Yes. This is the same argument that audio nuts make for analog recording. It's known to be bogus. (Yes, 16-bit CD audio has resolution problems for soft passages, and early filters for the sampling rate were not too good. We're past that.)
One of the few concrete examples of a complex analog computer system still used in recent decades was the F-16 flight control system. It's a four channel fly by wire stabilization and control system, all analog. It was, at the time, the most advanced flight control system, and it's still well thought of. That's from the 1970s, and modernized F-16s use a digital replacement.
For several decades, full authority digital flight control systems were disfavored in aerospace because there were no analysis techniques to be sure they were bug free. There are ways to analyze an analog computer system to be sure that the test case set is sufficient, and that behavior will be smooth continuous between the test case points. Eventually that problem was solved for digital flight control systems, and now everybody goes digital.
Analog computers still crop up in all sorts of places these days but have become very niche and are typically a small part of an otherwise larger system. It's rare to see a general purpose analog computer outside of a very small number of research labs and FPAAs exist but are expensive novelties. As digital processors continue to get better and we develop better ways to work with them, there just isn't the need for the cost and effort required to make analog computers.
Some common places where you might see one is in audio driver amplifiers where it is common to implement a bit of trans-linear logic at the output stage to reduce distortion. Same in some high quality power supplies.
Sometimes very high performance sensor systems will have an analog pre-processor to perform some calculation on the incoming signal before handing it off to the digitizers and DSP. Think multi-microphone arrays.
I think the question that comes to mind is whether it's possible to take GPU-styles architecture and give it 100x more power or more by replacing bits with approximate voltage levels that are "fuzzy" but statistical guarantees to their performance, along with gates that allow the values to be filtered back to zeros and ones.
The thing is the "neural architectures" seem to have been more or less failure through requiring the user to accept one particular neural net structure while the standard GPU seems to have succeeded through being generic-enough to use for a variety of tasks. So some sort of analogy-GPU should also have a similar generic model - but of course it seems likely that the creator of this stuff will want to impose their special model.
Edit: More or less like the D-wave "quantum computer" except not milking quantum hype and being reconciled to being understood as "massively analogue"
The "fuzzy but statistical guarantees" is basically "error correction". Which is (handwavily) the difference between digital and analog computing. What you are describing is the engineer's definition of a digital computer. Admittedly, there might be interesting performance gains if we use less stringent statistical guarantees... which happens to be what is happening each time we make the transistors smaller and more susceptible to noise.
C.f. the paper that introduced the distinction between analog and "statistically guaranteed" digital in the case of classical computers (before it people were arguing that you can not build a scallable classical digital computer because of noise): by von Neumann http://www.sns.ias.edu/pitp2/2012files/Probabilistic_Logics....
P.S. FYI DWave is not a scallable quantum computer. They have another "quantumy" word for what they do, so that they can keep the hype without angering people that are trying to build an actual quantum computer.
Yes. Concentrating on error correction reveals the difference between digital and analog computing.
Digital computers have error correction. Analog computers don't.
Some small number of problems might be solved efficiently using analog computers, but they will never take over the role of general purpose computers because of error correction issues.
Too bad the author didn't substantiate his claims that analog computing will make a comeback.
I found this article [0] with a real world application, but no performance comparison to a digital computer. I also take issue with the claim that a transistor has infinite possible states, and that we're ignoring most of them. This doesn't take into account real world limitations of components' precision and noise.
I agree that analog computing is probably more efficient, or can be more efficient if predictability is sacrificed.
However, there are societal implications to this. We prefer the processing steps be traceable and dissactable for accountability and managing the distribution of tasks/parts in terms of accountability. We may not accept a higher degree of "rogue machines" to gain average efficiency.
However, I suppose a given country or group could accept the tradeoff to gain a military advantage, which could spell chaos. They may accept a higher degree of rogue battle bots in order to win via average efficiency, or at least be willing to take the gamble that the theory is true. The chance of a high-stakes or borderline suicidal leader/dictator eventually coming on the scene is historically high, leading inadvertently to run-away human-flattening bots. Maybe that's the answer to the Fermi Paradox.
Perhaps I missed it, but the point of the article seems to be that eventually a complex "neuro" computer will be to complicated to understand and produce unaccountable results. The Author makes a few poor assumptions about analog vs digital computing and seems to ramble a lot, but ultimately, his main point doesn't have much to do with either.
Digital computation has been able to rise to this level of complexity because we can precisely predict/repeat outcomes because of exact numbers and boolean logic. I could not fathom anything like what we have using analog. I was reading the article, waiting for some type of plan for how to tackle analog computing, but it never came. A thought provoking article, but I'm not holding my breath for analog.
