0
$\begingroup$

The year is 2175, and at this point, there are compact ASIC systems (no larger than a 2024 smartphone) that use a combination of powerful dedicated processors and some quantum computing subsystems cooled and protected from the outside world using room-temperature superconductors and superfluid helium (an art mastered in the late 21st century). These ASICs use their immense processing power and quantum parallelism to do exactly one thing: simulate an entire human brain down to the molecular level, with its inputs being the brain scan of a now-deceased human (alive at the time of the scan) and a few random numbers to differentiate the simulated brain from the original, and the output being the signals emitted by the newly-conceived brain that are used to express the artificial general intelligence's behavior.

In short, I have a mixed classical/quantum computer system the size of a smartphone that is capable of molecular-level human brain simulation for the purposes of producing artificial general intelligence, and I have this technology by 2175.

Here's the question: is this a realistic timeframe given Moore's law and computing limitations? That is, is it feasible that a computer system this compact and this powerful could be produced by 2175? In context, this technology does exist by 2175 in the plot in question, and I would ideally like not to have to invoke technobabble or artistic license to make this technology a reality.

Some specifications:

  • Quantum computers can, by 2175, be made as powerful as modern (2024) classical computers are, so that 128 GQB (gigaqubyte) quantum computers are readily available.

  • Engineering of individual electrons is possible at this time, so that by manipulating an electron's spin a single atom can be used as a memory cell and graphene sheets can be used as memory arrays with the memory capacity in bits being equal to the number of atoms on the sheet.

  • Processing power progresses according to Moore's law until transistor size reaches the size of an atom, at which point it can progress no further.

$\endgroup$
13
  • $\begingroup$ Recommend this: en.wikipedia.org/wiki/Mind_uploading#/media/… Blue lines are estimates as to which processing benchmarks render brain emulation doable. It estimates a 1000 USD personal computer could perform full-brain emulation by 2100 presuming proteome simulation is needed for an accurate simulation. Although it assumes processing power doubles every 18 months rather than every 24 I still believe said 1000 USD PC would hit such benchmarks by 2175. I don't feel comfortable giving this as an answer but it implies yes in my opinion. $\endgroup$
    – KEY_ABRADE
    Jan 28 at 20:02
  • 4
    $\begingroup$ "I would ideally like not to have to invoke technobabble": The first paragraph of this question is technobabble, and of a rather obnoxious kind. P.S. Whatever path we will eventually take towards building artificially intelligent machines, simulating a slow, grossly inefficient and massively error-prone wetware system will not be it. $\endgroup$
    – AlexP
    Jan 28 at 20:49
  • 1
    $\begingroup$ @KEY_ABRADE: If the OP was referring to the simulation of human brains and leave it here, that would be fine and believable. Unfortunately, they are not. They are referring to the simulation at molecular level of a system comprising tens or hundreds of billions of billions of molecules. That's not a simulation; that's an emulation, and there simply no reason whatsoever to do that. Why emulate a slow and error-prone chemical system when its operation can be easily simulated by abstracting out the chemistry? $\endgroup$
    – AlexP
    Jan 28 at 21:07
  • 1
    $\begingroup$ Asking about the development of Hansonian ems is totally fine and I don't really get the argument there. The problem with this question is that it is subjective; the future of technological development is fundamentally unpredictable on timescales like this. You can't tell which answers might be more or less based in an accurate model of the future because none of them can have one. $\endgroup$
    – parasoup
    Jan 28 at 21:28
  • 2
    $\begingroup$ All due respect to @Pelinore's answer (I'm a EE), 99.9% of the world's technology was invented in the last 150 years. You've given us another 150 years. But to also be fair, you're asking a silly question. You've developed an entirely fictional technology to rationalize an entirely fictional consequence and now you want us to tell you the time frame is realistic. Silly. The movie The Martian is set in the "near future." Can we do it? Nope, but it's completely believable. And that's what we do here. Suspension of disbelief. You got that before asking. $\endgroup$
    – JBH
    Jan 29 at 5:44

3 Answers 3

6
$\begingroup$

Probably no, and the problem is not timeframe

These two are probably incompatible:

  • no larger than a 2024 smartphone
  • simulate an entire human brain down to the molecular level.

The brain is some 10 times heavier than a smartphone, so for every 10 atoms in the brain to model, you only have 1 or 2 atoms available on the computer.

For example, there are trillions of smallish molecules with no more than 20 - 30 atoms (eg. neurotransmitters), and you definitely need more than 2 - 3 atoms per each of them to

  • store, and
  • access, and
  • manipulate the information in complex ways with usable speed.

If you are using technology whose development can be extrapolated from what we have today, the ratio is unfortunately quite unfeasible due to physical restrictions. Smartphone-size device is probably too small to store such amount of information and process it without some handwavium. When it appears, that cannot be predicted.

