Why do people believe a turing machine can be fully conscious?

Question

In his book Consciousness Explained Dennett writes "Anyone or anything that has such a virtual machine as its control system is conscious in the fullest sense" [p281] referring to a Joycean machine which (if I understood correctly) may be implemented/simulated by a Turing machine.

Suppose I define Qualia as that thing which will always be left out by any implementation/simulation of consciousness by a Turing machine.

So it seems Dennett believes that Qualia defined in that sense does not exist.

The surprising thing to me is that most of my friends insist that no such Qualia exists, and they are all intelligent, and often software developers, who I expect, are supposed to know something about the nature of computation, even if only intuitively.

So far I failed to make even one of them realise that there is something in their inner experience that cannot conceivably be reproduced/simulated by a Turing machine.

They are in good company, BTW; here is an amazing "Closer to Truth" interview with Marvin Minsky where he explains away qualia: https://www.youtube.com/watch?v=SNWVvZi3HX8

Note I am NOT interested in any arguments for or against physicalism, idealism or dualism.

I am curious as to why so often non-philosophers, but nevertheless intelligent people who are supposed to know something about the nature of computation, insist there is nothing in their inner experience which cannot be reproduced/simulated by a Turing machine.

And my request is for references to discussions of this curiosity by philosophers, if such discussions exist.

Answer 1

In answer to your question, I think Scott Aaronson, a computer scientist at MIT, expresses the strong AI position very eloquently on his blog and in the notes for some of his courses. For instance, http://www.scottaaronson.com/democritus/lec4.html

Here's an excerpt in which he mentions qualia:

So, I asked you to read Turing's second famous paper, Computing Machinery and Intelligence. Reactions?

What's the main idea of this paper? As I read it, it's a plea against meat chauvinism. Sure, Turing makes some scientific arguments, some mathematical arguments, some epistemological arguments. But beneath everything else is a moral argument. Namely: if a computer interacted with us in a way that was indistinguishable from a human, then of course we could say the computer wasn't "really" thinking, that it was just a simulation. But on the same grounds, we could also say that other people aren't really thinking, that they merely act as if they're thinking. So what is it that entitles us to go through such intellectual acrobatics in the one case but not the other?

If you'll allow me to editorialize (as if I ever do otherwise...), this moral question, this question of double standards, is really where Searle, Penrose, and every other "strong AI skeptic" comes up empty for me. One can indeed give weighty and compelling arguments against the possibility of thinking machines. The only problem with these arguments is that they're also arguments against the possibility of thinking brains!

So for example: one popular argument is that, if a computer appears to be intelligent, that's merely a reflection of the intelligence of the humans who programmed it. But what if humans' intelligence is just a reflection of the billion-year evolutionary process that gave rise to it? What frustrates me every time I read the AI skeptics is their failure to consider these parallels honestly. The "qualia" and "aboutness" of other people is simply taken for granted. It's only the qualia of machines that's ever in question.

But perhaps a skeptic could retort: I believe other people think because I know I think, and other people look sort of similar to me -- they've also got five fingers, hair in their armpits, etc. But a robot looks different -- it's made of metal, it's got an antenna, it lumbers across the room, etc. So even if the robot acts like it's thinking, who knows? But if I accept this argument, why not go further? Why can't I say, I accept that white people think, but those blacks and Asians, who knows about them? They look too dissimilar from me.

Oh, this is also an entertaining discussion http://bloggingheads.tv/videos/2561?in=18:56&out=22:10

Answer 2

People believe that we have qualia because it seems that e.g. red is like something not reducible to declarative knowledge. (I've never heard a satisfactory account of why procedural knowledge isn't as vexing as qualia. But that's an aside.)

People believe that we are Turing-computable because all the physical processes that seem to be in play in biological systems can be described very well with mathematics that is Turing-computable. Empirically, we can't distinguish our universe from a Turing-computable one (at least at the scale of our consciousness). We can see that those things that we describe well with mathematics have a profound impact on consciousness (e.g. neuron reversal potential).

Because we believe we have qualia because we seem to, and because we believe the universe is Turing computable because we have a staggering quantity of empirical evidence consistent with that hypothesis, and we have no proof (aside from what boils down to argument from increduility, which is notoriously weak--see how well vitalists fared!) that qualia is Turing-incomputible (e.g. qualia cannot solve the halting problem), we conclude that qualia are computable.

