# What's it like to be an uploaded human with limited computing power?

In most science fiction where humans are simulated or uploaded onto computers, the humans are simulated with perfect fidelity. But this seems unlikely-- humans are very expensive to run. In Robin Hanson's Age of Em, humans that are too poor to run at full speed are just slowed down; probably the only reason that they aren't "lossily compressed" instead is that in that universe, we haven't made advances in neuroscience and must treat stored minds as black boxes.

In a world with the following:

• Abundant but finite computing power
• Trillions of simulated humans
• A value system that looks favorably on lossy compression of human thought, or the elimination of particularly expensive modules, to make room for more people
• The scientific understanding to do this

what mental faculties of the average person would actually be compressed/eliminated? Since computers are trillions of times more efficient than humans at arithmetic, imagine removing the ability for humans to do arithmetic, replacing that functionality with a software library. Or depowering parts of the brain related to social interaction when one is alone. There are surely better candidates. I'm especially interested in examples that result in interesting negative changes to subjective experience, that are plausible based on current scientific evidence.

Related question on Biology.SE

• In particular, most humans almost certainly do not have a special ability to do arithmetic, which could plausibly be turned off; our ability to do arithmetic is simply an application of our ability to concentrate on a task for more than a few seconds, and of our ability to remember and apply recipes. (Some few humans do seem to possess an innate special ability to do arithmetic; but those are very rare and definitely not neurotypical.) In general, the question assumes the existence of a high-level quantitative model of the human mind; no such model exists currently, much less a good one. – AlexP May 9 '20 at 23:41
• @AlexP most humans almost certainly do not have a special ability to do arithmetic in the context, this is to say "Most humans will benefit in being compressed and granted access to the arithmetic library of the hosting computer" :grin: – Adrian Colomitchi May 10 '20 at 0:27
• I would strongly suggest replacing hard-science (requires citations/evidence) with science-fiction. Because we haven't managed yet to model the brain of any mammal, much less a human, so in all honesty there can be no citation/evidence. – Adrian Colomitchi May 10 '20 at 0:33
• For one creative take on this question, watch Tom Scott's Life: the singularity (ruined by lawyers). Very entertaining, and references this idea directly. – Sam Weaver May 11 '20 at 3:11
• Did you watch Upload recently? – Thomas Weller May 11 '20 at 11:21

Lag.

The

world seems

really

fast.

and sometimes

you....

....

because you are

sharing

the server with

so

.....

....

• This explains the world speeding up on me last Tuesday. – user535733 May 10 '20 at 0:48
• If all humans and their simulated environment were slowed down they wouldn’t even notice it. – Michael May 10 '20 at 9:04
• @Michael they would, because the physics engine would now appear to run faster. And if they are synced with the physics engine, there will be significant discrepancies at chunk boundaries. If the electromagnetic field is simulated, there will be significant diffraction at the chunk boundary. If the body is simulated, crossing the chunk boundary will be quite unpleasant. Energy should still be conserved (the time warp is time invariant), but momentum won't be (a chunk with very slow time would essentially act like a solid wall). – John Dvorak May 10 '20 at 9:53
• ... and if you can predict the server load in each chunk and predict when the time factor changes, you should even be able to extract energy from those changes. Throwing a ball into a slow chunk and back after the chunk speeds up should do. – John Dvorak May 10 '20 at 9:55
• @Michael if different chunks - parts of simulated spacetime - run at different clock speeds, you will notice from the inside! If the time warp coefficient between two chunks changes you won't know if one slowed down or the other sped up, but with a large pool of chunks to compare the clock speed between, you will be able to paint a pretty good picture. – John Dvorak May 11 '20 at 9:34

In point of fact, what you are positing is exactly what the human mind/brain does. The circulatory system, as efficient as it is, can only provide the brain with part of the total energy it needs. To accommodate the deficit, the brain diverts blood flow to areas of the brain that are active. This is the principle of MRI scanning.

This is most evident in sleep, where major functions of the brain are shut down, and brain energy requirements are low. It is also the reason why someone might have to call your name several times, before they 'get your attention'.

Our brain has a very efficient central control system to monitor attention and priority brain functioning. It is akin to your AI using a control monitor level that 'loads' specific programs to do specific tasks into 'main memory' in order to execute them. Most computer hard drives might be full of hundreds of programs, or apps, that do various tasks, but your computer or smart phone only loads and executes a select few. It is not just computing power that is a limitation, but data and program storage power as well.

On the other hand, the body has several processors that run specific programs continuously, independent of the brain's control. Such things as breathing and heart beat, homeostasis, and balance. So to do modern computers. The printer, screen, remote keyboard, all have processors that run independently of the main CPU, to off-load sub-tasks. The computer CPU, for instance, does not move the cursor around, that is done by the graphics processor.

