Can minds be uploaded in computers?

Question

The idea runs around of loading minds up in computers. It's a recurring theme in SF culture and thought about the technological possibilities in the future. We can read on Wikipedia:

Mind uploading, also known as whole brain emulation (WBE), is the theoretical futuristic process of scanning a physical structure of the brain accurately enough to create an emulation of the mental state (including long-term memory and "self") and transferring or copying it to a computer in a digital form.

I wonder if this can be done, even in principle. It presupposes that mind can be extracted, collected, and injected. I think mind is bounded to a living brain, and the living body and world the body walks around in.

So what to think of the conjecture?

Answer 1

Anything thing can be done in principle, so long as one is comfortable with a certain amount of magical thinking. But with respect to mind emulation, there are (as best I can see) several obstacles that need to be overcome, none of which we are close to overcoming.

1. Computers as currently designed are quasi-linear: a small number of independent 'cores' with high processing speeds. The human brain, by contrast, is better modeled as massively multi-processed: a very large number of slow, simple 'cores' organized in groups and hierarchies. The first does not emulate the second well.
2. The human mind does not seem to store or retrieve information anything like the way modern computers do. Storage and processing are not separate areas; recall is more a matter of association than location.
3. The human brain is an analog device built around subtle electrochemical differentials. Converting it to a digital platform would inevitably be lossy. It is also (I imagine) a non-linear system subject to sensitivity to initial conditions: small initial digitization errors will rapidly produce large differences between from the original. Even if the first two points can be overcome, the resulting emulation would likely only resemble the original tangentially.

Computer intelligence, if we ever achieve it, is likely to be of a different nature from human intelligence. I doubt it will be any easier for a computer to think like a human that it would be for a human to think like a computer.

Answer 2

There are several different issues here:

(1) It it possible in principle to scan your brain and create a mechanism of some sort that would completely duplicate your behavior?

(2) If you could do (1) would the mechanism experience a sense of awareness? Is there something it would be like to be such a mechanism or would the mechanism just be a computer mindlessly executing an algorithm?

(3) If there is something it is like to be such a mechanism, is it the same as it is like to be you, or would it feel different to be such a mechanism?

(4) If what it is like to be such a mechanism is the same as it is like to be you, would the mechanism essentially be you?

With respect to (1) there are three schools of thought. Let's call them computationalist, organicist, and dualist. The computationalist says that the brain is just a biological mechanism running a computation, so that computation could be executed in other hardware. So the answer to (1) is "yes". An organicist would say that the brain is not just a wetware device running a program; it is a complex biological organ, part of a complex biological system, and we have no idea what features of that system are responsible for your specific behavior. It may be that there is no algorithm that can simulate that behavior, so an organicist would not claim to know the answer to (1). A dualist says that our behavior is not determined by the brain but by the mind, which is not physical at all, so they would say that simulating the brain could not perfectly simulate behavior.

For question (2), we have the behaviorists vs. the mentalists. A behaviorist would claim either that these internal mental states don't exist (they are some sort of illusion) or that the internal states are bound up with behavior so that exhibiting the behavior proves the existence of the states (and possibly even causes the states). A mentalist would say that behavior and internal experience are separable and one does not prove the other.

The mentalist has the better argument here. An illusion is a mental state. If you are experiencing an illusion then you are experiencing a mental state, so the idea that mental states don't exist doesn't make sense. The idea that behavior proves the existence of mental states or (even worse) causes mental states can be refuted like this: you can program a computer to write "hello world". It does not gain a mental state when you do this. You can program a computer to act like a therapist with a very simple program. You can program it to recognize faces with a lot more code. At none of these points is the computer having mental experiences. You can keep getting more complex until you are simulating a person. There is no point along that process where it makes sense to suggest some sort of miracle happens and the computer gains mental states.

Some behaviorists respond to this argument by asserting a sort of panpsychism--the belief that everything in the universe has some level of mental experience. This still leaves open the problem of how this panpsychic mind connects to human behavior. After all, just because the circuits and gates and wires have some minimal form of internal mental experience, how would they know that their purely physical properties are causing a video monitor in the next room to act like a human?

