Consciousness in Simulation theory & AI, why do some believe that it is even possible?

Question

There are many famous philosophers that assume that "consciousness can be create by calculations in a computer". For example: Nick Bostrom with the simulation theory [I greatly respect and love his books and work]

Why do people assume that? Is there a good argument for it?

Here is why I find that proposition to be absurd as a computer scientist.

1. Because all that your computer is doing is addition in a loop with if-else branching using electronic logic gates ... I can do the same calculations without electricity using mechanical gears or even better I can do the same calculations by putting people in hall and asking them to calculate on pen and paper simple additions, they are Turing-complete meaning I can calculate anything that your computer can, including running Windows Vista and any program possible ... now yes using people with pen and paper to calculate what Windows Vista would output on the screen would run at 1 frame per day but the point is, it is still running no different from how any program is running.
2. Let say that we scan every atom in my brain with it's precise position and momentum ... We run a program on a large computer that simulates real physics of the atoms in my brain. The behavior IE the apparent emotions and actions of real me and the digital simulated brain of me would be identical. However it is obvious to me that the digital simulated brain would not be conscious IE it would be a "Philosophical zombie" because the digital brain is just calculations of what a real brain would do ... If we take my freezer and run a physics simulation of the freezer would you expect your CPU to get cooler or something, no because it is just calculation of where the atoms will end up in the simulated freezer?
3. 2+3=5 did I just create consciousness by doing that!!?? So this website is also conscious? I wounder what is it like to feel to be StackExchange.com consciousness? I see this as absurdity.

Answer 1

You ask for "a good argument" for the idea that consciousness is possible in a simulation. I'm not sure that any such argument exists, because in the absence of any agreement as to what consciousness actually is, or any objective way to measure or define it, it is impossible to meaningfully argue.

However, I will attempt to present one interpretation of consciousness in which it is clearly possible within a simulation; I do not expect to convince you that this interpretation is correct, but perhaps it will help you to understand what someone like Nick Bostrom might be thinking when writing about simultationism or AI.

The behavior IE the apparent emotions and actions of real me and the digital simulated brain of me would be identical. However it is obvious to me that the digital simulated brain would not be conscious [...] because the digital brain is just calculations of what a real brain would do ...

I think this is the key point of potential confusion; what is it that you think a real brain does? From my perspective, the relevant function of a brain is to process data in order to make decisions, and it seems obvious that consciousness is an essential part of that decision-making process.

(An actual brain also performs various autonomic functions such as regulating body temperature and heartbeat, but I assume we can agree that those functions are irrelevant in this context.)

If we take my freezer and run a physics simulation of the freezer would you expect your CPU to get cooler or something, no because it is just calculation of where the atoms will end up in the simulated freezer?

A freezer's action is physical in nature - it transfers heat from one place to another. Simulating a physical action is clearly distinct from actually performing that action.

On the other hand, we do not draw any distinction between "adding two numbers" and "simulating the addition of two numbers", or argue that a computer is only capable of doing the latter. This is because addition is a conceptual action, not a physical one. All that matters is the algorithm, not the physical implementation of it - whether you use an abacus, a brain, or a binary adder, you're still adding numbers.

Decision-making - and therefore, in my interpretation, consciousness - is very clearly an action of the second kind, a conceptual one, and therefore independent of the physical implementation.

Again, I am not attempting to convince you that this interpretation is correct, merely that is is both self-consistent and plausible. Personally, I think Occam's Razor favours it over the idea that the consciousness has some sort of independent physical existence, or is an aspect of the supernatural, but that's just one opinion. What I would ask is: do you have a good argument for your interpretation of consciousness?

Answer 2

Nobody has ever found any credible evidence that the human brain is anything besides a very complicated computer running strange software. Nobody has ever found any credible evidence that humans are not the same thing as p-zombies. In particular, you don't have any credible evidence that you're not a p-zombie.

"But that's ridiculous," you may say. "It's completely obvious that I'm not a p-zombie."

