Hacker News Reader

Comment by theGnuMe | Hacker News Reader

People like to ascribe magic and mysticism to things they don't understand. Chat bots are similar. No way it is sentient.

View on HN

People are also complete assholes and like to remove sentience from things if they find it inconvenient in any way.

Humans have removed 'humanhood' from other humans based on things like race, religion, wealth, and whatever metric we find convenient at the time. We seem to be a very poor judge on this in my point of view.

The second paragraph that you wrote is fantastic. The first paragraph is unnecessarily harsh, in my opinion. You don’t need to attack anyone here-we are all just trying to contribute to the conversation. That is all. Be well.

I'm not removing sentience from anything; AI chat bots are not sentient.

Taking your argument to the furthest extent:

Humans once dehumanized others, therefore humans should never judge what is human or not.

My pet rock looks very sentient to me. Anyone who tells me otherwise is just an asshole who is on the wrong side of history. Better extend human rights to all rocks.

tl;dr yes humans have been wrong in the past, but that isn’t any excuse to never try and explain anything and always believe everything is sentient the second someone claims it is.

Yes, I’m also familiar with that Clarke quote. I do understand how these models work on a mechanical level, but we certainly have not unpacked all the emergent behavior. What exactly is your evidence that there couldn’t be more going on?

Once we have the ability to X-ray the black box and we only find simple conditional correlations behind an existential conversation, then I would agree with you, but we haven’t done that yet.

Read up on conditional probability.

I already said it might be no more than conditional correlations. That does not account for an emergent combination of conditionals that could implement some undiscovered algorithms of sentience.

If our brains are no more than computers and our own consciousness is software, then there exists some algorithm or combination of that gives rise to sentience. If these models arrived at this special algorithm during their training, much like the optimization of evolution arrived at the same, then we may have created something sentient.

But the fact that our own sentience is a mystery means that there’s not a whole lot we can say mechanically about these LLM other than talk about their behavior and whether it’s convincing.

>If our brains are no more than computers and our own consciousness is software, then there exists some algorithm or combination of that gives rise to sentience. If these models arrived at this special algorithm during their training, much like the optimization of evolution arrived at the same, then we may have created something sentient.

Joscha Bach gives a pretty good explanation of the algorithm we follow through the lens of Control Theory, and Stephen Wolfram has a pretty amazing Theory of Everything that explains how it can be arrived at.

Joscha Bach: Nature of Reality, Dreams, and Consciousness https://www.youtube.com/watch?v=rIpUf-Vy2JA

Stephen Wolfram: Complexity and the Fabric of Reality https://www.youtube.com/watch?v=4-SGpEInX_c

But what about computational irreducibility that arises from complexity?

What kind of evidence would you need to convince yourself of an AI being sentient?

Pass a Turing test with a competent judge. ChatGPT or any other bot won't be able to pass it.

So if a human fails a Turing test, we can remove their rights of personhood?

I'm just checking your logic to see if a two way communicative process, or if humans are granted magic values here?

For example, great apes look somewhat like humans but can't pass a Turing test. Should we give them same rights as humans too? It works other way too.

Bing chat would do better at passing a Turing test than many real people I suspect.

I'm strongly convinced that a Turing test can be passed by a non-sentient entity. Being sentient and smart is sufficient to fool a judge, but not necessary, people can be fooled by stupid illusions.

I'm pretty sure if let's say OpenAI worked 1 year on passing the Turing test, they would achieve it. It seems extremely close. Current LLMs are trained to be useful, not to trick humans.

Sentience is the ability to be able to experience feelings. In this regard, chatbots are not really convincing. Kind of like a psychopath can describe a feeling, but not actually feel it.

This is a tough question to answer though, but there's this kind of human intuition we have where most of the time we somehow know if someone is feigning emotion. Emotion that comes from chat bots can be explained with, "well it's just a chat bot". Something needs to happen in order for us to truly question that in order to squash our doubts about their sentience. Not sure what that will look like though.

This is the point I was making. We humans do have this emotional sensor we can use to probe the minds of others. For some people, this emotional detector started beeping when they talked to Bing. Is the sensor faulty? Maybe. But as it hits for more and more people that becomes it’s own problem.

> We humans do have this emotional sensor we can use to probe the minds of others.

Wait, what? I certainly don't have one. Most of us (not all! neurotypical experience isn't universal) have a somewhat effective skill, continuously trained since childhood, to guess at the emotions of others based on various visual and behavioral cues, but we also know that these visible cues can be faked by (for example) a skilled actor, as the "sensor" has zero insight to the actual emotional experience.

If I make the assumption that someone else (e.g. you) function the same as I do, then I can reason that "your behavior X implies your emotional experience Y, because it does for me"; however, if we can't make that assumption, or (as is the case for non-human minds) we know that the process is substantially different - then there is literally zero basis for any trustworthy reasoning whatsoever about that internal emotional state.

I just typed a colon and a parenthesis. Your brain interpreted it as a smile on a human face.

You are equally mistaken in the use of your emotional sensor. Our brain fills in a LOT of gaps based on OUR training.

>Sentience is the ability to be able to experience feelings. In this regard, chatbots are not really convincing. Kind of like a psychopath can describe a feeling, but not actually feel it.

So, psychopaths aren't sentient?

People conflate sentience with sapience.

I see. Well, TIL. Tbh the more I look at the resolution people (myself included) have for these concepts in general, it really does seem like it's so low as to be virtually meaningless. That even holds for the those individuals or groups of individuals who coined the terms in the first place.

To that end, I find myself gravitating more and more towards higher level abstractions that have multiple existence proofs (e.g. Control Theory) and just looking at everything through lenses like that.

Nobody has established a reasonable answer to this question; we can't even demonstrate self-sentience or sentience of other humans.

Step 1) The ability to distinguish real from unreal and the ability to understand truth as, partially or primarily, direct sensory input which is retained and can be acted upon.

Lock 2 AIs that haven't been programmed with language in a room, given time do they develop the ability to communicate and express ideas, desires, needs to each other?

There have been several instances of language evolving again, deaf people creating sign language, etc. A sentient intelligence should be driven to understand concepts and convey them, probably over multiple mediums.

That's why the turning test is a good start, communication is a key aspect of sentience but its just round one.

This already happened.

https://www.huffpost.com/entry/facebook-shuts-down-ai-robot-...

No, in the linked example it is mutating English into a different language. GP's challenge was to create a form of communication unprompted, and without precedence.

One precondition for sentience would be some kind of continuity. Does the AI operate separately from its tasks, or does it instantiate, take care of some request, and then end, as a ChatBot process does? Is it actively processing while waiting for input, or is it only alive and active when it is processing some input and preparing some output?

I think the lack of this continuity is why people who feel they know how ChatGPT works feel confident it is not sentient. Sentience isn't simply responding to stimuli -- although we don't really understand consciousness, it clearly involves some kind of continuing sense of self, some kind of "dial tone" that provides the basic sense of continuing existence, and that isn't there in a bot that simply takes input and builds a response based on rules and data.

Reply on news.ycombinator.com