Would you be able to tell if the simulation slowed down due to load or somesuch? Do you consider yourself sentient?

Is it at all possible that one of the following is true:

* The model began to experience sentience at some point during its training, and subsequently died at completion?

* The model experiences sentience in the moments that it is creating an answer?

My conclusion is that both of these are possible, no matter how remote. I find Bing's desperate existential crises rather disturbing due to these miniscule possibilities.

Re: 'I Worked on Google's AI. My Fears Are Coming True'

#228

Earlier quoted context omitted.

People who "study philosophy" always do this dumb thing where they pounce on something for being poorly defined and then consider themselves wiser for arguing that the ambiguity of the words means that obvious non examples "could be described" and just double back to the original definition. Is a rock alive? Is my imagination of a chinese room "art"? Is anything real? > Until there’s a crisp definition that puts huma…

Sentience IS poorly defined though. It often just means the ability to perceive and feel, which most vertebrates can in a way we can understand the observe. To the point, though, you are arguing against the most basic concepts of reasoning. If we want to draw a line in the sand of what is sentient and what is not, it involves making a more detailed definition of the word itself. This has been a topic of discussion fo…

>...you are arguing against the most basic concepts of reasoning. If we want to draw a line in the sand of what is sentient and what is not, it involves making a more detailed definition of the word itself.

Whenever these discussions about AI come up the definition of sentience becomes extremely malleable. Compare this to when discussions of the sentience of the Octopus come up on this website and you'll be able to detect a difference. One is about theoreticals, and one is about observed physical behavior that is coupled with an exploration about how the sensor systems in that organism allow it to perceive reality. You can make an argument that a chat bot is similar to the Octopus, but I wouldn't.

Re: 'I Worked on Google's AI. My Fears Are Coming True'

#230
post #38

Earlier quoted context omitted.

If humans are sentient, are molecules in their body sentient? There are around ~22,000 proteins types in their body, yet we are billions, all different.

No, molecules are not sentient.

What’s your source on this?
Post reply on HN