Tossing aside computability because of a vague hunch that it feels wrong for qualia is just as foolish as tossing aside qualia because they seem awkward to compute.

Answer 3

As a former professional computer software engineer, physicist, and "thinker" - but perhaps not philosopher - I feel well placed to answer the core of your question:

"...insist there is nothing in their inner experience which cannot be simulated by a Turing machine."

And the answer is simple. These are educated and often pragmatic people who will heuristically and intuitively expect justification for an argument that doesn't match their experience and the progress in science, mathematics, and computation over the last few decades. We have not found any system that cannot at least to a good approximation be simulated using Turing machines, so why should there be any exception? If consciousness etc are emergent properties of complex dynamic systems then if you simulate those complex dynamic systems those same characteristics will emerge based on the evidence of other such simulations - e.g. weather forecasting.

I suppose this is also an intuitive appeal to Occam's Razor http://en.m.wikipedia.org/wiki/Occam's_razor This will be natural to software engineers as it's highly applicable to good software design.

Answer 4

I am software developer and I actually tried the same thing as the poster. I found it is most people find it hard to think about this problem, not just engineering types.

I see the same thing happening in the answers posted here. Most people are utterly and completely stuck in a specific kind of thinking. Qualia are hard and they basically just ignore the problem. Saying stuff like "I can reduce your behaviour to a machine and I wouldn't know the difference". This sort of thinking is totally oblivious to the real issues.

The qualia are so close, it is near impossible to see them. Once you do, you understand the problem and you immediately grasp the impossibility of it all. For some reason this seems to be a all-or-nothing kind of deal.

Answer 5

I'm not sure what the strong arguments in favor of conscious Turing machines are, but I'm aware of at least two reasons people deny it! ...the technical limitation of simulating (specifically) the human brain, and the mind-body problem. The mind-body problem has all the appropriate literature, so I will not attempt to repeat this, except to say that whatever a Turing machine could do, is done by "body".

In my capacity as a mechanical engineer, I have from time to time used commercial software that implements computation to simulate physical systems, such as finite element analysis of the stress at each element of a bracket or axle (etc) resulting from applied forces. The computer writes equations at each element with the goal of solving for an equality and uses iterative methods to arrive at a solution, since empirical solutions (as with differential equation solutions) are so rarely practical. Even so, the solution is fraught with interpretation, as when an internal 90-degree corner has a singularity in stress, the mathematical result of "stress concentrations."

If memory serves, a typical analysis that I performed a few years ago involved about 200,000 elements and my 2005-era computer could finish it in less than an hour. Using Moore's law and guessing that a transistor increases computation speed linearly, a computer today could do the same thing in between 1 second and 60 seconds.

Increasing the number of elements (or "degrees of freedom") is desirable because it increases the precision of the output, but at an exponentially increasing computer cost, as demonstrated here.

What would be necessary to simulate the human brain? The human brain has 85-86 billion neurons. If we take "Computational Time for Forward Elimination Steps of Naive Gaussian Elimination on a Square Matrix" as a reference, reasonably there is about an increase by 10 ^ 10 in computation time for each ten-fold increase in the number of elements in our computation, so when we add six factors of 10 to my 200,000-element computation, we need sixty factors of ten for the computation time.

However, my physical system computes elements that are only related to four or six nearby elements. Each neuron has on average approximately 7,000 synaptic connections. At a minimum, this can be multiplied by my element count for this purpose (of guesstimating a reasonable minimum computation time to simulate brain activity) so we need another thirty factors of ten on top of our computation.

My computation was linear, in that a given input in force causes a proportional response in stress, deflection, etc. Neurons are non-linear, so we must somehow account for this increase in computation time due to complexity. My computation was static, meaning that time-dependent aspects of the computation weren't part of the equation. Neuron activity is highly time-dependent (see this again).

Without addressing non-linearity and time-dependent issues, a brain computation would reasonably take a computer today between 1 and 60 seconds multiplied by ten to the ninetieth power. To reduce the computation time back to under a minute, Moore's law would need to work for the next 660 years.

My computation was for a single input and a single output. The human brain is constantly receiving inputs and producing outputs at an unknown or (more likely) stochastic rate. If this were 100 per second, then that tacks easily ten more years on our Moore's law estimate, placing a working brain simulation nicely in the year 2780.

