Free Will

[ucim]

What's the point? It's not going to be significant on any moral axis, for instance.

Sure it is. It is far less immoral to cause unnecessary pain to a creature that is not conscious, for example. If machines evolve (or are designed) to become conscious, then there are moral implications about turning them off or disassembling them. And when they become smarter than us, but have already been put in control of things we can no longer handle (stock trading, nuclear defense, the energy grid...), they may decide that we are not "conscious" enough. There will be a narrow window in which we can argue with them, and then it will be over.

If your definition for consciousness contradicts our moral intuitions about what counts as a "person" in one direction or the other, we'll just decide that "consciousness" isn't the deciding factor and default to another axis on which to define "our kind".

But we're not trying to define "our kind". We're trying to define a certain essence of self-awareness, feeling, sense of self, which we bundle together into one word which we use to help define how we interact with them.

ETA: One difficulty we have is that humans are at the top of the intellectual scale of life on earth. We are not used to thinking of other living things as conscious, or even as having the ability to think. This is evident in the way people treat other races, religions, and even the other sex. "They" are clearly inferior. However, had we evolved as the second most intellectual creatures on earth, perhaps as pets to some higher life form, and thus already used to the idea that we aren't the only ones that can {whatever}, this question might not even come up. Of course they are conscious, and yes, we are conscious too, and so probably are {other life forms} and even {some other machines}. Beware the smugness of the peak.

Jose

[tomandlu]

However, had we evolved as the second most intellectual creatures on earth, perhaps as pets to some higher life form, and thus already used to the idea that we aren't the only ones that can {whatever}, this question might not even come up.

Jose

OT, but worth pointing out that mankind has spent an inordinate amount of energy trying to convince themselves that this scenario is the truth. It doesn't seem to have helped much.

[ucim]

OT, but worth pointing out that mankind has spent an inordinate amount of energy trying to convince themselves that this scenario is the truth. It doesn't seem to have helped much.

Because it's not true. Even granting the religious aspects, those only enter in "the next world". We can thus preserve our smugness while remaining humble.

Jose

[Eebster the Great]

For the record, I agree that a single apparently coherent text conversation isn't enough to determine consciousness. I also agree that Bayes should inform our initial assumptions as always.

However, a prolonged series of conversations about a range of topics is very strong evidence of consciousness, and importantly, this doesn't mean the lack of such a conversation is strong evidence against consciousness.

If the Turing Test can neither prove that something is a person when it passes nor that something isn't a person when it fails, then it's not a great test. And I would be skeptical that the evidence is "very strong." Alexa can have a conversation on a wide range of topics, but Alexa is not conscious, and I think its lack of consciousness is not merely a factor of complexity. I think the very way it operates is fundamentally different from the way conscious beings operate. That doesn't mean it can't be intelligent or that other machines can't be conscious, just not that one. I think intelligence is insufficient to demonstrate consciousness.

Eebster seems to have made that mistake twice, once when askingwwhether we thought a person incapable of conversation wasn't conscious, and again when asking whether ucim assumes people are p-zombies. Even if I thought a single conversation could prove consciousness, it wouldn't mean I thought a failed conversation disproved it.

ucim was the one who brought up p-zombies, and the point of bringing up the converse was to show that conversational ability is not really closely related to consciousness at all, and that the Turing Test was, like I said, just a rule of thumb, not a suitable definition.

[tomandlu]

Because it's not true. Even granting the religious aspects, those only enter in "the next world". We can thus preserve our smugness while remaining humble.
Jose

Oh, I don't know. Prior to the NT, there was no afterlife that one might look forward to, and God was forever rubbing our nose in our mistakes, or ordering us up mountains or into dung heaps, before ending with the equivalent of 'LOL - MY DOG IS SO FREAKED OUT!'.