$\endgroup$
1
  • 2
    $\begingroup$ Trillions pffft. Three quarters of the brain is water, so that for a typical 1.2 kg female brain there are some 25 trillion of trillions of water molecules. And each of them has a position (3 numbers), an orientation (three numbers) and a velocity (another three numbers). $\endgroup$
    – AlexP
    Jan 29 at 14:39
2
$\begingroup$

No this timeframe is not realistic.

Integrated circuit - Wikipedia

Moore's law is an observation of how fast we are getting the known established technology of silicon transistors smaller, it says nothing about what the limit is and it says nothing about what we might use in it's place to surpass that limit so for the purpose of your assumptions drawn from it it is meaningless.

It does not and has never applied to anything other than silicon transistors.

theoretical limits

transistors are now approaching their theoretical limits when it comes to the sizes of their gates. Below roughly 5 nanometers, silicon can no longer control the flow of electrons from sources to drains because of a quantum-mechanical effect known as tunneling.

How small are transistor gates now?

Wafer World 24 Nov 2021

They're currently roughly 7-10 nanometers in size, and they're on track to shrink even further to 5 nanometers. Most devices at this point are sub-100 nanometers.

We are practically at the theoretical limits of the materials we are using already and should probably be there within the next 20 years alone.

Applying Moore's law beyond the point at which theoretical maximum of the material is reached by what would seem to be more than an extra 130 years and thinking they will keep getting smaller despite having reached that absolute limit (130 or more years before) is just not going to work.

$\endgroup$
6
  • 2
    $\begingroup$ OP noted that "processing power progresses according to Moore's law until transistor size reaches the size of an atom, at which point it can progress no further", which implies they aren't sticking to Moore's law all the way. Moreover, TSMC is producing 3 nm transistors and plans to produce 2 nm transistors, which suggests quantum tunneling is not a hard barrier to sub-5 nm transistors. $\endgroup$
    – KEY_ABRADE
    Jan 28 at 20:28
  • $\begingroup$ Hmm .. I'll have to stop editing and do some more current reading .. cheers for the comment > after one last small edit to take the sting out of the last bit ;) $\endgroup$
    – Pelinore
    Jan 28 at 20:30
  • $\begingroup$ Hmm (again) The term "2 nanometer" or alternatively "20 angstrom" (a term used by Intel) has no relation to any actual physical feature (such as gate length, metal pitch or gate pitch) of the transistors I'm not so sure now that you have anything there ;) will have to dig deeper but that doesn't sound promising for your comment @KEY_ABRADE $\endgroup$
    – Pelinore
    Jan 28 at 21:13
  • $\begingroup$ @KEY_ABRADE: The 5 nm process does not make 5 nm transistors, not even close. See the nice Wikipedia article which kindly provides physical sizes; for example, transistor gate pitch remains somewhere around 50 nm, and a (four-transistor) SRAM bit-cell is about 150 by 150 nm. $\endgroup$
    – AlexP
    Jan 28 at 21:21
  • $\begingroup$ @AlexP Ta [thumb] $\endgroup$
    – Pelinore
    Jan 28 at 21:22
0
$\begingroup$

Nothing to do with Moore's law but I feel it could be done before then.

The first quantum calculations on atomic orbitals using the Hartree-Fock method were largely done by Hartree's retired father using a mechanical calculator. The modern computer is billions of times faster these days, but the real advances in comnputing speed and complexity have come from advances in computing techniques, rather than just making the basic logic elements smaller and faster.

Take modern machine learning, for example. That was predicted since the first computers, but it has only started giving practical results in the last few years. We have developed programs that allow us to design learning systems and train them, and these tools have helped us design new tools. These advances are mostly very specific pattern-matching algorithms that provide you with other data values like the ones you trained it to recognise. This is not intelligence, but you could probably make intelligence from enough of these. And when you have developed them, you can embed the trained model in something a lot smaller. If you take a picture on your phone, the face recognition software may have come from a machine learning model.

I have an intelligent computer. It is highly parallel. It weights about 3 pounds. It runs on less than 100 watts. I keep it under my hat. Intelligence can be done with much more modest resources. At present we use computer tools such as GPUs that were designed for other things, and program them in ways that allow us to manipulate and understand the learning process.

It is interesting to speculate how small we might make such a brain if we could get around the limits of organic life. You could have have molecular-level circuits. You could have a bus architecture that allows high connectivity without actually having neuron A touch neuron B. You might have a 1 gram brain that works on microwatts. If we want this badly enough - say for exploring the stars using solar sails - we could probably do this a lot earlier than 2175.

$\endgroup$

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .