Consciousness experiment

[modernPrimitive]

As a "derail" inspired by the thread "Does Consciousness Survive Death?" I'd like to propose the following "thought experiment" for discussion:


PART A: software simulation of "consciousness"

1) Suppose that scientists manage to create a complete and accurate computer simulation of the human brain using purely software simulation. In other words a program runs on a super-computer that simulates all mechanics involved in the workings of the brain. This would include modules that simulate everything from the behaviour of neurons, neural networks, specialised functioning in areas of the brain and even a simulation of the human body and the five senses (since for the sake of completeness for exmaple one would include unconscious signalling to and from the body, such as breathing, danger etc etc)

2) This program would then implement a simulated environment module of the life of an average person from birth to death, including all details, people, school, jobs etc etc, which would feed sensory information to the first module. It's imoprtant to realize that while there would not be mechanical implementation of the five senses in order for the first module to interact with the outside world, this would be unnecessary as this "faked" information would be fed directly to it from the environment module - visual, auditory, tactile information etc etc.

The question is:


Would this computer simulation actually become conscious, aka SELF-AWARE? It might exhibit external signs of being self-ware but would it truly be self-aware?

PART B: Hardware simulation of "consciousness"

Now imagine that exactly the same simulation is built except that it is built around hardware - aka robot style - in other words, actual nano-modelling of neurons, organs, mechanical eyes etc etc. This simualtion would not need an environment module as our robot would interact directly with the real world.

Would this type of "simulation" become self-aware?


If your answers differ for PART A and B, why do they differ?

[Marduk]

1) Maybe
2) Maybe
How do you tell? :huh:

[modernPrimitive]

1) Maybe
2) Maybe
How do you tell? :huh:

Yeah, granted that's difficult.

I thought some people might start with the argument that these simulations might exhibit external signs of self-awareness.

However some physicalists in the thread "Does consciousness survive death?" have claimed (directly or indirectly) that either of these would produce self-awareness. I'm inviting them to explain.

[skepticalbip]

Your questions seem to assume (but don't define) "consciousness". Without nailing down the meaning of the word, neither question has any meaning.

Sorta like asking if a farfnel has all the characteristics of farfnels except it is brown, is it still a farfnel? It would depend on whether not being brown a necessary trait of farfnelness.

[modernPrimitive]

I would say that due to my bias based on how I see computers versus robots (hollywood movies) I would instinctively be inclined to consider simualtion B to be more likely to become self-aware than simulation A.

And yet there should be no difference.

[modernPrimitive]

Your questions seem to assume (but don't define) "consciousness". Without nailing down the meaning of the word, neither question has any meaning.

Sorta like asking if a farfnel has all the characteristics of farfnels except it is brown, is it still a farfnel? It would depend on whether being brown a necessary trait of farfnelness.

"Self-awareness". In the sense that humans have it.

[skepticalbip]

Your questions seem to assume (but don't define) "consciousness". Without nailing down the meaning of the word, neither question has any meaning.

Sorta like asking if a farfnel has all the characteristics of farfnels except it is brown, is it still a farfnel? It would depend on whether being brown a necessary trait of farfnelness.

"Self-awareness". In the sense that humans have it.
That seems to be just a synonym, not a definition. To be honest, I have no idea what either the words consciousness or self-awareness means specifically. Are either a dog or a snail or a chimp self-aware? Are severely mentally retarded humans self-aware? What exactly is the dividing line?

[Deleet]

Don't know. What evidence are we supposed to believe on? If any.

[Sapho]

If it said it was, we would probabnly have to accept it as true. Only the computer/robot would know.

[Janus]

Yes to all questions. As to how we would know the AIs are self-aware, that they behave as though they're self-aware would be good evidence, but not conclusive. To know for certain we would need to understand the mind to such a degree that we understand what it is that makes self-awareness possible. If the AI has this something, then it's self-aware.

[modernPrimitive]

I was thinking how one would code the part of the simulation that would be responsible for making decisions. Conscious decision making is not a random process - indeed I concluded that the decision making process of humans and animals is at the most primal level to with the aleviation of suffering. (yes Gautama Buddha hit the nail on the head)

Even boredom could be considered a type of suffering which might affect the decision to seek out entertainment. Suffering would be the driving process in all decisions.

