Artificial Consciousness
October 4th, 2007 –
What about consciousness? The sense of being one person and not somebody else? The sense of being a captive and concerned audience to the happenings of your particular self?
I think Singularitarians like myself tend to ignore this question — not because it’s unimportant, but because we don’t see it as relevant to our core concerns. Conscious or not, recursively self-improving artificial intelligence presents a risk and opportunity impossible to understate.
And yet, many are concerned that the “higher” human abilities, such as creativity and aesthetic appreciation, are inseparably connected to one’s conscious experience. Artificial Intelligence is insufficient, according to this view, because it could only emulate the types of thinking that are most obviously logical and analytical. To get the rest you would need some sort of Artificial Consciousness, too.
The implication that tends to follow here is that even if AI is possible, AC — Artificial Consciousness — may not be. Rather than jump right in to a rebuttal, however, let me toss this question to the cognitive science experts in the room: To what extent, if any, are human brain functions relating to creativity and aesthetic appreciation linked to those behind our perceived senses of identity and free will? Has the research progressed far enough to tell us?
I’m willing to hazard that an assumption this widespread could be a reflection of an actual relationship in the hardware, at least in the case of humans. So I’m not so quick to dismiss the above reasoning entirely.
But regardless of how research answers that question, we must be wary of assuming that mental capacities are linked in human brains because of properties inherent to intelligence-in-general.
We must remember that our brains are not so different from those of species unable to appreciate a good Picasso. Our newer abilities are owed to existing brain structures recently re-purposed by small evolutionary changes.
Imagine spending your childhood in a home built from an abandoned missile silo. You might have assumed that long ladders are indispensable for connecting large rooms, and that a simultaneous turning of keys by two people on opposite sides of the kitchen is the only reasonable way to start a dishwasher. How many linkages between our intellectual features are similarly silly?
Are some connections truly necessary? I answer with a skeptical maybe. It could well be that any AI capable of love will also have a kind of consciousness. But at this point in time I don’t know how to test that assumption. And apart from the obvious philosophical questions this raises, I’m still not convinced it matters.
As I was recently telling a colleague, I’m confident that all of my mental abilities, both logical and artistic, are owed to the structure of the matter in my brain. “And if it’s all in there, then I see no reason to argue that certain aspects of it will be reproducible on another substrate while others will not. Indeed, for all I know, AC may actually be simpler than AI. Maybe we’ve been creating Artificial Consciousness since 1893 and just haven’t realized it yet because toasters can’t cry.”
As far as I can tell, the main problem with consciousness is not creating it, but defining it… Until there’s an agreed-upon, testable definition, we will not know whether toasters have it.
Some people define consciousness as a subjective experience that can only be detected by the conscious agent. That’s the kind of untestable definition that I’m talking about. It is useless.
To me, intelligence and consciousness are the same thing. If I say I’m conscious of something, that just means I’m thinking about it. If I’m self-aware, well, that means I’m thinking about myself. I know some facts about the entity called “myself”, and my plans for the future include this entity. Don’t know what more there is.
The main issue about consiousness is that it can analyse concepts rather than objects.
Words for instance, are not objects but rather concepts. And we can analyse words. (look at the “What’s really red about red?” comment.)
A phrase is a fabrication of ours that links multiple concepts together, something which can only be observed in humans. An animal’s intelligence can allow it to learn about how to react to a certain situation. But you’ll never hear that animal talk about the situation.
I also agree that consiousness is only an advanced version of animal intelligence. It’s focus is however on intellectual objects rather than just stimuli.
In order to have an “AC”, you would need a computer so advanced that it can program itself, I doubt we will every achieve such a thing. To create AC would require us to at least be capable of fully understanding the brain in it’s very last details. The incredible network of brain cells is nothing more than a much greater application of a computer’s “If, Then, Else”. Those cells mostly work by generating(then) or inhibiting(else) “charges”, which has always reminded me of the “if-then-else”.
(Foot note : I’m not an expert on the matter. My knowledge is limited to a few biology classes in college and a somewhat greater amount of education when it comes to programming.)