For (3), I believe most behaviorists and computationalists would say that since the mechanism is running the same algorithm as your brain, it has the same internal experiences as you. A mentalist who acknowledges the possibility of internal experience in a mechanism would say that even if it starts out in an exact same state as your brain was in when it was scanned, the following experiences of the mechanism immediately begin diverging from your experiences so the mechanism quickly becomes different from you. In other words, the mentalist would say that it is the past experiences of the mind which determine what it is like to be that mind, and not just the algorithm that is running to produce the mind.

For (4), some behaviorists would say that since "what it is like to be you" is entirely determined by your behavior, there is no significant difference between you and the mechanism. Some behaviorists would say you should have no fear of having your brain destroyed at this point because "you" would continue on in the computer. A mentalist would point out that even if there is something that started out exactly like the "you" at the time of the scan, it quickly became its own self. So destroying your brain at this point would kill the original you, leaving a different self running in a mechanism. This different self might have your memories and may act somewhat like you, but it is a mimic; it is not you.

Answer 3

Answer stitched together from original comments of the OP as best I could, quoting directly without edits (except typos):

The physical state of my brain gives me consciousness. A simulated physical state is not a real brain state though. And it needs the real stuff to be conscious. Simulated spiked currents are not real spiked currents. A simulated storm doesn't blow. A dreamt lightning doesn't really flash. A dreamt person isn't really conscious, no matter how real he behaves. It's the very matter that you simulate that is needed for mind, not the causal structures. The causal structures in my brain can only be found in a real brain. Similar causal structures doesn't imply the same properties. Only the properties of causal structures are equal. How can a processes of a brain be copied in a computer, which is a very different material structure?

The obvious question is of course, can you replace a neuron by a an artificial structure? A single neuron can't even be created in a lab. Let alone 80 billion in a working body. So how can an artificial structure replace it? If not, then the argument is invalid. The only thing having the same functionality as a neuron is...a neuron. These can't be constructed. If someone could show me how to replace my neurons I would be convinced. Then there would be a re upload to the new brain. Until such a device hasn't actually been constructed we must assume it doesn't exist.

I think there are better reasons to think that no physical process can be simulated on a computer. Contrary to the brain, which can do that perfectly. Maybe you can create consciousness (which I doubt, as I think it's a natural appearing process) but how can you upload yourself? I can't imagine to go on living like that.

It's clear that if someone scans my brain and simulates it on a computer I can live side by side with the computer. When I die, I won't live on. Maybe other people think it's me but I won't notice anymore. If I would see with a scan that there is electronics inside, I would know that there is no real brain inside, but a simulated one. So without a mind. The argument that if other people can't know the difference isn't a a strong enough argument. I don't think that behavior is a sign of consciousness. A body might show no signs of consciousness and still be conscious. Behavior can deceive.

A simulated brain is by definition not a brain. A simulation is not the same as the thing simulated. Which means a simulated mind is not a mind.

I believe in the afterlife in the sense that I believe there is an infinite series of big bangs. Every time a new universe takes off there will evolve a new version of me and you and everyone. All matter in this universe will be turned into photons which loose all their energy and in a new big bang a new history of particles will come to be leading to my birth. I think my mind will be uploaded again in a brain in a body in the new universe. The gods made matter to evolve like that. They had their reasons. I think only naturally occurring reality simulating creatures can be mindful.

Summary

1. It is impossible, in principle, to reconstruct a working biological brain in a computer simulation, because some (unmentioned) physical properties of the brain must be assumed essential to the emergence of mind, because only nature so far has produced minds.

2. It must be assumed impossible to recreate even a single neurons properties artificially, because it has not been done before, and seems quite difficult really if you think about it.

3. Computer simulations might produce similar outcomes as reality, but not whatever mind is. So from the outside, a computer-driven agents behavior might be indistinguishable from a conscious mind, but that could be an illusion. It must be assumed that only nature can create mind, because we can so far only observe nature producing mind.

4. Some mix of new-agey esoteric beliefs that are altogether way more incredible than brain copy and inconsistent with previous statements.

Answer 4

In Science Fiction and in principle? Yes. In practice? No.

The problem is not that you would need to simulate every neuron in the person's brain and all of the sensory input it receives from the body. The problem is the initialisation of the brain state. The behaviour of the brain is (IMHO) deterministic, but highly sensitive to it's initial state, and we won't be able to measure that without destroying or modifying the state we are trying to measure (possibly the process requires some dicing? ;o). It's a bit like Heisenberg's uncertainty principle?

Answer 5

Some hurdles are embodied cognition, and extended cognition.

Our bodies have many interacting levels, like the gut-brain axis where digestion and mood are linked, which may be difficult, or even undesirable, to emulate. Also our hormone systems, and evolutionary pressures we are unaware of like the proposal economic inequality increases likelihood of being violent, proposed to be adaptive in relation to competition for fewer secure niches. In terms of embodied cognition, a great deal of our biology would need to be simulated for the same kind of mind. It has been observed that we don't make aeroplanes that flap, or cars that gallop, so we might simply choose not to do this, and accept simulated minds as similar but different to organic minds.

Extended cognition relates to how the things we interact with in the world, tools like calculators, and social systems like money, augment and structure our cognitive abilities. In this sense, not only the brain would have to be emulated, but modes of interaction like hands, which shape our minds. Ultimately, this might have to involve an android or cyborg body. There would be a lot of challenges involved, not least having enough energy storage - a human brain is estimated to be equivalent to all the computing power in the world sometime around the turn of the millenium, but runs on only 20 watts - very hard to emulate! And again, we might simply choose not to do this. For instance, the human speed of physical reactions is quite slow, and we might simply prefer to allow higher speeds of interaction, even though they would produce different kinds of mind. Or we might choose to have lots of useful software, augmenting the simulated mind's skills in the way a calculator extends our own.

Mechanisms of consciousness are an issue. It's a highly speculative area, so probably not worth going too deep into details. Penrose with OrchOR suggests the Human Brain Project may be underestimating human cognitive power by at least two orders of magnitude, because of the harnessing of quantum processes in neurons. For me the really interesting point Penrose makes, is how poorly we understand memory, and that brain damage cases suggest that memory is somehow stored in a distributed way. One suggestion is memory as a kind of 'holographic image', where the whole image is stored everywhere, but each part also contributes to a higher resolution whole. See the Holonomic Brain theory. It's interesting how poorly understood memory is. I'm not stuck on these or other mechanisms, I just point at them as indicative of how the uncertainties about mechanisms likely involve many orders of magnitude, and unknown amounts of time to solve.

It's unlikely any mechanisms of consciousness generation, memory, embodied or extended cognition, pose fundamental issues for simulating human brains, because of the idea of universal computation and Turing Machines (inc quantum ones). If a series of operations can run on one machine, however complex, we expect another machine of a different type to be able to simulate it, with limitations only relating to speed.

But, these issues may pose substantial hurdles in terms of the scale of the challenge, and make it of unknown degree - for instance the energy density might make it very hard to fully localise such a brain in a human sized body. So, we can expect a long phase of not-very-human synthetic consciousness, possibly with traits or characters that make them seem more human than they really are. We might claim a brain has been uploaded, there may be subjective continuity, but the hardware would actually see it develop quite differently to if it had stayed in a body.

It's likely the phase of experimentation will fundamentally change humanity, for instance getting to experience life without the impact of hormonal cycles. We may find that a loss, in subtle ways in the long run, and simulate them. But we may not, and just as we don't expect to make aeroplanes that flap, or horses that gallop, simply embrace synthetic minds as a new thing, cultivated from human minds, but not any longer human.

Perhaps the picture of convergence and merging of human minds and Artificial General Intelligences, as advocated by for instance the Neuralink project, may be more appropriate.