Therefore the simulation would need to be capable of experiencing suffering and I think if we could recognise attempts to aleviate suffering on the part of the simulation that might be an indication of self-awareness.

This hopefully gives us a means to identify self-awareness, aside from any strict definitions.

[modernPrimitive]

Yes to all questions. As to how we would know the AIs are self-aware, that they behave as though they're self-aware would be good evidence, but not conclusive. To know for certain we would need to understand the mind to such a degree that we understand what it is that makes self-awareness possible. If the AI has this something, then it's self-aware.

Ok.

However when scientists run say a weather simulation (or as I recently read a simulation of the evolution of dark matter) we do not consider the simualtion to be the thing itself.

Why would this be different?

[Janus]

Yes to all questions. As to how we would know the AIs are self-aware, that they behave as though they're self-aware would be good evidence, but not conclusive. To know for certain we would need to understand the mind to such a degree that we understand what it is that makes self-awareness possible. If the AI has this something, then it's self-aware.

Ok.

However when scientists run say a weather simulation (or as I recently read a simulation of the evolution of dark matter) we do not consider the simulation to be the thing itself.

Why would this be different?


Well, the way you've described both simulations makes them more than simulations, it makes them re-creations. Copies, really. Usually, a simulation is an abstraction of reality, that is, it doesn't replicate everything that's going on, it only simulates some aspects of what is going on in the hope that this will be enough to accurately predict the future. As I said, a simulation that simulates everything is more than a simulation, it's a copy.

So what you're really asking, if I've understood correctly, is if re-creating the human brain and everything it does in a different 'substrate' would make any difference. I don't see why it would.

[Simen]

Yes to both questions: consciousness is substrate independent. A recreation of the brain and its input will be conscious if the brain is, which it is. I believe. Since the experiment hasn't been done it's not like we have any emperical data to argue over.

[Simen]

Yes to all questions. As to how we would know the AIs are self-aware, that they behave as though they're self-aware would be good evidence, but not conclusive. To know for certain we would need to understand the mind to such a degree that we understand what it is that makes self-awareness possible. If the AI has this something, then it's self-aware.

Ok.

However when scientists run say a weather simulation (or as I recently read a simulation of the evolution of dark matter) we do not consider the simualtion to be the thing itself.

Why would this be different?

In what way is this a simulation of a brain? It is a brain.

[WCH]

According to Jung, babies aren't self aware at first... instead they come to be as they mature a little. I can't remember what the average cutoff age was he suggested, maybe six months or something?

[Abel]

Yes to all questions. As to how we would know the AIs are self-aware, that they behave as though they're self-aware would be good evidence, but not conclusive.

Why do you say it wouldn't be conclusive? Recognizing oneself in a mirror, using "I" as a reference to oneself, understanding that one's consciousness is finite are all behaviors, right?

To know for certain we would need to understand the mind to such a degree that we understand what it is that makes self-awareness possible. If the AI has this something, then it's self-aware.

I don't think self-awareness can be defined by the human experience alone. Within our own species, even, there's a broad definition that can be applied to the term, don't you think?

[Janus]

Yes to all questions. As to how we would know the AIs are self-aware, that they behave as though they're self-aware would be good evidence, but not conclusive.

Why do you say it wouldn't be conclusive? Recognizing oneself in a mirror, using "I" as a reference to oneself, understanding that one's consciousness is finite are all behaviors, right?

Yes, but I can imagine a machine that would be programmed to act as if it recognizes itself in a mirror and understands that its consciousness is finite, even though it doesn't really recognize or understand anything.

We can already see a precursor of this scenario by looking at chatbots. Of course they're not very advanced so they won't fool you into thinking you're chatting with an intelligent being for very long, but they do demonstrate by extrapolation that it's possible for AI to fake self-awareness.

Of course, in the thought experiment laid out by the OP, if we know that the 'simulation' is perfect, then I agree that behavior that seems self-aware is conclusive evidence of true self-awareness, because each one of us knows that the pattern of the human brain does create true self-awareness. I was talking about AI in general.