I suspect that the 'excess' precision of digital computers could be able to be retargeted towards some other use, negating any benefit that a analog computer had.
> It is entirely possible to build something without understanding it. [...] Our relationship with true A.I. will always be a matter of faith, not proof
WTF. I have no problem with building something without understanding how or why it works, but I do have problem of using something without at least some sort of guarantee on its behavior.
You may, but the vast majority of the population is already way past that with an iPhone and search bubbles, apps . . ..
And honestly, even if you're in tech, this feild is so broad and so many people are doing so many cool things that there's almost no way to keep up with it unless you're a ludite. If the case we are talking about involves building a single giant computer that no one knows how to use, you won't have much say in that.
The main problem with old school analog computers is that they were based on electricity which ended up causing rift (imprecision that gets worse over time).
The claim about relative performance of continuous variable models versus circuit models is completely unsubstantiated. Especially given that most uses of the continuous variable systems is to encode discrete qubits on top of them with GKP/cat/binomial codes. Is there any Complexity Theory work published about continuous variable models?
And photonic systems do not magically fix the noise issue. Noise grows in a fast non-linear fashion with the size of the system, so the "constant factor" noise suppression you gain from switching to photonic systems is quickly washed away.
> Especially given that most uses of the continuous variable systems is to encode discrete qubits
Current uses, maybe.
> Is there any Complexity Theory work published about continuous variable models?
"Complexity and Real Computation".
> And photonic systems do not magically fix the noise issue. Noise grows in a fast non-linear fashion with the size of the system, so the "constant factor" noise suppression you gain from switching to photonic systems is quickly washed away.
Is there anything published on the fact that this cannot be overcome?
Just looking at the wiki pages for Real Computation is enough to see it is not a physically realizable model. For more detailed discussion of the problem you can see the essay "NP-complete Problems and Physical Reality". To quote from it "The problem, of course, is that unlimited-precision real numbers would violate the holographic entropy bound".
At every level of physics there is a bound on precision, from boring things like classical macroscopic thermodynamics and noise, to quantum noise, to bounds that emerge in speculative theoretical physics. Basically, anything capable of encoding an infinitely precise real number in a finite amount of space will collapse and form a black hole.
In case this is not convincing enough, to your question about how this noise can not be overcome: if there was a method that can overcome the noise asymptotically in photonic systems, then that method would work in electric systems too. And there is actually such a method: turning the computer into a digital computer thanks to error correction codes.
The claim that Real Computation can be realized in our universe is comparable to the claim one can construct a perpetual motion machine or some other generator of free energy. They are both preposterous given our understanding of physics. And yes, I would celebrate if either of one turns out to actually be possible, but incredible claims require incredible evidence.
> Just looking at the wiki pages for Real Computation is enough to see it is not a physically realizable model.
This computer is literally at the physical limit reality can take. This number is actually quite high, although not unlimited. If you could make a computer faster by 10^30, ok I might not need unlimited. What if this number could be 10^300?
Furthermore, and this is important, this computer allows for a fundamentally different type of computation.
> At every level of physics there is a bound on precision, from boring things like classical macroscopic thermodynamics and noise, to quantum noise, to bounds that emerge in speculative theoretical physics.
Correct, this can be accounted for. Don't press on details, you won't be satisfied with the answers. Can we talk about would be possible if this computer were possible and work backwards? As a mental exercise.
Also what if I'm fundamentally more interested in probabilistic computation and this error can actually be a foundation of my computation.
> then that method would work in electric systems too.
Electricity is fundamentally more "unstable". This error compounds. This of the difference in attenuation rate in electric vs optical media. Where does this attenuation come from?
> And there is actually such a method: turning the computer into a digital computer thanks to error correction codes.
Nope. Think of it as averaging unstable signals. The result is very much continuous.
You are getting too hung-up on the infinity aspect. Let's talk more about what sort of programming model this would allow for.
A Turing machine technically has a tape of infinite length, while current computers don't. Does that mean that no Turing machine has ever been constructed? Does the answer to this question matter?
Also, and this is important, compared with a normal computer, this computer would not heat up. Current CPU's can't get much larger because they can't dissipate heat fast enough. What if you could have a CPU of the size of a cube of the volume of one cubic meter?
> Can we talk about would be possible if this computer were possible and work backwards? As a mental exercise.