Yes, well, lots of falsehoods are completely obvious! For example:

• It's completely obvious that the earth is motionless.
• In the checker shadow illusion, it's completely obvious that the square in the shadow is lighter (meaning that it's printed with lighter ink, or that it's displayed using a greater pixel brightness) than the square outside of the shadow.
• If you flip a fair coin a large number of times, it's completely obvious that the difference between the number of heads and the number of tails will oscillate around 0.
• If a test for a given disease is accurate 99% of the time, and somebody takes the test and the result is positive, it's completely obvious that there's a 99% chance that the person has the disease.
• Suppose that there is some man who is older than some woman. It's completely obvious that the woman can never, later on, be older than the man.
• Suppose that there are two objects, and we want to increase the distance between them. It's completely obvious that the only way for that to happen is for at least one of the two objects to move.

All of these statements have seemed completely obvious to some person or another at some point, and yet all of them are false.

So it's clear that merely being completely obvious isn't good enough reason to believe something. In order for us to be justified in thinking that humans have a kind of consciousness that p-zombies do not, we ought to have some kind of corroboration of this notion. Some additional evidence for it.

But there's no additional evidence for it.

Answer 3

Materialism perspective

One of the arguments for the possibility of simulated minds comes from the assumption that in general, physics can be simulated. If all physics can be simulated, then it must be conceptually possible to simulate (among other things) all the processes happening on Earth, including biology, cells, organisms, living beings and also including human-like beings.

The materialism (or physicalism) perspective assumes and asserts, in essence, that "physics is all there is to reality", so if the physics of these beings - including the physics in their brains - are correct, then all their external and internal behavior would also be "correct". Thus, asserting that these simulated beings are substantially different from us in some aspect requires asserting that our mind function involves something more than "just" the physics and chemistry happening in our brains, which the materialism approach fundamentally denies as an unreasonable extraordinary assumption with no supporting evidence.

Answer 4

What is consciousness?

There isn't a commonly accepted definition of consciousness (that I'm aware of).

This lack of a precise definition presents the first and biggest problem in providing a concrete answer to "can artificial intelligence be conscious".

The Wikipedia page on animal consciousness goes into a bit of detail about this. It lists one example in 2004 where eight neuroscientists said the following in "Human Brain Function":

... we do not know whether consciousness can emerge from non-biological systems, such as computers... At this point the reader will expect to find a careful and precise definition of consciousness. You will be disappointed. Consciousness has not yet become a scientific term that can be defined in this way. ...

How can you say something is or is not conscious if you don't know what exactly consciousness is?

How do we measure consciousness?

So we don't have a definition, but let's consider how it might be measured.

Both of the below present some problems, but I don't think there are other ways to go about measuring consciousness (they cover a fairly broad range of possible ways to measure it).

So we may never be 100% sure something is conscious, even with a formal definition and a way to measure it.

But let's see what these ways of measuring might say about artificial intelligence.

Physical internal analysis

One way to measure it would be to look at the inner workings of an entity (e.g. a brain scan or map or looking at the architecture or flow of a computer or computer program).

Measuring it in this way would give the most conclusive answer, but it's also really, really hard. What if you can't see these inner workings in enough detail? What if you can't fully understand it? How would you even translate what you're seeing into an answer to "is this thing conscious"? If you can't fully understand it, would you be able to compare it with an entity that works fundamentally differently? There is still a ton we don't understand about what's going on in the human mind, so we're not really at a point where we can use that to tell whether something completely different is conscious.

Although one could actually make a fairly strong case here that some artificial intelligence can be conscious (in theory). Assume "souls" (as in non-natural parts of our minds) don't exist or aren't required for consciousness (this argument wouldn't work too well without this assumption). So brains are just biological systems that are often created in nature during reproduction. There doesn't seem to be any good reason to believe it's fundamentally impossible to artificially create something that can be naturally created, or to create something that functions exactly the same. In fact, artificial neural networks are, as the name suggests, created to be structurally similar to how brains works (but of course built from code instead of biology). Although they are much, much simpler, since there isn't enough computing power in the world to come close to what a human brain does (this seems to be more of a practical challenge than a theoretical problem).