Back On T, if we go with the assumption that consciousness is an evolutionary adaption, and that sensation rather than intelligence is its defining characteristic, then maybe part of the pathway to creating or understanding AI (more accurately AC - artificial consciousness - as a prelude to AI), is to focus on thinking about and reproducing the evolutionary pathway?

As a reasonable (imho) requirement for a useful (in evolutionary terms) consciousness, I would say you need some form of optional response, and the ability to like/not like particular stimuli. These can be as basic as you want (I'm tempted to add "some form of memory", but I'm wavering)

The main stumbling block imho is how to provide meaningful stimuli. Let's say we create an adaptive program whose function is to create and solve simultaneous equations. That's the easy part, but now we want to encourage it to produce equations whose solutions are all prime numbers - how can we meaningfully reward or punish the computer that is running the program depending on its output? What would be the programmatic equivalent of serotonin? I think if we could solve that, we'd be half-way towards AC, and then be half-way towards AI.

So, and yet again imho, maybe current AI research is something of a cargo cult - trying to produce a conscious machine by emulating intelligence and complexity, which is a result, not a cause.

As a minor aside, with regard to AI, I could never decide if Data (ST) was a more or less interesting character than he appeared to be. I suppose, in a sort of reverse begging the question - that question does actually resolve itself internally quite satisfactorily.

[tomandlu]

If the Turing Test can neither prove that something is a person when it passes nor that something isn't a person when it fails, then it's not a great test. And I would be skeptical that the evidence is "very strong." Alexa can have a conversation on a wide range of topics, but Alexa is not conscious, and I think its lack of consciousness is not merely a factor of complexity. I think the very way it operates is fundamentally different from the way conscious beings operate. That doesn't mean it can't be intelligent or that other machines can't be conscious, just not that one. I think intelligence is insufficient to demonstrate consciousness.

This is almost exactly the conversation and debate that my son and I just had about this! We both thought the evidence was stronger than you did, but still, essentially, rather unsatisfactory. I mean, basically, if you put a keyboard and a toddler in one room, and a computer program in another, and you have to decide which room to fill with cyanide, the toddler is screwed.

[Meteoric]

Well, yeah. That's how tests work. Generally, tests provide evidence, not proof. See: math tests, blood tests, smog tests, citizenship tests, etc. A student who passes a math test is reasonably likely to understand the material, but they may have merely cheated, or crammed, or otherwise scraped by without really grokking what was going on.

[gmalivuk]

↶If the Turing Test can neither prove that something is a person when it passes nor that something isn't a person when it fails, then it's not a great test.

As Meteoric points out above me, that's how all tests work.

the point of bringing up the converse was to show that conversational ability is not really closely related to consciousness at all

Bringing up the fact that streets can be wet when it's not raining doesn't show that rain is not really closely related to wet streets.

[morriswalters]

This is almost exactly the conversation and debate that my son and I just had about this! We both thought the evidence was stronger than you did, but still, essentially, rather unsatisfactory. I mean, basically, if you put a keyboard and a toddler in one room, and a computer program in another, and you have to decide which room to fill with cyanide, the toddler is screwed.

He/she could be deaf, mute, and blind and still be human and conscious, as well as screwed. Not every thing that is conscious can use a keyboard. His argument wasn't meant to serve that problem. He limited it to make it easier to define the problem, if I understand what I read. If money was no object you could buy a system now to watch your child and to listen for distress including brain activity as well as monitoring respiration, with good reliability, probably better than most humans. As he seemed to define it that device would be thinking in some limited fashion. This links to the original paper as a PDF.

Computer Machinery and Intelligence

and the point of bringing up the converse was to show that conversational ability is not really closely related to consciousness at all

There is evidence that while language is baked into the brain that you can develop without it. But it may be that there is some underlying structure which language supports, some people call this mentalese.

Certain deaf adults who neither have capability to learn a spoken language nor have access to a sign language, known as home signers, in fact communicate with both others like them and the outside world using gestures and self-created signing. Although they have no experience in language or how it works, they are able to conceptualize more than iconic words but move into the abstract, suggesting that they could understand that before creating a gesture to show it. [9] Ildefonso, a homesigner who learned a main sign language at twenty-seven years of age, found that although his thinking became easier to communicate, he had lost his ability to communicate with other homesigners as well as recall how his thinking worked without language. [10]

[ucim]

ucim was the one who brought up p-zombies

No, that was in response to somebody else's post, and had nothing to do with conversation. AISI, the Turing Test is an amusing game, little more. It gives a starting point for thinking about these things, but should be discarded as soon as meaningful conversation begins. And there's no such thing as a p-zombie. If it can act completely like it has some property, it has that property.

Prior to the NT, [...] God was forever rubbing our nose in our mistakes

Only in the story books. There are no contemporaneous non-religious nonfiction texts showing humans coexisting with deities, and certainly none that also deal with people asking the consciousness question. People were never actually in the (shared) position of being an inferiorly thinking species. (Though recent politics is making me rethink this.

Back On T, if we go with the assumption that [...] sensation rather than intelligence is its defining characteristic

I would go with that. Consciousness is fundamentally about the self-awareness of sensation. Enough "intelligence" is necessary to be aware of oneself; more is irrelvant. Sensation is the key. Consider what is is we "turn off" when we put somebody under general anesthesia. Consider what "disappears" when we're asleep (and how it returns when we're woken up... we haven't lost all awareness or alarm clocks wouldn't work).

As a minor aside, with regard to AI, I could never decide if Data (ST) was...

I never cottoned to the idea of an "emotion chip" - it made no sense. And people are trying to do this kind of thing now, trying to make computers think "like people".

First, why deliberately cripple a perfectly good machine by giving it all our shortcomings?

Second, emotions aren't added on top of thinking, rather, thinking happens on top of emotions. So, it's all backwards anyway. Emotions are the DOS of Windows.

Jose

[Copper Bezel]

Second point is valid and why the story concept doesn't really work even in principle for a loosely speculative story, but I think the point would be contested whether "human emotions" as circumscribed in the series are exclusively a set of shortcomings.

I think consciousness as you're thinking of it is still a nebulous and arbitrary thing, exactly as much as anyone else's intuition of what defines "our kind" vs. "others", but it's good you're starting to sketch it in a little. I happen to agree with almost all of it. There's no logical basis for that belief.

It's a minor point, but I also don't know why you think it's relevant that religions, which the majority of people in recorded history have lived according to, don't involve any real-life actual deities. Your conflation of the history of religion with Christianity and its teleological tendencies aside, a fair number of those people definitely believed in real, conscious entities of a kind different and superior to themselves, and tomandlu's quip was as a consequence both apt and incisive and definitely didn't warrant your uncorrection.

[ucim]

a fair number of those people definitely believed in real, conscious entities of a kind different and superior to themselves, and tomandlu's quip was as a consequence both apt and incisive and definitely didn't warrant your uncorrection.

No offense was intended; I hope none was taken. My point was just that those real, conscious entities were also seen as "different" from us (and other life forms) in a fundamental sense. A dog might look up to us, but still (I speculate) would see us as one of many other kinds of animals they encounter all the time.

I think consciousness as you're thinking of it is still a nebulous and arbitrary thing

Yes, it is still undefined. However, it is constrained.

Jose

[Eebster the Great]

the point of bringing up the converse was to show that conversational ability is not really closely related to consciousness at all

Bringing up the fact that streets can be wet when it's not raining doesn't show that rain is not really closely related to wet streets.

But I never argued that the Turing Test was completely useless, just not great. It produces a large number of type I and type II errors. I did say, multiple times, that it could be a handy rule of thumb, just not much more than that. It certainly isn't strong enough to outweigh the very low prior probability I would assign consciousness to a machine being tested, so even if I encountered a machine that passed the Turing Test with flying colors tomorrow, I still would not believe it was conscious. That means that for present-day purposes, it's not very valuable.

[gmalivuk]

I don't think it's fair to say it's a bad test on the basis that it's testing for something so unlikely.

A test with 99.9% specificity and sensitivity seems like it would be a pretty good test, even if we used it to test something with such low prior probability that most positive results are still false.

[tomandlu]

I don't think it's fair to say it's a bad test on the basis that it's testing for something so unlikely.

A test with 99.9% specificity and sensitivity seems like it would be a pretty good test, even if we used it to test something with such low prior probability that most positive results are still false.

But hasn't the Turing test been passed by decidedly (imho) non-sentient programs? I mean, "I am Napoleon" is a claim Napoleon would indeed make, but that doesn't mean I'm likely to give much weight to such a statement, should I come across someone who makes it.

I'm not necessarily disagreeing. We might want to require an AI to pass the Turing test (assuming the right sort of AI), but on its own it looks to me like it wouldn't tell us much. Even a negative isn't particularly informative if we've conceded that an AI (or just plain 'I') could fail.

It seems like the Schrodinger's cat of AI studies - more interesting for the questions it raises than for itself.

[Eebster the Great]

I would dispute that the Turing Test would be all that specific, though I suppose it depends on the population under study. The problem is that we may simply become skilled at creating AIs that can pass the test long before we become skilled at creating conscious AIs, which in practice would mean that it would for a long time have a specificity of 0 for machines. (Of course, it may correctly identify many humans as people, but so would all kinds of silly tests.)

[Copper Bezel]

a fair number of those people definitely believed in real, conscious entities of a kind different and superior to themselves, and tomandlu's quip was as a consequence both apt and incisive and definitely didn't warrant your uncorrection.

No offense was intended; I hope none was taken. My point was just that those real, conscious entities were also seen as "different" from us (and other life forms) in a fundamental sense. A dog might look up to us, but still (I speculate) would see us as one of many other kinds of animals they encounter all the time.

Sorry if I was unnecessarily snippy. I'm finding the tone of this thread somewhat hard to track with.

The perspective of actual pets on the matter is difficult to speculate on. My guess would still be the popular wisdom that they see us as different versions of themselves rather than different versions of - well, whatever it means to a cat or dog to consider another species inferior and other, which is where things get tricky, because we don't know at all what that subjectively means to them and whether it's the same as our perception of "animals".

[tomandlu]

Sorry if I was unnecessarily snippy. I'm finding the tone of this thread somewhat hard to track with.

I don't think there's a tone - more of a speculative meandering, and I've rather confused things by diverting a thread on free will into a general discussion on consciousness, so apologies for that.

Thinking about it, and if you assume that the first religions were forms of animism, then it would appear that one of our instinctive problems is not so much refusing to consider the conscious interests of other organisms, but of granting that consideration too liberally and too wide (alternatively, we lived in harmony and respect with nature, just like in Avatar, and yada, yada, yada).

[Copper Bezel]

I think there are open questions relating to consciousness, which makes it a more interesting topic than free will.

We certainly assign intentionality to damn near everything by default, and that's never stopped. We get angry at animals and mobile phones equally when they misbehave. I do think that actually believing them to be conscious agents might be another layer that could, depending on how it's placed, cause us to see ourselves and our human specialness differently. I guess it's hard to speculate coming from a Western cultural background, where our religious and cultural weirdness definitely works in the opposite direction instead.

[gmalivuk]

I would dispute that the Turing Test would be all that specific, though I suppose it depends on the population under study. The problem is that we may simply become skilled at creating AIs that can pass the test long before we become skilled at creating conscious AIs, which in practice would mean that it would for a long time have a specificity of 0 for machines. (Of course, it may correctly identify many humans as people, but so would all kinds of silly tests.)

I mean, I guess you can go ahead and imagine any future specificity you like, though at present it's basically 100%. And it would never fall to zero unless we started making *all* machines non-conscious test passers.

