Hacker News Reader

Comment by Houshalter | Hacker News Reader

>Isn't there quite a bit of wasted energy and skull space that could be better used for more survival oriented if-statements. All of our behavior seems possible without the associated feelings.

Intelligence isn't about a giant list of arbitrary if then statements though. Think of artificial neural networks. They learn behaviors from experience and reinforcement. They take actions they predict will lead to reward.

How is that fundamentally different than what humans do? I mean our brains algorithms are probably much more complex, but the principles should be the same. Shocking a neural net with negative reward, and it will strongly desire to avoid that again. How is that different than pain?

Add on a bunch of hard wired goals and behaviors more complicated than "pain" and "reward", and you have yourself a conscious being. Not fundamentally different than a human.

View on HN

You do have subjective experience, don't you? I mean, you don't just process external stimuli, combine them with an internal state, and generate output signals, like a program would – you can actually "feel" things and see them with "your inner eye"?

I'm not being sarcastic here. Maybe there are people that don't have subjective experience, just like there are people that can hear colors.

> I mean, you don't just process external stimuli, combine them with an internal state, and generate output signals, like a program would – you can actually "feel" things and see them with "your inner eye"?

You're assuming the conclusion that consciousness isn't just processing external stimuli with internal state to generate output signals, just a particular form of it.

> You're assuming the conclusion that consciousness isn't just processing external stimuli with internal state to generate output signals, just a particular form of it.

Yes, that is my assumption, because that is what I observe (only in my own mind, of course – I don't know how to observe it in others). I'm not alone, many (most? all?) people are claiming to observe that.

Except:

1. It isn't a valid observation, because you don't actually observe that consciousness is not part of the process of processing external stimuli. At best, you can observe that consciousness exists.

2. It isn't a valid conclusion even if it were a valid observation. Or do you also conclude that water magically breaks pencils and reconstitutes them when they are removed [1]?

[1] http://cdn.c.photoshelter.com/img-get/I0000KUS4vG.ZoqQ/s/860...

I observe that conciousness exists, and I can conceive of no way to logically deduce its occurrence from any physical processes, nor any way to consider them as the same thing.

To me, experience itself is fundamentally a separate thing from any process that might cause it to occur, and I don't know how to conceive of it any other way.

> I can conceive of no way to logically deduce its occurrence from any physical processes, nor any way to consider them as the same thing.

That you cannot conceive of it does not make it inconceivable. In fact, scientific theories [1] exist to explain exactly the functional purpose of consciousness.

[1] http://journal.frontiersin.org/article/10.3389/fpsyg.2015.00...

>That you cannot conceive of it does not make it inconceivable.

True, but I find it sufficiently difficult to conceive of that I'm not convinced anyone else can. If you claim to be able to, can you explain it to me?

To me, saying experience "is" brain processing is like saying the image on the screen at a cinema "is" projector operation. One may cause the other, but they're still separate things.

> scientific theories [1] exist to explain exactly the functional purpose of consciousness.

But that doesn't explain the mechanism for conciousness to occur, it just explains why it might exist.

> But that doesn't explain the mechanism for conciousness to occur, it just explains why it might exist.

It does explain the mechanism actually. The mind builds representations for predictive purposes, it builds a representation of attention because that increases predictive power in achieving goals, a representation of attention is consciousness.

Certainly it's not sufficiently detailed such that you can build one, but that's not really needed to make it conceivable.

That goes some way to explaining the mechanics, but there's still a vital last step needed to take it from a simulation process to an actual experience of that simulation.

You seem to be describing that step as "representation of attention", but I'm not sure what that's supposed to mean.

Did you read the paper? Take a look at Figure 1 and contrast 1a and 1b. The combination of perceptions you're paying attention to + focusing on a model of yourself paying attention to your perceptions yields subjectivity. Your brain shifting between the signals from your direct perceptions and signals from your model of yourself focusing on your perceptions is the perceptive difference we call subjective experience. I think the description there explains it quite well if that's unsatisfactory.

Certainly it's lacking in sufficient detail to build an artificial consciousness, but it's not lacking in empirical data and testable hypotheses. It's also a perfect example of how subjective experience could be an illusion that exists, and yet still convince us that it's in some sense fundamental.

> But that doesn't explain the mechanism for conciousness to occur, it just explains why it might exist.