Observational analysis

The other way to measure consciousness would be through observation, e.g. looking at how it behaves in response to stimuli.

We can compare this to the closely related topic of self-awareness. This is generally measured exclusively through observation. One common way to measure this it by putting entity in front of a mirror and seeing whether it recognises itself. You could also potentially ask it questions (if able) to see if its responses demonstrates self-awareness.

This has what I'd call "the simulation problem". If I am conscious, and something acts exactly like I would, is that thing also conscious? Since we're only measuring consciousness through observation, the only possible answer we can give here is "yes".

Now artificial intelligence might still be a long way off from what humans are capable of, but they have made quite a lot of progress towards what humans are capable of in many complex areas (such as natural language processing and computer vision). Of course there is still a long way to go, but I don't see a reason to believe that a computer or other artificial device is fundamentally incapable of simulating the behaviour of a human. For reference, I am a computer scientist working with artificial intelligence.

So, with this type of analysis, we would conclude some future artificial intelligence can be conscious (in theory).

Answer 5

Saying that consciousness is not possible in a simulated being has strange consequences as well. Suppose that we create a simulation that is very much like a human being. Say, we do your case 2 on someone who has never been interested in consciousness yet, and then unfortunately the real person dies, but the simulation carries on and then becomes interested in consciousness (the real person would have too, but passed away before that point). Let's say the simulation doesn't have access to any literature or anyone to talk with about this, but discovers many of the same concepts itself (since the real person might have done the same) -- it comes up with the idea of qualia, the inverted spectrum argument, the knowledge argument, etc. It argues fiercely in a similar way that you do. Do we really think it could have discovered these concepts without having access to real consciousness?

It seems that when I first started thinking about consciousness, having immediate access to the real thing -- conscious experience -- very much affected my thinking; it was the data used by my (I hope) logical reasoning. So how can the simulation end up reasoning the same way without also having access to it?

Answer 6

You are correct in finding this position absurd. And it's not just philosophers and comp. scientists but also physicists, like Brian Greene, who openly admit that what they think, is that consciousness is "created" by computation/symbol manipulation. There are three points I'd like to make.

1. They never define what is this "consciousness" that they hope to replicate. If you are under the impression that my consciousness is a accumulation of atoms in a very narrow range of trajectories, then, according to the people in the materialistic camp, replication of that arrangement should replicate me. Here's the problem. I can take you as a person, completely disintegrate you into your atoms, then add more atoms, and reconfigure you into the same arrangement. This is happening all the time anyways, we are eating food, excreting, losing skin, hair, losing some neurons, etc. Now when I do reassemble you into two people, whose eyes are you looking out of? If you say person 1, it makes sense to ask the difference in material of person1 and person2 in order to pinpoint that exact process that is causing you to be attached to the body of person1 and not person2. If this sounds silly to you, it's because it is. Consciousness is a process that isn't owned by us, nor created by us, because if that was the case, my consciousness would have a label that makes it different from consciousness of another person who'd have another label attached to it. But it's not like that.

2. Take anesthetic gases and you are supposedly "unconscious", so it makes sense to postulate that your consciousness was caused by processes hindered by the anesthesia. However that process is also material, so there's nothing stopping us from replicating it. In fact, that process is found in every single brain. The contents of consciousness might be different but the consciousness of the contents is universally identical. You're conscious of X means you're conscious. That's the bottom line. Consciousness can be said to be equivalent to knowing. Knowledge might be replicated in books, videos etc. The knower might be replicated as a dog, cat, a man or a woman. But the act of knowing itself is untouched by all identifiable parameters.