[Eebster the Great]

Ah shit I meant sensitivity. The idea is that it would start giving positive results for some machines even before a single machine was actually conscious.

[gmalivuk]

No, you meant specificity, and I know that was the idea and I explained why I think you're wrong.

You're talking about false positives, which are so far exactly zero unless there's been some new development I don't know about.

False negatives are much more common because as you pointed out before, there are plenty of reasons why a conscious person might not be able to carry on a conversation. However, babies probably constitute the overwhelming majority of such cases. If we restricted ourselves first to entities capable of producing multiple words in some existing language, the false negative rate would drop quite a bit.

[Copper Bezel]

False positives are exactly the thing, I mean, sensitivity is false vs. true positives, and there are no true positives. Glancing at the Wikipedia page, there are definitely bots that can convince humans that they're humans, which is a nonzero false positive rate.

[morriswalters]

I've read through the paper multiple times and in it he never uses the word consciousness, except in his response to the Argument from Consciousness. He uses the word thinking.

I do not wish to give the impression that I think there is no mystery about consciousness. There is, for instance, something of a paradox connected with any attempt to localize it. But I do not think these mysteries necessarily need be solved before we can answer the question that we are concerned with in this paper.

The point of the Imitation Game was not to be perfect, but to do it well as a human under his specific conditions. To say that any interaction with some type of software has not convinced you that it is conscious is a bias of prior knowledge. Given that you didn't know that Siri was an Apple software product, could you ever rally know that the speech didn't originate with a human named Siri, given sufficient improvement in the software? And he never gives any thought to the idea that a construct such as Siri could be output unrelated to the overall process that might be consciousness in a machine.

[gmalivuk]

↶
False positives are exactly the thing, I mean, sensitivity is false vs. true positives, and there are no true positives.

No, sensitivity is true positives vs. false negatives. That is, it's the proportion of actual positives that the test identifies as positives.

Glancing at the Wikipedia page, there are definitely bots that can convince humans that they're humans, which is a nonzero false positive rate.

That comes down to precisely how you're defining the test. No machine has ever convinced a majority of judges that it's human, for example. And at least from the transcripts I could find, the Loebner Prize appears to be awarded based on the responses given to a predefined set of 20 questions, which is hardly a full and varied conversation.

[Copper Bezel]

No, sensitivity is true positives vs. false negatives. That is, it's the proportion of actual positives that the test identifies as positives.

Ah, gotcha. That makes sense now, then, sorry.

And he never gives any thought to the idea that a construct such as Siri could be output unrelated to the overall process that might be consciousness in a machine.

I'm really of two minds about this part. On the one hand, I don't think there's a shortcut to simulating all the properties of self-awareness without actually creating it. On the other, it seems shortsighted to designate almost any subset of those properties and assume it's impossible to simulate them, because your machine can always be designed to study for the test. And of course, at some point we get back into the argument over how much you need to really count.

[ucim]

On the one hand, I don't think there's a shortcut to simulating all the properties of self-awareness without actually creating it.

I agree here.

On the other, it seems shortsighted to designate almost any subset of those properties and assume it's impossible to simulate them, because your machine can always be designed to study for the test. And of course, at some point we get back into the argument over how much you need to really count.

The dilemma goes away if you consider consciousness to be a non-binary property, that appears by degree. Even humans possess various degrees of consciousness throughout the day.

Jose

[Copper Bezel]

My issue with treating consciousness or awareness as a grade is that whatever we're interested in is probably not a linear property, so it's really a bunch of intersecting axes, not just one. Something can be more or less conscious, certainly.

[morriswalters]

Without language is it possible to self reflect, to be aware of being aware in the way we seem to be? How might you do so? The quote about the young man raised without language raises this point, or at least suggests it. Was he self aware prior to getting language?

[ucim]

Without language is it possible to self reflect...