Certainly! This type of thought experiments is how much of physics progresses. But their results should be contemplated seriously. Assuming it is possible to prepare a system that works like your computer, it will immediately collapse into a black hole, because it breaks the holographic principle (and plenty of other bounds).
Similarly we can imagine what happens if FTL was possible: time paradoxes. Or if we could measure both position and momentum: UV catastrophe.
> You are getting too hung-up on the infinity aspect.
The infinity aspect is what makes Real Computation more powerful than other types of computation. It is also what makes it impossible in our universe. It has absolutely nothing to do with the type of (countable) infinity that is the length of a Turing machine tape. You can make asymptotic statements that are useful about Turing machines or logic circuits. The only useful thing you get out of Real Computation is strictly after you take the limit to infinity. Otherwise you have old boring less-powerful analog computers (which are nonetheless marvels of engineering and their creators deserve praise).
> Electricity is fundamentally more "unstable".
This is just nonsense.
By the way, the majority of continuous variable hardware is actually in the microwave regime, which is closer to the fields of electrical and radio engineering, than to the fields of lasers or THz systems.
Lastly, yes, I completely agree that such hardware might be interesting in many cases. But claiming that Real Computation is possible (as oppossed to just admitting that small analog computers are occasionally useful) is far fetched. Claiming that such hardware can be built when it flies in the face of everything we know about physics is like claiming you can build a perpetual motion generator. As I said, extraordinary claims require extraordinary evidence.
Please believe me, it is extremely rewarding to learn the details of these arguments, much more rewarding than the existence of Real Computation. In comparison with the real world, "Real Computation" is a boring cop-out. The last strip here describes the idea well: http://calamitiesofnature.com/
> Certainly! This type of thought experiments is how much of physics progresses. But their results should be contemplated seriously. Assuming it is possible to prepare a system that works like your computer, it will immediately collapse into a black hole, because it breaks the holographic principle (and plenty of other bounds).
You are condescending.
> Similarly we can imagine what happens if FTL was possible: time paradoxes. Or if we could measure both position and momentum: UV catastrophe.
This has been done extensively and not much new can be brought to the table. Not many people talk about photonic analog computation, so indulge me.
There are people who research this field extensively (continuous variable quantum computation). Your whole argument just relies on a section from a Scott Aaronson paper. Can I ask what other research do you use as a foundation of your worldview?
Are you familiar with work of say Alessio Serafini?
You read too much theory of complexity, it doesn't have all the answers. Can I ask about your stance on say Lie theory as it relates to quantum physics and why exactly is it that Lie theory provides a good foundation reasoning about quantum phenomena?
Can we talk about anticommutavity? And infinitesimals?
> The infinity aspect is what makes Real Computation more powerful than other types of computation.
10^300 is infinity. 10^3000 is infinity. 10^300000000 is infinity. It's that you can make each atom do a lot more computation and much more valuable computation. I'm not interested in your arguments about infinity. I legit don't care, stop trying, I find the premise of the question flawed.
And last but not least it is "Real" computation. It's the limit of computation you can "fit into reality". Are you legitimately saying that current architectures are at the physical limit that reality can take? There's nothing realer than reality.
Who has actually attempted to build this computer? Like we haven't really given this a good shot. Things are impossible until they are not.
I certainly did not mean to, and I would blame the limits of this medium for that false impression. I was serious in my comments about thought experiments.
> Can I ask what other research do you use as a foundation of your worldview?
Aaronson is good at explaining things, so I use him as a reference, but look at any quantum computing textbook and the same argument will be present there. I am not familiar with Alessio Serafini, but there is plenty of other work in the field, from Xanadu Inc. to GKP's to encodings used at the institute where I work.
> it doesn't have all the answers.
This is always true, but you have given fewer answers.
> Can I ask about your stance on say Lie theory as it relates to quantum physics and why exactly is it that Lie theory provides a good foundation reasoning about quantum phenomena? Can we talk about anticommutavity? And infinitesimals?
Sure? If you are doing a litmus tests of my knowledge, yes, I do understand those topics and would be happy to discuss them. You can find my email on my profile page. I am sincerely always interested to talk about this with curious people.
> 10^300 is infinity. 10^3000 is infinity. 10^300000000 is infinity.
This is really missing the point. When things grow exponentially (like compounding errors) the numbers you are quoting are small.
> I'm not interested in your arguments about infinity. I legit don't care, stop trying, I find the premise of the question flawed.
This is simply intellectually dishonest. How do you expect a discussion to reach the truth if you disregard arguments you dislike?
> Are you legitimately saying that current architectures are at the physical limit that reality can take?