So, basically, anything in the human and the human simulator can be shut down under certain conditions, provided essential services are off-loaded and you have an independent always-functioning control processor. When I am listening to music, I close my eyes. When I am reading, I shut down attention to hearing. When I am sleeping, I shut down mechanisms for maintaining balance. When I am daydreaming, I shut down both hearing and seeing. In point of fact when I meditate, I pretty much shut down everything. Except (unless I am in a coma) the central background monitoring system - the one that brings me awake at a sudden noise - and the autonomic processors and sub-systems (my liver still processes stuff, for instance).

Given the size of humans, and thus the availability of data storage space everywhere, including spaces in the feet, torso, and so on, I suspect that future computer-humans will store most of their functionality into programs that are brought up to the main processor as needed. Just as in parallel distributed processing, the trick will be in the system monitor program being able to load the programs as necessary.

And given what we have done with cloud systems, I suspect a lot of routines will be offloaded into cloud systems, sort of like the chrome books today.

You're going to run into two problems trying to idle or compress brain systems.

## Evolution Has Already Favored a Lean System

Our brains use heuristic shortcuts, as well as a task manager which itself idles and brings into activity different subsystems. This is to minimize energy costs; there's a finite number of of thoughts you can manage on a handful of nuts and berries or a mouthful of animal flesh. It is for this cause that people are bad at the sort of "brute force" computing that machines do, though people often falsely attribute this to us just not being very smart. Beating nature tends to be more difficult than people anticipate.

But what about the subsystems we don't need because we're in a simulation?

## The Brain Has a Habit of Sprawling Into Its Own Subsystems!

Close your eyes and picture the last thing you ate. If someone was doing a brain scan as you did that, your visual processing center just lit up (and maybe another system or two). Even profoundly blind and/or deaf people end up using the unoccupied sense-processing real estate, for other tasks. If you eliminated "unused" regions of the brain, there may be unanticipated consequences for basic function.

Did that one guy use his visual center to help him find the right word? Idling that subsystem just gave him aphasia. Plus, he can't perform visual abstraction tasks anymore. And that autistic lady who wasn't even good at recognizing faces (different, dedicated subsystem)? She is no longer able to quickly distinguish between images of different bacteria! And your blind mathematician has suddenly lost his mathematical intuition.

Worse than that, when systems fail, the brain rewires itself, so forcing subsystems to shut down will have permanent structural implications as the remaining structures change and adapt to losses, even if those losses were intended to be temporary.

Where Am I?

If you're exploiting brains to work symbolically rather than in an actual environment — that is to say, walking around in a virtual world and doing things isn't too important for the people organizing this so much as getting the thoughts they need — you can probably (in the vague science-fiction sense, not the one that I can back up with evidence) get rid of most of the functionality of the brain stem and cerebellum. You don't need a heartbeat, you don't need to feel pain, you don't need to know where your hands are.

If the people in charge didn't even need to give their emulated friends images to work with, and are planning on using the brainpower for purely abstract reasoning, with some risk, your hypothetical neuroscientists might be able to work outward from there, shutting off sensory processing in the occipital and parietal lobes.

The results would be far more unnerving than just being a "brain in a vat"; these people would lose track of where they were, not know whether or not they had bodies, and wouldn't understand what vision or touch could provide for them. But you could still feed them conceptual data and get responses back, like querying some server, assuming there are things you need to poll these minds on that you couldn't just get from computers.

Compress individual viewpoints and knowledge.

A common design principle for software engineering is to not repeat yourself, and this would be useful in your scenario. Instead of individuals having their own opinions on a subject, which are all pretty similar to their peers anyway, the individuals could simply be subscribed to a common "ism". For example, not everyone needs a highly personal and unique reasoning around the story of creation. So, just assign one of 'n' creation myths to that person, which is the closest match to their previous belief.

As another example, for those not actively interested in politics, just assign them the generic democrat, republican, anarchist belief pattern, and do away with the tiny individual details. They will all have the exact same responses to political questioning. This is a highly efficient type of lossy compression which can be applied to tastes in music, food, childhood memories, sexual fantasies, and so on. The trick is to make people more generic, more identical, and reduce the differences between them as you recycle the resources. It allows people to retain their depth and complexity, as its not the thought processes but the diversity of opinion which gets compressed. The simulation computers respond to Trumps latest tweet once, calculate the result, and all the Trump supporters think exactly alike on the matter.

I myself am not particularly interested in nature, and being able to identify individual birds, leaves and animal droppings. So just assign "generic city dweller" module to me to cover that area. If i do take a nature walk I'll recognize a couple of common plants and birds, but I would have to really look carefully to spot that my knowledge of this is exactly the same as some other city dweller. That is the lossy compression at work.