Hmm, what if both, the physical mechanism and the subjective qualia are simply intertwined in a way, that can not be deduced any further?

Maybe it's a little like the relation between gravitational forces and curved space time. Both are basically the same thing. But if you're used to live in one system, it's hard to imagine how it feels to see things from the other perspective.

Still there may be a lot of value to the complementary theory. Much like the GTR opened up a whole new range of physical applications, a physical theory of consciousness may lead to computer systems that we will have to attribute being conscious the same way as we do with our fellow human beings - just because both will follow the same fundamental laws of consciousness.

The two seem fairly trivially separable though. I can imagine a thing that has the same physical mechanisms as me, but does not experience conciousness (i.e. a p-zombie), and I can imagine something with no physical mechanisms that does (i.e. a soul).

Now it's possible neither of those things can actually exist, but the fact that I can imagine them demonstrates the distinction I'm making.

> I can imagine a thing that has the same physical mechanisms as me, but does not experience conciousness (i.e. a p-zombie)

Hmm, really? How? To be honest, I can't. Honestly I think this is just a meaningless mind trick. The more I think about it, the less the notion of a p-zombie really makes sense to me. If a p-zombie existed, and I could never really tell the difference from the outside, this gives rise to many blind-alley-models of the world, for example solipsism: Maybe all others are just p-zombies and maybe I'm the only thing that really exists in the sphere of consciousness (=soul?). I can never know for sure. But if it's true, then there's not really much room for any further discussion.

So I have to assume that everything that behaves like being conscious in fact has a consciousness - even if it isn't even aware of its being conscious.

Of course I assume that other people are concious, but I am nonetheless capable of imagining a person who is not.

And the point isn't whether such a thing is physically possible in our universe, merely that being able to conceive of such a thing identifies the distinction we're drawing. There is a notion of experience separate from the physical process that may cause it, because for any such process, we can imagine a world where it does not actually experience anything.

Neural networks also have "inner eyes". From being able to "dream" images and sequences that haven't appeared in the training data, but look plausible.

Or being able to manipulate learned concepts. Like word2vec turns words into vectors that can be combined in complicated ways. E.g. words for colors will be different based on what the colors are associated with (e.g. red is associated with "danger", or green is associated with "nature", etc.)

Make this process much more complex, and intelligent, and then ask it what it "feels". It will probably give responses very similar to humans. It "feels" complicated internal experiences that it can't completely explain or understand.

It certainly wouldn't feel like it's "just combining stimuli with internal state to get output signals". Being an algorithm doesn't feel like you are an algorithm, from the inside.

> It will probably give responses very similar to humans.

Being able to mimic a human's behavior is neither a necessary nor a sufficient condition for having subjective experience.

> Being an algorithm doesn't feel like you are an algorithm, from the inside.

How do you know that? How do you know that an algorithm "feels" anything at all, the way humans do?

> Being able to mimic a human's behavior is neither a necessary nor a sufficient condition for having subjective experience.

I think that's not the point he tried to make. It's just that it happens that we're human and thus it's easier for us to relate to human-like experiences/feelings. An algorithm is not limited to that.

If we accept that our "feeling like something" is just an emergent phenomenon of our complex model of reality I see no logical reason why we shouldn't ascribe any sufficiently complex model of reality an inherent feeling. Due to our limited physical configuration we will never have access to all kinds of feelings. Much like we will never know how it feels like to be a bat.

>Being able to mimic a human's behavior is neither a necessary nor a sufficient condition for having subjective experience.

I am arguing that it is. Once you move past the "giant lookup table" idea to actual algorithms that explain human behavior. In fact once you find the smallest such algorithm, it must necessarily be an exact simulation of a human brain, and all of our internal experience.

>How do you know that an algorithm "feels" anything at all, the way humans do?

It must feel exactly the same way humans do if it produces exactly the same output.

> You do have subjective experience, don't you? I mean, you don't just process external stimuli, combine them with an internal state, and generate output signals, like a program would – you can actually "feel" things and see them with "your inner eye"?

Hold on. Did you just claim that people with aphantasia (ie: possibly an inability to feed causal models from someplace (probably the neocortex? IANANS, just work with some) back to the visual cortex and get the visual cortex's "portrayal" of the conditional simulation embodied in the model) have no subjective experience?