No, I am just saying that any technique that might be used to improve practical analog computers would be already of use in digital computers. I am also saying that making a better real world analog computer has nothing to do with the theoretical model of Real Computing which is unphysical. It is not a continuous change - the gulf is discrete and qualitative.
> Who has actually attempted to build this computer?
Analog computers in various media, including light, have been built for a century (more recently with light). Xanadu seems to be trying to commercialize them in the case of quantum continuous variables, and they will probably produce some interesting hardware, but claiming that their (potential) hardware is more powerful than the usual model of quantum computing (also only potential for now) is unscientific. My employer (Yale Quantum Institute) is also employing continuous variable systems in other contexts.
With respect, I think most comments here are missing Dyson's point (perhaps because it was somewhat poorly made).
I don't think his point is about whether the integrator and analog electronics will be resurgent in the next century, and analog hardware will be common.
I think Dyson is talking about the complex network of the modern world, where humans interact with computing machine, and with each other - humans influence computers, and computers come back around and influence humans. I think his "future of computing" is a future where human society and culture is decided based on the interplay between humans and our machines, and this decision is an "analog computation" made by a massive scale hybrid computer that no one has intentionally designed or understands.
You can see examples of this already, with youtube recommendation engines influencing the belief systems of millions (billions?) of people across all kinds of subjects, and with our thoughts frequently dominated by whatever happens to show up on our phones.
Maybe I’m just simple but couldn’t you say it was always case that the human hive mind is a bit like a giant analog computer ? Not sure why the digital stuff is important for the analogy.
> In analog computing, complexity resides in network topology, not in code. Information is processed as continuous functions of values, such as voltage and relative pulse frequency, rather than by logical operations on discrete strings of bits.
...
> Individually deterministic finite-state processors, running finite codes, are forming large-scale, nondeterministic, non-finite-state metazoan organisms running wild in the real world. The resulting hybrid analog/digital systems treat streams of bits collectively, the way the flow of electrons is treated in a vacuum tube, rather than individually, as bits are treated by the discrete-state devices generating the flow. Bits are the new electrons. Analog is back, and its nature is to assume control.
> Say, for example, you build a system to map highway traffic in real time simply by giving cars access to the map in exchange for reporting their own speed and location at the time. The result is a fully decentralized control system. Nowhere is there any controlling model of the system except the system itself.
...
> Even in the age of all things digital, this cannot be defined in any strictly logical sense, because meaning, among humans, isn’t fundamentally logical. The best you can do, once you have collected all possible answers, is to invite well-defined questions and compile a pulse-frequency weighted map of how everything connects. Before you know it, your system will not only be observing and mapping the meaning of things, it will start constructing meaning as well. In time, it will control meaning, in the same way the traffic map starts to control the flow of traffic even though no one seems to be in control.
In these passages, "The Computer" that's running is the network, a meta-computer, made up of the individual "computing elements" if you will that are the physical pieces of hardware that most anyone would point to and call "a computer". And as you've recognized, that's Dyson's point: No matter what those little elements are made of, the overall picture is analog. If we want to understand how "Algorithms" are influencing the world, we can't think of them as "discrete computing algorithms". And finally, the emergent behaviour out of this can lead to large-scale control effected on our society without being designed in, and even if nobody is thinking to effect control.
The article seems to be redefining analog and digital to mean something different than what they're traditionally taken to mean. It's not clear to me that the new definition implicitly being proposed is very useful or that if it is useful it is helpful to try and repurpose these words for it.
That's no different to conventional history and culture. It just has computers in it.
Dyson seems to have only just worked out that evolution, culture, and politics are emergent.
Which is why the argument makes no sense. Either digital systems add something new and unique to the mix, or they don't.
If they don't, then what's the point of the piece?
If they do, blather about analog metazoans is irrelevant. The problem becomes one of understanding emergence without obfuscation and hand-waving.
In fact so-called emergent behaviour is not the problem. Not even slightly.
The biggest digital systems we have now have been consciously and deliberately designed to monitor and manipulate human behaviour.
There's nothing emergent, surprising, obscure, or quasi-mystical about either the outline goals or the specific techniques used to achieve them.
Instead of resorting to obscurantism, it would be far more useful to invent a new kind of "digital democracy" where the goals of large software corporations were open to democratic oversight.
Of course that's not going to happen while free market excuses and rhetoric are used to keep democratic oversight well away from these corporations.
> no different to conventional history and culture. It just has computers in it.
I would argue that it is different, because it has computers in it.
> The biggest digital systems we have now have been consciously and deliberately designed to monitor and manipulate human behaviour.
They've been designed to make users click on this or that or watch or scroll more, but I disagree that all the consequences of those designs were or are understood. I don't think anyone intentionally designed a system to make kids watch creepy YouTube videos endlessly or not get vaccinated or think the earth is flat, they designed a system to "maximize engagement" and these things happened (maybe or not as a consequence).
There is this concept of emergent control called a "metasystem transition" [1] which is what Dyson is really talking about. Conventional history and culture has never before been shaped by the - ever shortening, ever accelerating - techonomic feedback loops that we see today. Cybernetics has been summoned like Yog Sothoth in "The case of Charles Dexter Ward".
>Which is why the argument makes no sense. Either digital systems add something new and unique to the mix, or they don't.
I'm sorry but this runs counter the observed historical reality.
For reference, consider data transmission interfaces (USB, SCSI, memory busses, display interfaces, etc): "Is parallel better than serial? Is serial better than parallel?". The history clearly indicates that, every few generation of interfaces there's a big switch-over between parallel and serial, and back, and fore again. The approaches simply have different trade-offs, and at different point of evolution of our understanding and available technology it makes sense to pick one over another. Neither of them is inherently "ultimately better".
Likewise digital vs analog computing. It used to be[1] that analog (calculations and computations) ruled the day. In case we forgot already, there were not only analog (mechanical and electrical and electronic) programmable computers and fixed-function, complex calculators, but many an everyday object are analog electronic calculators in time and frequency domains. Cue the humble superheterodyne radio, and the analog color TV. Cybernetics used to be a thing, too.
Then came the digital electronic computation, which allowed us to translate lambda-calculus theory verbatim into discrete bits (sorry!) of control words and data. However digital is a leaky abstraction over switching electronics [2]; all the transistors aren't digital, and - Moore's law notwithstanding - digital view of analog semiconductors is modern equivalent of squaring a circle. Or describing astronomy with ever more epicycles. It works, but at the cost of the ever-growing complexity, and a bit of suspension of disbelief.
We already are having renaissance of continuous-valued computation: the present day crop of early quantum computers. Both the computing elements (quantum gates) and also the control programs and data are in the complex number domain.
On the other end of the spectrum, we're using ever more neural networks, which are analog abstractions over our digital computers. At some point somebody will start asking questions. Maybe even invent (or make feasible) the long-awaited memristor.
I'm sure we'll be shifting back to digital in a few decades though.
--
[1] glossing over discrete computation approach of the early pioneers like Charles Babbage
[2] and switching electronics, whether semiconducting or valve-based, is merely a leaky abstraction over high-gain analog amplification, but I digress
Just a few thoughts I've been mulling for a while about this topic:
Machine learning is something that I believe can take advantage of analog computing. A machine learning algorithm does not need highly precise or accurate representations. Most current implementations of such processing units use fewer bits (usually 8).
However, even if we use fewer bits, the engineering effort (design, layout, lithography, etc.) that goes into making the processing unit still assumes that those few bits are error free. The manufacturing process treats it like any other digital circuit. It assumes data processing part should be fault free (e.g. treat MSB and LSB the same). Digital circuits also demand higher power compared to analog versions.
If an analog circuit can be designed for such algorithms, not only could it be much faster, it will probably consume far less power. With a super high bandwidth consuming little power, an analog processing chip may give us a much better playground to try advanced algorithms. The materials can then be optimized and we might end up with something like a brain.
Brains (all animals) process far more information for the power they consume.
Digital circuits give us low level reliability and so they are really good for simple control. Analog/biology don't give us that. But they can give us a high level reliability while delegating the low level reliability to digital counterparts.
I think you’re wrong about the ML precision. You need highly precise for most recursive machine learning tasks because you’re compounding errors otherwise.
Typically you can’t even use floating point representation: not accurate enough.
Disagree. https://arxiv.org/abs/1805.08691 demonstrates 8-bit architecture for a pre-trained CNN provides more than acceptable results with lower latency and higher throughput than a higher precision version.
Analog can actually be much more precise than 8-bit since there's no quantization noise (yes, this is not A2D, but any intermediate results can take middle values that will only make the final value more precise than digital signal).
Perhaps I'm missing the more grandiose point, but I think his message is simple: digital computing is a handy abstraction for humans who like to count, but computing is essentially analog.
This layman likes to refer to something about genetic algorithms and magnetic flux for reference on this.
Quantum, Analog, and Fuzzy Logic - are different terms that I believe some bright person in the future will prove as describing exactly the same underlying phenomenon.