Naturally, upper class individuals could emerge, where they acquire more computing resources to simulate more unique and individual nuanced versions of one or more areas or opinions. In such a situation, it would be a luxury item to have a custom personality that stands out from the generic masses.

I think you're examining the wrong thing. It's much easier to simulate a human mind than it is to simulate a universe the human mind observes.

Humans - particularly young humans - have eyes and brains capable of pretty high FPS. And when you throw in the fact that eyes don't just detect in 'frames', but have a sensitivity that is continual/blurring (aka, your eyes detect a bright light for longer than the photons are actually hitting it.) Realistically, you're going to have to simulate whatever the eye is looking at with very high frame rate.

... and that's combined with a very high degree of precision. The total megapixel size of your area of focus (ignoring the peripheral vision) is up to 15 megapixels. That's twice the area of a 4k display.

Then you come to the lighting/visuals itself. Video games have come a long way, but there hasn't been one produced that is indistinguishable from reality.

And finally... the physics of the universe the human is experiencing. How detailed of elements are you simulating? If they hold a mound of sawdust in their hand and blow on it, are you simulating every particle? If they walk along the beach, is each particle of sand simulated? Are the particles displaced by their footprints kept? What about when it's raining - are you simulating each water droplet and how its water molecules interact with each grain of sand? When they boil a pot of water to make pasta, are you tracking each individual steam molecule so you know whether or where it condenses? For what it's worth... we're still at a point where we're bragging about simulating a single molecule. Whereas... we're actually taking stabs at being able to simulate the human brain already.

I think you'll find that it's not "de-rez'ing parts of a brain" so much as its "de-rez'ing parts of the universe". Nobody's at the beach? Don't track the sand particles, and regenerate/extrapolate once someone enters the scene. Only one person at the park? Don't simulate each individual grass blade, and use a simulated effect for everything more than 50 meters away. Etc.

• I don't think OP mentioned that they live in a simulation - just that their brains are simulated. They might, for example, have remote controlled robotic bodies; or their senses might be replaced entirely with something else. – Vilx- May 10 '20 at 21:51
• Eyes work at about 60FPS but they immediately discard almost all the information they receive. Even some image processing such as edge and motion detection is done in they eye itself before the result is sent to the brain. Only the central portion of the visual field is transmitted in anything like high-resolution. – Ben May 12 '20 at 15:18
• @Ben - look at the megapixel article I linked. At the very top, it says "576 megapixels" - which, after the effects you're talking about, reduces it down to "up to 15 megapixels". 15 megapixels is the reduced amount. – Kevin May 12 '20 at 16:21

That kinda depends on what the simulation is like.

Not a complete answer, but if the mind is uploaded, then there are a whole host of things that need not be done, I think. Like controlling the heartbeat and breathing and digestion and various reflexive responses, etc. All the subconscious stuff. No point in calculating things related to body control when there is no body to control. So if your simulation involves a brain floating in "nothingness" then you can pretty much shut down all senses and balance control and what not else.

The remaining few senses that are used for communication can be wired more directly and thus require less processing.

# You're drunk - go home

With insufficient computing power your 'brain' struggles with fine motor control, your speech becomes slurred and you become clumsy. Your reactions slow down as each computation takes longer. Your decision making becomes worse as you have less processing power to make good choices.

Without the high level processing the world feels calmer, no more micro-calculations buzzing through your mind; the edges of objects get blurred, which is frankly a blessing having been looking at the world in ultra-high definition all week and noticing individual specs of dust on objects all around the room. At the extreme end you are immersed in a world of block-colours, soothing noises and a base feeling of happiness. Imagine going to the pub, but they 'pay' you to drink!

People in this world love spend a Friday night on low-power, so they can switch off, relax and hangout with friends. No-one remembers much the next morning.

Addiction is still a problem, underclockers mumble at you on street corners. "Got some credits? I've got this job lined up tomorrow, but I need enough credits to process well. Help a guy out. Get me get back on my feet". Their friend nearby is slumped against a wall in a 33MHz trance.

Slow

A "kind" society might just slow down the overall computing speed of any individual process (person) and their experienced world to the point that it seems normal to them, but is maybe 1/60th speed to an outside observer.

Avoid it with limiting abilities and emotions, by providing an "in-life shop". You want to feel love? Pay for it, loads. You want to have a good sense of orientation, that's not for free obviously. Maybe you even can earn money, by voluntary shut down a ability of you basic set your are provided with by law. Shut down an arm for cash.

All perks, abilities and so on cost money, and there are many. Most people only can afford a few of them, keeping the system requirement low. Only the few rich ones, can afford a high system load. This keeps the system usage in bounds.