To know for certain we would need to understand the mind to such a degree that we understand what it is that makes self-awareness possible. If the AI has this something, then it's self-aware.

I don't think self-awareness can be defined by the human experience alone. Within our own species, even, there's a broad definition that can be applied to the term, don't you think?

Sure. I don't understand what this has to do with my post, though.

[SteveP]

John Searle argues that a computer simulation of a brain can no more create "real" thoughts than a computer simulation of a liver produce real bile.

He supports this argument with the well known "Chinese Room" thought experiment, in which a man ensconced in a box receives patterns and diagrams on pieces of paper, and with the help of a large instruction manual draws more patterns and passes them out. The patterns are in fact written chinese, and the man appears to understand chinese even though he has no understanding of what he is doing.

This is of course in response to the Turing test, in which it is argued that if a person is unable to distinguish in communication between two "black boxes", one being another person and the other being a computer (the tester knowing that one of the boxes is a computer but not knowing which), then the computer must be considered to be conscious.

This seems reasonable; if we were to substitute Searle's real liver with a black box that performs exactly the same functions, producing real bile and all the rest of it, then the rest of the body wouldn't care and the black box might as well be a real liver.

However the Turing test assumes that the only qualification required for a referee trying to determine if a black box is conscious is that the refereree himself be conscious.

Arguments of solipsism might throw doubt on this assumption. Implicit in the Turing test is the (pretty much Skinnerian) position that consciousness is specified by an entity's external behaviour. This seems to conflict with what we know (or at least, what I know) about our subjective experience of consciousness.

Dennett (IIRC) argues that even if the man in Searle's chinese room doesn't himself understand chinese, the combination of man and instruction manual does understand chinese.

As for me personally, I am firmly in the "materialist" camp in that I ascribe consciousness to purely physical, chemical/electrical/cellular activity in the brain (not quantum!), however I'm not a fan of Skinnerism... so whether the Turing test is valid I'm not sure..:huh:

[Michael Sansom]

The OP goes right to the heart of a naturalist approach to
consciousness. There is fame and fortune awaiting anyone
who can provide a comprehensive and testable answer.

Though a materialist/naturalist re the human mind, I have no
real idea about the necessary and sufficient conditions for
any material system to be conscious. I have one certain
example of sufficiency: the human brain but don't know how closely (or not)
we would have to emulate the various structures, processes and cell types
in order to produce a conscious device. The answer might require such
close modelling that the device would be a brain. If not, then we simply do not know.

On the practical side, I'd give any combination of hardware/software
that passed the Turing test the benefit of the doubt (pending more study).
Though I much admire Searle's work, I'd currently bet that Dennett is closer to
the truth.

[modernPrimitive]

I was thinking how one would code the part of the simulation that would be responsible for making decisions. Conscious decision making is not a random process - indeed I concluded that the decision making process of humans and animals is at the most primal level to with the aleviation of suffering. (yes Gautama Buddha hit the nail on the head)

Even boredom could be considered a type of suffering which might affect the decision to seek out entertainment. Suffering would be the driving process in all decisions.

Therefore the simulation would need to be capable of experiencing suffering and I think if we could recognise attempts to aleviate suffering on the part of the simulation that might be an indication of self-awareness.

This hopefully gives us a means to identify self-awareness, aside from any strict definitions.

hmmm....the problem with this is that it would be relativey easy to implement using current technology and yet the sim would not become self-aware for the simple reason that it would be executing a pre-written function that is telling it how to behave rather than giving it the choice as to how to behave.

Determinism vs free-will is the crux here. Accessing past memories and basing future decisions on these is quite simple - for example not touching a hot plate based on past experience - but the code would always produce a perfectly expected outcome - effectively the sim is behaving "as a robot" - doing as it is told to do.

A simplistic version of this could easily be written right now however it would be no more sophisticated than knowledge management / dashboard applications.

Admitedly this isn't exactly the same thing as simulating the brain but rather attempting to simulate the thought process of humans. I suppose two entirely different things at the end of the day.