I don't see why not. Language is for communication between individuals, not for introspection. Like everything else, it can influence introspection, but i see no reason that it is necessary for it to happen in the first place.

Jose

[eSOANEM]

Eh, I'm not so sure. The evidence from congenitally deaf people who were not taught to sign at a young age seems to suggest that language plays a majorly fundamental role in all human cognition including introspection.

[morriswalters]

Language is for communication between individuals, not for introspection.

Certainly language is used for communication. But it doesn't preclude other uses. I tried to come up with an example. The best I could do, revolve around arithmetic. If you don't have a formal structure you can't precisely add large numbers, or even think well about them. Without language stars might amaze you, but with language you can attempt to count them.

[ucim]

Without language stars might amaze you, but with language you can attempt to count them.

But consciousness is about being amazed, not about being scientific.

The evidence from congenitally deaf people who were not taught to sign at a young age seems to suggest that language plays a majorly fundamental role in all human cognition including introspection.

Got a citation that relates to introspection?

Jose

[eSOANEM]

I'm struggling to find one explicitly stating introspection, but the effects of not teaching congenitally deaf children to sign on cognition as a whole are well accepted amongst the deaf community & those working with them and necessarily imply an impact on introspection.

[Copper Bezel]

It's very hard to isolate a person from language completely, and just as hard to isolate cases of total language deprivation from extreme social deprivation and neglect. Human brains are very, very deeply wired for language. Even in these cases of a deaf child growing up in a remote village in a culture without literacy and so on, they tend to develop some degree of home sign language with their families.

And I don't really know what specific kind of introspection you're looking for that is more like being in awe of the stars than counting them.

[ucim]

To be conscious implies the ability to feel, not the ability to think. Awe is a feeling, as is pain and joy. When we decide it's ok (or not ok) to boil a lobster alive, it's not based on whether they can count, but whether or not they can feel. If there is any research that points to the idea that teaching somebody to think (or sign, or speak) gives them the ability to feel (as opposed to the ability to analyze and communicate that feeling), I'm listening. Otherwise, count me unconvinced.

Jose

[morriswalters]

When we decide it's ok (or not ok) to boil a lobster alive, it's not based on whether they can count, but whether or not they can feel.

Without higher thought the only thing that will save that lobster is me just having eaten. That and my shellfish allergy. However some like Pinker suggest that higher level consciousness might require language.

[ucim]

Without higher thought the only thing that will save that lobster is me just having eaten. That and my shellfish allergy. However some like Pinker suggest that higher level consciousness might require language.

You seem to be equating consciousness and thought, or thought and feeling. I do not. That's our difference. Do babies not need anesthesia because they can't do calculus?

Jose

[Xanthir]

I mean, that was a legitimate position held by doctors for a while, and still factors into the ethics of anesthetics - is a "you're unconscious" anesthetic morally equivalent to a "you're paralyzed and will forget this" anesthetic?

(The answer is "no" in both cases - the stress of extreme pain is reflected in the body in more ways than just literal memory, so an amnestic, whether chemical- or baby-brain-based, is not actually equivalent.)

[eSOANEM]

↶
To be conscious implies the ability to feel, not the ability to think. Awe is a feeling, as is pain and joy. When we decide it's ok (or not ok) to boil a lobster alive, it's not based on whether they can count, but whether or not they can feel. If there is any research that points to the idea that teaching somebody to think (or sign, or speak) gives them the ability to feel (as opposed to the ability to analyze and communicate that feeling), I'm listening. Otherwise, count me unconvinced.

Jose

You said introspection not sentience. Introspection is an aspect of cognition, sentience is the presence of an aspect of cognition (perception and/or emotions). I made no claims (and would not care to) about the effect of preventing a child from learning any language on their sentience because sentience, like consciousness, is inherently untestable (cf p-zombies), and denying it to other humans leads to all sorts of eugenics stuff I don't find ethical.