3. Gödel's incompleteness theorems highlight the fact that no formal system of symbols can ever be consistent and complete at the same time. This is obvious even if you're not a trained logician. You can keep on questioning your axioms and construct sentences that put two of your axioms against each other. The way I like to think about it is that think of yourself as an observer, made of the same symbols that you're trying to observe and then coming to a conclusion about the symbols that you're observing. Where is ground of understanding to be found then? Once you come to the conclusion that leaves are green, you ask what is green? Green is certainly not out-there. Green is also not perceivable unless you have the right cone cells in your retina. Green is also not perceivable if your brain doesn't know how to form a context within which it can contrast what green means. To perceive green means to perceive that which is not green. But then how can you ever know fundamentally what green is, without its context or boundaries? You can't. Gödel shows exactly this semantic with logic. You cannot assume a set of axioms to explain everything. They will collapse somewhere, and if they do, you never have a complete symbolic explanation of everything. Then there's the other interpretation that Roger Penrose likes to give: If you can construct a liar's paradox like sentence, which is perfectly consistent with the algorithm to construct sentences in your system, but which, nevertheless, the system can't prove or decide upon, then there's a factor in the human mind such that it can see the validity of the statement and step out of the infinite regress of proving that statement. A computer wouldn't be able to answer this and would go on an infinite loop. Computers cannot understand sarcasm.

The materialistic paradigm makes the mistake of taking conscious behaviour as equivalent to consciousness. It's not. There's no reason for it to be like that. The problem is, the other possibility that consciousness is prior to matter is so radical that it would upturn fundamental science on its head: think of something that is more fundamental than physics, giving emergence to physics as physics emerges chemistry and subsequently biology. Scientists are just not sold enough to give up on that because they operate under the paradigm where they expect to shuffle symbols and expect an answer. Consciousness reserves the power to completely take yes for a no and no for a yes. Such kind of inconsistency is not tolerated in any science that I know of. So it's unlikely consciousness will even be recognized as a radically different phenomena in the current science. The hard problem continues to persist because it's the result of not accepting consciousness of something different from matter.

Answer 7

this is a debate that has been going on for hundreds if not thousands of years, and will go on for many more.

there are many arguments on both sides of the "trench" and I assume that you are familiar with some of them given that you know of the term philosophical zombie.

My take on this question is that it is possible that you are aware of something obvious that most people appear to be blind to. Religious traditions across the world and for thousands of years report that "awareness" and "blindness" by different names.

For example in Hindu Advaita Vedanta this is referred to as being able to see the divine nature of the mind and of ultimate reality. Such people are literally called seers - rishi - and there are parallels in other traditions.

However, having a position on this core question of philosophy of mind does not in itself entail seeing or not seeing since there are many "wrong" reasons and false arguments for either position. We have an infinite capacity to endlessly debate anything and bury it under mountains of arguments and jargon.

Finally, this explanation would naturally seem nonsensical to most people, but this unfortunately cannot be avoided.

A few days ago I have uploaded a paper I have written about this, and in which I review this problem and the multitude of opinions about it at length, and I shamelessly plug it in here. It is called "A key hidden in plain view":

https://philarchive.org/rec/AIDAKH

I would love to hear what you think of it (note I am not a professional philosopher).

Answer 8

I think that it is certainly possible to construct a device that is conscious, and to construct a true Artificial General Intelligence (I don't know if the latter requires the former, and I don't think anyone else knows this either). After all, a human is itself such a device.

However I also see no evidence whatever, a lot of contrary arguments, for believing that our current technologies are anywhere near achieving this. I do not even see any reason to believe that any extension of our current technologies would achieve this. The only devices that we know do achieve this are brains, and brains don't work precisely like computers. It can even be argued that the more powerful we make our computers the less like human brains they become, not more - they are already far faster than brains and far exceed them in any task involving calculation, and this hasn't thusfar made them any more conscious than an abacus as far as I can see.

Note - a great book on this that I read recently is The Promise of Artificial Intelligence by Brian Cantwell Smith. It is a descendant of the phenomenological critique of AI the Hubert Dreyfus articulated in the 70s (I think) but more sympathetically phrased (I'm a massive fan of Dreyfus but his acerbic mockery needlessly alienated the AI community in my view and prevented reception of his ideas). Note that Cantwell Smith is more focused on making an AGI and the challenge there, rather than on making a consciousness. As I said earlier, I don't think it is clear whether these different challenges or the same one.