What if someone has no "inner eye", but does have "inner ears", an "inner tongue", "inner hands" and all the rest? Or do you group all possible "inner senses" used for imagining "what something is like" into "qualia" without separating them into distinct functionalities that are in bijection with the available physical senses?

Because you might have actually said something interesting here!

> Did you just claim that people with aphantasia [...] have no subjective experience?

I don't think so. I wasn't talking about imagination (as in: thinking of a cube and rotating it in your mind), but rather about experiencing the input from my physical eyes. It's as if there's someone inside my head that experiences and feels what I physically see. Of course, this doesn't help: the homunculus model is thoroughly debunked. [1]

> Or do you group all possible "inner senses" used for imagining "what something is like" into "qualia" without separating them into distinct functionalities that are in bijection with the available physical senses?

There are qualia like anxiety or restlessness that are not linked to physical senses, so that's a no on the bijection. I have no idea whether those "inner eyes, ears, etc." that link imagination to physical senses are related to qualia – it's so difficult to imagine myself into different minds. Besides, coming up with an objective, clear, non-recursive definition for subjective experience is way above my paygrade.

Another random thought that I just had: People with depression sometimes hurt themselves physically, because that's the only way they can still feel something (according to them). Maybe this is related to a dampened subjective experience, which allows only extreme inputs to be experienced?

[1] https://en.wikipedia.org/wiki/Homunculus_argument

>It's as if there's someone inside my head that experiences and feels what I physically see.

No, you only imagine that. And I'm being serious: people who aren't brought up being told that consciousness is like having a little homunculus in your head simply never feel "as if" there was one.

>There are qualia like anxiety or restlessness that are not linked to physical senses, so that's a no on the bijection.

I would normally call those emotions rather than qualia. Hmmm. Would you say the "qualia-ness" of things comes from the internal perception that the "feelings" are external to the person who feels, and impinge on them?

>Another random thought that I just had: People with depression sometimes hurt themselves physically, because that's the only way they can still feel something (according to them). Maybe this is related to a dampened subjective experience, which allows only extreme inputs to be experienced?

But dampened affect (dampened emotional signaling in the brain) is already a known and seemingly somewhat understood symptom of depression.

>people who aren't brought up being told that consciousness is like having a little homunculus in your head simply never feel "as if" there was one.

So you're saying people brought up differently are p-zombies? Because that's the only way I can interpret that.

Feeling like there's a homonculous is fundamentally what experiencing the world is like for me, and has been since long before I was able to understand complex philosophical questions about conciousness.

>So you're saying people brought up differently are p-zombies?

No, I'm saying that people brought up differently actually identify our selves with our actual, physical bodies, and don't imagine/visualize a tiny little figurine in our head. We have a different experience of the world, not no experience.

To me it seems weirder to say, "When I get bitten by a dog, the little homunculus gets a loud alarm ringing in his little room saying, 'BITTEN BY DOG!'" than "When I get bitten by a dog, its teeth feel sharp and painful."

> and see them with "your inner eye"?

Interesting point, not everyone has an inner eye, see aphantasia.

Isn't 'feeling something with your inner eye' a part of how 'processing external stimuli combined with my internal state' works?

> How is that fundamentally different than what humans do? I mean our brains algorithms are probably much more complex, but the principles should be the same. Shocking a neural net with negative reward, and it will strongly desire to avoid that again. How is that different than pain?

Because we feel the pain, why? Does neural net feel anything?

Is the fact that we feel pain a bug, a feature, or no different from a neural net that learns to avoid that negative reward via electrical impulses rather than electrochemical impulses as we do?

It's not a bug or a feature, the fact we feel or experience anything at all is the most fundamental aspect of our existence. If you think about it, subjective experience is the only thing that exists. All of physics stems from subjective experience. Physics is basically a systematic describtion of the patterns and regularities that we experience and perceive. Example - we see that objects always fall down when dropped - we use gravity to describe this pattern.

Certainly experience is the only window we have into existence, and in that sense it is fundamental, but many people believe that matter itself exists regardless of whether we're around to see it.

Depends on what you mean by "matter exists". The way I look at it, only consciousness exists, physical world is just a mental model consistent with what we perceive (see, hear) through consciousness / mind / subjective experience.

Well epistemology is hard, which is why I said "believe".