(Also, if this could be accomplished, the next step in human evolution might be proving that the entire universe is a giant Analog computer, but that's Sci-Fi at this point in time...)
There is no such thing as "analog". Everything in this Universe is digitized, down to elemental particles inside atoms. What you experience as "analog" is just digitization with a very fine grain, or if you like it, with better sampling. So no, the future of computing is not analog at all, it will still be digital, just with better sampling, aka quantum computing.
48 comments
[ 2.3 ms ] story [ 101 ms ] threadOne of the few concrete examples of a complex analog computer system still used in recent decades was the F-16 flight control system. It's a four channel fly by wire stabilization and control system, all analog. It was, at the time, the most advanced flight control system, and it's still well thought of. That's from the 1970s, and modernized F-16s use a digital replacement.
For several decades, full authority digital flight control systems were disfavored in aerospace because there were no analysis techniques to be sure they were bug free. There are ways to analyze an analog computer system to be sure that the test case set is sufficient, and that behavior will be smooth continuous between the test case points. Eventually that problem was solved for digital flight control systems, and now everybody goes digital.
Some common places where you might see one is in audio driver amplifiers where it is common to implement a bit of trans-linear logic at the output stage to reduce distortion. Same in some high quality power supplies.
Sometimes very high performance sensor systems will have an analog pre-processor to perform some calculation on the incoming signal before handing it off to the digitizers and DSP. Think multi-microphone arrays.
I think the question that comes to mind is whether it's possible to take GPU-styles architecture and give it 100x more power or more by replacing bits with approximate voltage levels that are "fuzzy" but statistical guarantees to their performance, along with gates that allow the values to be filtered back to zeros and ones.
The thing is the "neural architectures" seem to have been more or less failure through requiring the user to accept one particular neural net structure while the standard GPU seems to have succeeded through being generic-enough to use for a variety of tasks. So some sort of analogy-GPU should also have a similar generic model - but of course it seems likely that the creator of this stuff will want to impose their special model.
Edit: More or less like the D-wave "quantum computer" except not milking quantum hype and being reconciled to being understood as "massively analogue"
C.f. the paper that introduced the distinction between analog and "statistically guaranteed" digital in the case of classical computers (before it people were arguing that you can not build a scallable classical digital computer because of noise): by von Neumann http://www.sns.ias.edu/pitp2/2012files/Probabilistic_Logics....
The paper that did the same for quantum computers 50 years later: by Shor http://www-math.mit.edu/~shor/papers/good-codes.pdf
P.S. FYI DWave is not a scallable quantum computer. They have another "quantumy" word for what they do, so that they can keep the hype without angering people that are trying to build an actual quantum computer.
Digital computers have error correction. Analog computers don't.
Some small number of problems might be solved efficiently using analog computers, but they will never take over the role of general purpose computers because of error correction issues.
I found this article [0] with a real world application, but no performance comparison to a digital computer. I also take issue with the claim that a transistor has infinite possible states, and that we're ignoring most of them. This doesn't take into account real world limitations of components' precision and noise.
[0] https://news.mit.edu/2016/analog-computing-organs-organisms-...
This is impossible right now, no one is really making analog computers to make this comparison.
https://arstechnica.com/science/2018/07/neural-network-imple...
https://pruned.blogspot.com/2012/01/gardens-as-crypto-water-...
However, there are societal implications to this. We prefer the processing steps be traceable and dissactable for accountability and managing the distribution of tasks/parts in terms of accountability. We may not accept a higher degree of "rogue machines" to gain average efficiency.
However, I suppose a given country or group could accept the tradeoff to gain a military advantage, which could spell chaos. They may accept a higher degree of rogue battle bots in order to win via average efficiency, or at least be willing to take the gamble that the theory is true. The chance of a high-stakes or borderline suicidal leader/dictator eventually coming on the scene is historically high, leading inadvertently to run-away human-flattening bots. Maybe that's the answer to the Fermi Paradox.
WTF. I have no problem with building something without understanding how or why it works, but I do have problem of using something without at least some sort of guarantee on its behavior.
And honestly, even if you're in tech, this feild is so broad and so many people are doing so many cool things that there's almost no way to keep up with it unless you're a ludite. If the case we are talking about involves building a single giant computer that no one knows how to use, you won't have much say in that.
https://hn.algolia.com/?query=adamnemecek%20analog%20quantum...
The main problem with old school analog computers is that they were based on electricity which ended up causing rift (imprecision that gets worse over time).
A photonic, analog, quantum computer (also called continuous variable quantum computer) https://en.wikipedia.org/wiki/Continuous-variable_quantum_in... is possible and would run circles around discrete quantum computers (those with qubits).
And photonic systems do not magically fix the noise issue. Noise grows in a fast non-linear fashion with the size of the system, so the "constant factor" noise suppression you gain from switching to photonic systems is quickly washed away.
Current uses, maybe.
> Is there any Complexity Theory work published about continuous variable models?
"Complexity and Real Computation".
> And photonic systems do not magically fix the noise issue. Noise grows in a fast non-linear fashion with the size of the system, so the "constant factor" noise suppression you gain from switching to photonic systems is quickly washed away.
Is there anything published on the fact that this cannot be overcome?
At every level of physics there is a bound on precision, from boring things like classical macroscopic thermodynamics and noise, to quantum noise, to bounds that emerge in speculative theoretical physics. Basically, anything capable of encoding an infinitely precise real number in a finite amount of space will collapse and form a black hole.
In case this is not convincing enough, to your question about how this noise can not be overcome: if there was a method that can overcome the noise asymptotically in photonic systems, then that method would work in electric systems too. And there is actually such a method: turning the computer into a digital computer thanks to error correction codes.
The claim that Real Computation can be realized in our universe is comparable to the claim one can construct a perpetual motion machine or some other generator of free energy. They are both preposterous given our understanding of physics. And yes, I would celebrate if either of one turns out to actually be possible, but incredible claims require incredible evidence.
This computer is literally at the physical limit reality can take. This number is actually quite high, although not unlimited. If you could make a computer faster by 10^30, ok I might not need unlimited. What if this number could be 10^300?
Furthermore, and this is important, this computer allows for a fundamentally different type of computation.
> At every level of physics there is a bound on precision, from boring things like classical macroscopic thermodynamics and noise, to quantum noise, to bounds that emerge in speculative theoretical physics.
Correct, this can be accounted for. Don't press on details, you won't be satisfied with the answers. Can we talk about would be possible if this computer were possible and work backwards? As a mental exercise.
Also what if I'm fundamentally more interested in probabilistic computation and this error can actually be a foundation of my computation.
> then that method would work in electric systems too.
Electricity is fundamentally more "unstable". This error compounds. This of the difference in attenuation rate in electric vs optical media. Where does this attenuation come from?
> And there is actually such a method: turning the computer into a digital computer thanks to error correction codes.
Nope. Think of it as averaging unstable signals. The result is very much continuous.
You are getting too hung-up on the infinity aspect. Let's talk more about what sort of programming model this would allow for.
A Turing machine technically has a tape of infinite length, while current computers don't. Does that mean that no Turing machine has ever been constructed? Does the answer to this question matter?
Also, and this is important, compared with a normal computer, this computer would not heat up. Current CPU's can't get much larger because they can't dissipate heat fast enough. What if you could have a CPU of the size of a cube of the volume of one cubic meter?
Certainly! This type of thought experiments is how much of physics progresses. But their results should be contemplated seriously. Assuming it is possible to prepare a system that works like your computer, it will immediately collapse into a black hole, because it breaks the holographic principle (and plenty of other bounds).
Similarly we can imagine what happens if FTL was possible: time paradoxes. Or if we could measure both position and momentum: UV catastrophe.
> You are getting too hung-up on the infinity aspect.
The infinity aspect is what makes Real Computation more powerful than other types of computation. It is also what makes it impossible in our universe. It has absolutely nothing to do with the type of (countable) infinity that is the length of a Turing machine tape. You can make asymptotic statements that are useful about Turing machines or logic circuits. The only useful thing you get out of Real Computation is strictly after you take the limit to infinity. Otherwise you have old boring less-powerful analog computers (which are nonetheless marvels of engineering and their creators deserve praise).
> Electricity is fundamentally more "unstable".
This is just nonsense.
By the way, the majority of continuous variable hardware is actually in the microwave regime, which is closer to the fields of electrical and radio engineering, than to the fields of lasers or THz systems.
Lastly, yes, I completely agree that such hardware might be interesting in many cases. But claiming that Real Computation is possible (as oppossed to just admitting that small analog computers are occasionally useful) is far fetched. Claiming that such hardware can be built when it flies in the face of everything we know about physics is like claiming you can build a perpetual motion generator. As I said, extraordinary claims require extraordinary evidence.
Please believe me, it is extremely rewarding to learn the details of these arguments, much more rewarding than the existence of Real Computation. In comparison with the real world, "Real Computation" is a boring cop-out. The last strip here describes the idea well: http://calamitiesofnature.com/
You are condescending.