“Some people define consciousness as a subjective experience that can only be detected by the conscious agent. That’s the kind of untestable definition that I’m talking about. It is useless.”
The definition is useless because it defines consciousness in terms of consciousness (i.e. “conscious agent”). If you could detect whether an agent detects itself to be conscious (e.g. watch their brain for thoughts “only I can detect this subjective experience I am feeling”). Of course, then you need a definition of agent, and a method for determining when an agent realises ve is conscious.
The idea is the fact that humans worry about consciousness, but plausibly many AI designs (e.g. AIXI) wouldn’t, is evidence that humans are conscious.
OTOH, I’m dealing playing with words I don’t properly understand here, so take this with the appropriate grain of salt.
“To me, intelligence and consciousness are the same thing. If I say I’m conscious of something, that just means I’m thinking about it. If I’m self-aware, well, that means I’m thinking about myself. I know some facts about the entity called “myself”, and my plans for the future include this entity. Don’t know what more there is.”
The fact that people get confused over whether they have conscious experience, wondering what is *really* red about red. This doesn’t seem like a necessary property of all reflective planners, but a specifically human quirk.
It’s best to keep these words distinct and not use “consciousness” for the more straightforward self-awareness.
“The fact that people get confused over whether they have conscious experience, wondering what is *really* red about red.”
I’m sorry, i’m not following you. People are conscious because they’re afraid they might not be?
Yes. Many AI designs would not have this confusion.
“Yes. Many AI designs would not have this confusion.”
How does your conclusion follow from this fact?
Actually, it doesn’t.
This doesn’t show humans have consciousness as some humans think it. It shows humans have some quirk which causes them to sometimes get confused about this. A straightforward expected utility maximiser would not get confused like this.
My feeling is that consciousness (subjective experience) is an inexorable result of possessing a degree of general intelligence (or “general-purpose” computation) above some threshold; probably along a continuum. As far as I understand it at this point, minds only work via determinism. Even though it will be a probabalistic AGI, the AGI mind is still “caused” and “bound-by” the low-level procedures. Qualia are constructed from “assembled” throughputs. Another reason, that I’ve been preaching lately, is that “Free Will” doesn’t exist except as an illusion. So there would be no selection pressure (or adaptive advantage) for evolution to “tack-on” consciousness as an add-on to general intelligence (through some sort of additional specialization). The most logical explanation in my opinion is that human consciousness is an unavoidable consequence of human general intelligence. (That basically they are one and the same, as Tiiba points out.) And it would be the same for an AGI. BTW, although I believe that consciousness exists along a continuum, I don’t accept panpsychism (the belief that all things/objects have some degree of consciousness). I think that some threshold must be passed. And I have first hand experience with this; I’ve been under general anethesia during which time I can assure you that I had no subjective experience of any kind, even though my body and brain were perfectly intact (as material objects). I believe that we should carefully proceed under the assumption that any functional AGI will have subjective experience of one sort or another.
What is your definition of free will? Make sure you use a falsifiable definition (what is a being that does have free will like?)
I like to define it as the ability to optimize the universe towards the agent’s goal state. Anything that stands in my way - prison, bias or laziness - makes me less free. The laws of physics do as well, but just because I’m bound by some simple rules, that doesn’t mean I’m a slave.
On the other hand, I don’t like qualia, because the only definition I’ve heard for them is all-exclusive.
“What is your definition of free will? Make sure you use a falsifiable definition (what is a being that does have free will like?)”
This is actually difficult to address, because often when someone asserts that they or something else have free will, it’s a mushy, poorly defined, and emotionally infused conception. My perception is that when most people assert “free will”, they automatically assume that it is some sort of mystical, causeless freedom of desire that doesn’t require any justification within “scientific” reality. My assertion is that dynamic human desires follow the rules of causality. That the mind, and the desires of the mind follow from an algorithmic process. In short, that our dynamic “will” doesn’t magically emerge from the aether. That they, our dynamic desires, are caused by something, and that we can exploit the rules of causality to create a Friendly AI. In my opinion, there is no being that has a dynamic “will” that doesn’t also abide by the rules of causality. I don’t dispute that it is possible for a human to “feel” varying degrees of “freedom”, such as imprisonment as you mentioned.
Looks like current AI is not really taking into account the specificities of human consciousness (HC).
HC is a characteristic of human intelligence, and consequently should be studied by AI.
HC is mainly about self-consciousness and phenomenal consciousness.
Self consciousness can be defined as “the possession of the concept of the self and the ability to use this concept in thinking about oneself”. It is different from phenomenal consciousness which is more oriented at subjective feelings and can be understood as “experience; the phenomenally conscious aspect of a state is what it is like to be in that state”.
HC covers free will and emotions. And these components are key to AI when looking at the risk for AI to become unfriendly to humans.
Research activities are on going on HC in philosophical approaches to a science of consciousness (http://www.consc.net/mindpapers/) and also, at a lesser extend, in the field of artificial consciousness (http://www.consciousness.it/CAI/CAI.htm).
I feel that SIAI activities should consider HC more explicitly.
If you are interested about an tentative understanding of HC based on self-consciousness in a way that could perhaps be transposed to computers, you can have a look at
(http://crmenant.free.fr/AAAI-2007/Artificial-Consciousness.Christophe-Menant.pdf).
I see consciousness,intelligence, and awareness as different things. I can be conscious without being aware and inteligence is debatable. Pesonally I can’t see consciousness arriving from where it was not already. I theorize that consciousness of some sort is inherent in all things although there may be no awareness. I also think that existance is not a state of matter but rather matter is a state of existance and existance itself is consciousness. I do believe that with sufficient advances that artificial intelligent systems may become self-aware like us and that a non-aware consciousness exists inherently, with the degree of intelligence depending on the complexity of the system.
At a certain level of system complexity the consciousness of individual components merge sort of like magnetic fields of pieces of magnetic materials. My view is more philosophical than scientific I know but it makes sence to me.
My two cents…
I think that consciousness is the result of a long sequence of (standard) calculations revealing a calculated probability distribution of what the immediately preceding “input” (of any particular type) “had to be” (ie. was most probable). Its surely got to be one of the deepest mysteries of the brain - how such very subtle differences in algorithm structure (eg. between different brain modules) can cause such a wide range of perception, from vision to emotion to memories and all that other stuff. Very weird. However, I suspect that it won’t be necessary to solve *all* the mysteries of the human brain in order to make an AGI. We really only need to capture the salient aspects, and we’ll have powerful software models to help us out along the way.
Also along these same lines, one can point to the dramatic changes in subjective experience brought about by alcohol or illegal drugs - which I presume exert their effects by subtly (and temporarily) changing the relevant algorithmic “structure” of the brain.
I had a spare moment so I thought I’d elaborate on what I wrote above. I think that consciousness is a matter of “calculating-out” a “signal” from the “noise”. Where the “noise” is the set of all possible conscious perceptions. And the “signal” is the perceived “input” that is revealed (by calculation) to have the the highest assigned probability of being the “correct” one. [I wish I could put this more clearly using language].
It would be nice to know at what probability value a given throughput would become a conscious perception. It would have to be less than 100% probability - but it’s hard to say more than that. Just to throw out a guess, I imagine it would be something quite close to 100% probability without quite reaching it. Only algorithms beyond a certain level of complexity would be able to “reveal” the perception with a sufficiently high probability value.
Ahh, speculation is simply too much fun - probably to an unproductive degree, even.
Whatever its definition i think its correct to assume that consciousness is a ‘conscious thing’ and who it to say that it has violition other than our personal aims and that perhaps it doesnt want us to understand it until it is ready for us to do so. The big question to me is, where are all the priests and spritual people in this research , or do the people with computers think that consciousness belongs to them? The fact that techcnically there is no definition of consciousness makes me wonder why researches think they are researching consciousness in the first place, its like the dot com boom, their faking it to get grants.