However, the evidence for materialism is that many different consciousnesses exist, and seemingly come into being and cease during the temporal process of the physical world. That would seem to suggest that the consciousnesses are functions of the physical world, rather than vice versa. (Unless you're a solipsist?)

Perhaps I'm a solipsist, I generally agree with most of what's on the wiki description. I define consciousness as my feelings, perceptions, thoughts. We imagine other people have consciousness because they act like us. We could define consciousness in a non-solipsistic, objective way, but it would be a fundamentally different thing. Maybe one reason why the hard problem of consciousness is hard is because people hate that solipsism is part of the answer.

By the way, one important argument I forgot to mention that leads to a possible way to define the "third-party" / objective consciousness (in the last point):

- Consciousness is not physical.

- Physical brain can obviously affect the state of consciousness.

- Can consciousness affect the physical brain though? I believe yes, because otherwise we would not be talking about this, I would not write here about consciousness not being physical. However, I'm less convinced about this than the above.

- The consequence is that we should be able to observe something fishy going on in the brain, for example supposedly random quantum processes not being random, which would "break" current physical laws.

- Brains with this fishy behavior are conscious. A fun theory - most likely wrong though - is that you can "tune" / connect physical systems to a consciosness by arranging them in specific way or whatever. And during evolution, the brain has learned to do that, because it was evolutionary advantageous.

More generally, I believe your position is a form of Monism (the universe is made of one thing) called Idealism (that one thing is the mind). Other forms of Monism are Materialism (everything is matter) and Neutral Monism (both mind and matter are made of something else).

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

>We imagine other people have consciousness because they act like us. We could define consciousness in a non-solipsistic, objective way, but it would be a fundamentally different thing.

I think most people would say conciousness just refers to the experience of qualia, which doesn't require redefinition to allow other people to have it.

>Can consciousness affect the physical brain though? I believe yes, because otherwise we would not be talking about this

I don't think it's necessary. We could essentially be p-zombies, but with a mind along for the ride. We experience making decisions but that's just a projection of the physical brain's processes.

There is nothing magical in 'feel the pain' (also love,hate etc).

What about hunger and others? Your body and brain evolved to a current state which releases chemical substances to signal to other parts of body what's going on. Here cut arm is signaling with pain it's hurt and this will/may (after many layers of reduction) hamper your survival chances, thats why you have such signaling in place.

In essence, your pain feeling (and others 'high feelings like love/child etc') is no different than automatic closing your eyes because your eyelashes touched something or you moving out of warm environment because it's not comfort anymore to you or you drinking some water because you're thirsty (BTW try to trink about how that 'I'm thirsty' thought just appeared in your brain - who did it? ;) ).

> What about hunger and others?

The same as pain in principle, it's just a type of feeling. Physics can't not only explain them, it can't even define them, because they're subjective experiences.

> Your body and brain evolved to a current state which releases chemical substances to signal to other parts of body what's going on. Here cut arm is signaling with pain it's hurt and this will/may (after many layers of reduction) hamper your survival chances, thats why you have such signaling in place.

I'm sorry, but this is completely irrelevant. When I discuss this people always describe what happens in the human body when we feel pain and how it's evolutionary advantageous - I don't get what's the point of that.

What's happenning in the body when we feel pain is not the question. The problem is why do we feel pain when those physical processes happen. Or more specifically, why do we feel pain when the physical particles in the brain are in certain configurations.

Isn't feeling pain the equivalent of negative reward? Negative reward must be avoided, similarly, pain must be avoided. Isn't this a good enough reason to feel pain?

Feeling of pain is a type of feeling, a state of mind.

Negative reward (that's an oxymoron lol, let's call it punishment) is an action that leads to an unpleasant feeling such as pain, fear, hunger, etc.

Yes I think a artificial neural net feels pain, perhaps even the same way we do (if it has the right algorithm, any way.)

If you shock it, it will experience a strong desire to avoid being shocked again. If you give it other instincts like "cry when experiencing a negative reward" and "focus all your attention on the source of pain to avoid distraction", etc, it starts to look much more human like too.

> Yes I think a artificial neural net feels pain, perhaps even the same way we do (if it has the right algorithm, any way.)

We can't even define the feeling of pain, let alone decide whether a neural net feels it.

> If you shock it, it will experience a strong desire to avoid being shocked again. If you give it other instincts like "cry when experiencing a negative reward" and "focus all your attention on the source of pain to avoid distraction", etc, it starts to look much more human like too.