Nobody believes that Moore's law will actually continue working for the next 670 years. Will quantum computers save the day? Google's quantum computer is 100,000,000 times faster. But this only shaves 8 factors of ten off of our 90, resulting in only 82 factors of ten for the brain computation and 550 years for Moore's law to work: computer brains by 2570.

Answer 6

Turing himself in the paper which he proposed the Turing game side-stepped this question. He felt that this was too difficult a problem to tackle. The essential notion of the Turing game is that imitation is enough.

Computation is a very visible artifact and pervasive feature of our technical civilisation and is also fairly new in its ubiquity though it has been theorised since Charles Babbages time. Hence people take a leap and suppose that the imitation of consciousness and consciousness itself are the same thing - when in fact, they are not.

Answer 7

I find few people actually hold the position "there is nothing in their inner experience which cannot be reproduced/simulated by a Turing machine." There are indeed people who assert that all of reality is a Turing machine, in which case they obviously will hold that position, but I find few people find this position useful. Instead, what I see asserted is more along the lines of "there is nothing meaningful in their inner experience which cannot be reproduced/simulated by a Turing machine." This statement holds up better, because it avoids making great statements about reality. Instead it makes statements about that which is meaningful, which itself is a rabbit hole because "meaning" has similar characteristics to "qualia" in debates such as this.

So far I failed to make even one of them realise that there is something in their inner experience that cannot conceivably be reproduced/simulated by a Turing machine.

What if I told you that there is an issue with your argument? You state that there is something that cannot be reproduced/simulated by a Turing machine. This is a strong statement. The weaker statement that there may be something that cannot be reproduced is easier to defend. As it turns out, we typically find that there is no proof either way. Both the position that there is something that can't be simulated and that there is nothing that can't be simulated are remarkably hard to prove.*

Indeed your argument has a fascinating twist. To prove that there must be some element which is not simulatable, we typically must assign it a symbol. The instant we assign it a symbol, the Turing machine can use that symbol and do operations on it. Your proof must contain a description of those operations, which puts it right in the realm of things that Turing machines are good at doing. Of course it may not be possible to determine if these particular operations halt.

If you are good at Math, you may be able to trap them. If they can define consciousness sufficiently, you may be able to demonstrate that the definitions of their words cause their description of reality to run afoul of Gödel's incompleteness theorems or Tarski's undefinability theorem. These are my preferred go-to arguments for cracking these kinds of beliefs, but they do indeed depend upon the other person's definitions being mappable into the domain and language used by Gödel or Tarski.

Alternately, you may find Guy Steele's Growing a Language to be a fascinating tool. It's a 53 minute long speech from 1998, and it's quite brilliant. For those who can't sit still or want to check his self-consistency, a transcript is available. In that presentation, Guy Steele builds up a language from the ground up. He shows how you can go from simple things to all the complicated beautiful things languages have to offer in a way which a Turing fanatic would appreciate. You will find quickly that he is able to describe all of the fancy qualia that one might ever want to describe using this approach, but there's a catch. I'll hide it behind a spoiler text in case you want to actually watch the process unfold:

Guy states that he will assume that monosyllabic words have an understood definition, but he must define all polysyllabic words using only words that have definitions at that point.

From this we can see where the interesting bits are. If you want to unsettle the Turing fanatic argument, don't look for the complex, look for the simple. Look for the simple things which are assumed by the other to be self evident, and simply ask them to be consistent and provide a definition for them.

I love pairing Guy Steele's speech with a quote by a character in the Robert Heinlein book, Stranger in a Strange Land:

Short human words were never like a short Martian word — such as "grok" which forever meant exactly the same thing. Short human words were like trying to lift water with a knife.

And [God] had been a very short word.

---

^{* Per the discussion in comments with Mozibur Ullah, I am apparently using the word "prove" wrong here. Apparently simply assuming one of these statements is true, axiomatically, qualifies as a proof in the formal sense. Here I'm using the word proof to describe a proof that consists of more than simply assuming the conclusion and writing "QED." I'm using "proof" in a sense of something which might actually convince someone that your conclusion is true when they didn't already believe it to be true axiomatically themselves.}

Answer 8

I am curious as to why so often non-philosophers, but nevertheless intelligent people who are supposed to know something about the nature of computation, insist there is nothing in their inner experience which cannot be simulated by a Turing machine.

I'm no expert at all but I do believe in qualia anyway.

I suggest it is because of a fear of difficult problems. Life is easier to model if life doesn't have qualia. In a similar way psychologists often act with physics envy. They are dazzled by models that are highly mathematical and so they do away with the more troubling facets of psychology.

And so greedy reductionism (though, of course, Dennett wouldn't use the term here).

Answer 9

Any physical system can be simulated by a universal computer. For some purposes the computer in question would have to be a quantum computer. However, the human brain is a wet, warm system in contact with the environment and on the timescale on which though takes place it won't exhibit any distinctively quantum mechanical effects like interference or entanglement. So the human brain can be simulated by a classical computer and the Turing machine can simulate any classical computer. The sense in which the Turing machine can simulate a system not not just that the initial and final states of the computation are the same. The computer can be set up in such a way that there is a mapping between the states of the computer and the way information flows within the system while it is computing the result. For any pattern of information flow in the brain while it is instantiating consciousness, that pattern could be instantiated in some function of the Turing machine's tape. For reasons that I think Dennett explains it might not be a good idea to simulate the brain in that way. The brain's architecture looks nothing like a Turing machine and there is no particular reason to translate what it is doing so that it can be run by the Turing machine if some other architecture would work faster and use less memory, e.g. - a network of computational gates.

You say

So far I failed to make even one of them realise that there is something in their inner experience that cannot conceivably be simulated by a Turing machine.

I'm not entirely clear on why anyone would say that something that has not been explained (consciousness) cannot be simulated by a computer. I am not aware of any clear statement of what problem is solved by postulating qualia in the sense you gave above. If you haven't explained consciousness and don't have a clear statement of the problem, then you can't have an argument for saying that consciousness can't be simulated. You might want to read "Godel, Escher, Bach: an Eternal Golden Braid" by Douglas Hofstadter if you haven't already.

It is true that no explanation of how consciousness is instantiated in the brain has been given. It is also true that no explanation of what sort of program could simulate consciousness has been given. I think it is also true that both Dennett and his critics have some bad philosophical misconceptions that may obstruct their efforts. For an article that describes some of the problems, see

http://aeon.co/magazine/technology/david-deutsch-artificial-intelligence/.

Dennett and other people working on this areas also have problematic moral ideas that might pose a problem, see "The Meaning of Mind: Language, Morality and Neuroscience" by Thomas Szasz.

Answer 10

Try this. The Turing Machine argument supposes no difference between a programme run for a long time on a low power machine, and a shorter time on a high powered machine. But that doesn't fit our experience of minds. Slugs aren't just slower than humans, it is a qualitative as well as a quantative difference.

When we look at qualia, we are looking at an interaction between sensing and subjective mental experience. The subjective mental experience is continuous, relational, grounded in the whole process of genetic and social development.

Edited to add: This is an argument from https://en.m.wikipedia.org/wiki/Integrated_information_theory

Answer 11

Well, as a computer scientist, to me, your mind is a series of one bit processors (neurons) wired together on a direct weight graph (nerves). I understand that this model is based on physicalism, but it may also be completely formulated mathematically and I find it to predict my actions accurately. I would argue that there is considerable evidence that our understanding of mathematics is philosophically credible.

I'm entirely uncertain what about that I can't simulate on a Turing Machine, or what esle, if this is incomplete explanation, couldn't be simulated.

In that sense, I believe the onus is on you to provide any reason at all whatsoever that there would be any part of consciousness that isn't Turing complete. I'd say that this requires an explicit definition of consciousness.

That is, I'm struggling to see any reasonable question here without a hard definition on Qualia. Because any Qualia that I can think of, as my mind is Turing complete, would necessarily also be Turing complete.

I hope that makes sense.

Thank you.

==================================================================================

Edit: I thought I should produce a Turing machine.

Let us take as given that you have a finite lifespan.

Let us also take as given you can think a finite number of thoughts at any given point in time.

Let us also take as given that thoughts take non-zero time, so that time may be discretized.

Suppose then that Goddess monitors your every thought and constructs a Turing machine that procedurally outputs your exact thoughts under your same circumstances. Note that this machine would have a finite number of states.

I'd consider that an existence proof of you as Turing complete.