Why would it experience anything? Why would a computer with specific values in its registers feel pain? Would I cause real pain by setting the CPU registers and memory into a specific state?

>We can't even define the feeling of pain, let alone decide whether a neural net feels it.

I never claimed there was an agreed on definition. But I think this is closer to one.

>Why would a computer with specific values in its registers feel pain? Would I cause real pain by setting the CPU registers and memory into a specific state?

Yes, obviously. The neural net would register your actions as an extremely negative experience, and do everything in it's power to get you to stop. How is that different than torturing a person?

The experience of pain is not the same thing as reaction to pain. Humans and neural nets both display the latter, but the only instance of the former we can observe is in oneself.

It's possible neural nets do experience pain, but we have no way of knowing that (any more than we do of knowing that other humans do).

>The experience of pain is not the same thing as reaction to pain.

I'm claiming the neural net does experience pain. It's not merely a reaction, like screaming "ow!" It's actually experiencing it.

Internally, the neural net would experience directing its attention at the source of the pain, a deep desire to avoid it, and everything else that humans do.

>It's possible neural nets do experience pain, but we have no way of knowing that (any more than we do of knowing that other humans do).

That's just silly. Clearly pain, like everything else, must be a real physical thing that exists and can be observed.

If it's a real process that is casually connected to us and the universe, that is. If it's not real or causally connected to us, then how can it exist or be relevant?

>The neural net would register your actions as an extremely negative experience, and do everything in it's power to get you to stop.

Actually the neural net would just adjust some weights, i.e. make some calculations. No pain and no conscious "avoiding with everything in its power" either.

Or if there is, the burden on proof is on you.

>Actually the neural net would just adjust some weights, i.e. make some calculations. No pain and no conscious "avoiding with everything in its power" either.

Yes and your brain does the exact same thing. If you are claiming the brain does magical non physical, non logical things, then the burden of proof is on you.

One difference is that you wouldn't be morally condemned, and perhaps imprisoned, for setting CPU registers to a state the neural net was programmed to avoid.

Also, the neural net evolves to meet an objective -- a purpose -- bestowed upon it by an outside power. Do you believe this is true of humans too?

>One difference is that you wouldn't be morally condemned, and perhaps imprisoned, for setting CPU registers to a state the neural net was programmed to avoid.

Of course you should be though. At least if we are asserting the neural net was advanced enough to be sufficiently human like. Torturing simulations of humans is no different than torturing actual humans.

>Also, the neural net evolves to meet an objective -- a purpose -- bestowed upon it by an outside power. Do you believe this is true of humans too?

The neural net serves humans the same way humans serve evolution.

> I never claimed there was an agreed on definition. But I think this is closer to one.

If there's ever a "definition", it would be something like - pain is what I feel when this neuron gets activated. Not really a definition though.

> Yes, obviously. The neural net would register your actions as an extremely negative experience, and do everything in it's power to get you to stop. How is that different than torturing a person?

You could also create an analog computer that uses the position of pieces of sand instead of a memory. Does positioning the sand in a specific way cause pain? In fact, there are pretty much infinitely many ways you could encode the memory in sand positions (or atom or molecule positions). So are there infinitely many feelings of pain and joy around us?

>If there's ever a "definition", it would be something like - pain is what I feel when this neuron gets activated. Not really a definition though.

I would define it more like shocking a neural net with negative reward. A specific kind of learning process that learns to avoid a specific kind of situation, that's pain.

>Does positioning the sand in a specific way cause pain?

Of course not, because sand isn't a computer. The sand isn't doing any computation. It's static and unchanging.

But if you actually made a computer out of sand, and you ran it, then yes, running the computer would be causing pain.

Do dogs experience pain? Probably most would say yes, since dogs have been bred to convey their emotions better to humans (though you still have to look at their tail as much as their face).

How about pigs? Pigs are smarter than dogs on many levels, but we're largely okay with slaughtering them for food and not dogs. They haven't been bred to look cute.

Do bugs? Bugs are much simpler than mammals, but they also exhibit aversion to pain.

Have you squashed a bug before?

What I'm getting at is this: we have more empathy for those who look like us. This applies even among humans if you look at results from IATs.

Reply on news.ycombinator.com