> Similarly we can imagine what happens if FTL was possible: time paradoxes. Or if we could measure both position and momentum: UV catastrophe.
This has been done extensively and not much new can be brought to the table. Not many people talk about photonic analog computation, so indulge me.
There are people who research this field extensively (continuous variable quantum computation). Your whole argument just relies on a section from a Scott Aaronson paper. Can I ask what other research do you use as a foundation of your worldview? Are you familiar with work of say Alessio Serafini?
You read too much theory of complexity, it doesn't have all the answers. Can I ask about your stance on say Lie theory as it relates to quantum physics and why exactly is it that Lie theory provides a good foundation reasoning about quantum phenomena?
Can we talk about anticommutavity? And infinitesimals?
> The infinity aspect is what makes Real Computation more powerful than other types of computation.
10^300 is infinity. 10^3000 is infinity. 10^300000000 is infinity. It's that you can make each atom do a lot more computation and much more valuable computation. I'm not interested in your arguments about infinity. I legit don't care, stop trying, I find the premise of the question flawed.
And last but not least it is "Real" computation. It's the limit of computation you can "fit into reality". Are you legitimately saying that current architectures are at the physical limit that reality can take? There's nothing realer than reality.
Who has actually attempted to build this computer? Like we haven't really given this a good shot. Things are impossible until they are not.
I certainly did not mean to, and I would blame the limits of this medium for that false impression. I was serious in my comments about thought experiments.
> Can I ask what other research do you use as a foundation of your worldview?
Aaronson is good at explaining things, so I use him as a reference, but look at any quantum computing textbook and the same argument will be present there. I am not familiar with Alessio Serafini, but there is plenty of other work in the field, from Xanadu Inc. to GKP's to encodings used at the institute where I work.
> it doesn't have all the answers.
This is always true, but you have given fewer answers.
> Can I ask about your stance on say Lie theory as it relates to quantum physics and why exactly is it that Lie theory provides a good foundation reasoning about quantum phenomena? Can we talk about anticommutavity? And infinitesimals?
Sure? If you are doing a litmus tests of my knowledge, yes, I do understand those topics and would be happy to discuss them. You can find my email on my profile page. I am sincerely always interested to talk about this with curious people.
> 10^300 is infinity. 10^3000 is infinity. 10^300000000 is infinity.
This is really missing the point. When things grow exponentially (like compounding errors) the numbers you are quoting are small.
> I'm not interested in your arguments about infinity. I legit don't care, stop trying, I find the premise of the question flawed.
This is simply intellectually dishonest. How do you expect a discussion to reach the truth if you disregard arguments you dislike?
> Are you legitimately saying that current architectures are at the physical limit that reality can take?
No, I am just saying that any technique that might be used to improve practical analog computers would be already of use in digital computers. I am also saying that making a better real world analog computer has nothing to do with the theoretical model of Real Computing which is unphysical. It is not a continuous change - the gulf is discrete and qualitative.
> Who has actually attempted to build this computer?
Analog computers in various media, including light, have been built for a century (more recently with light). Xanadu seems to be trying to commercialize them in the case of quantum continuous variables, and they will probably produce some interesting hardware, but claiming that their (potential) hardware is more powerful than the usual model of quantum computing (also only potential for now) is unscientific. My employer (Yale Quantum Institute) is also employing continuous variable systems in other contexts.
I don't think his point is about whether the integrator and analog electronics will be resurgent in the next century, and analog hardware will be common.
I think Dyson is talking about the complex network of the modern world, where humans interact with computing machine, and with each other - humans influence computers, and computers come back around and influence humans. I think his "future of computing" is a future where human society and culture is decided based on the interplay between humans and our machines, and this decision is an "analog computation" made by a massive scale hybrid computer that no one has intentionally designed or understands.
You can see examples of this already, with youtube recommendation engines influencing the belief systems of millions (billions?) of people across all kinds of subjects, and with our thoughts frequently dominated by whatever happens to show up on our phones.
> In analog computing, complexity resides in network topology, not in code. Information is processed as continuous functions of values, such as voltage and relative pulse frequency, rather than by logical operations on discrete strings of bits.
...
> Individually deterministic finite-state processors, running finite codes, are forming large-scale, nondeterministic, non-finite-state metazoan organisms running wild in the real world. The resulting hybrid analog/digital systems treat streams of bits collectively, the way the flow of electrons is treated in a vacuum tube, rather than individually, as bits are treated by the discrete-state devices generating the flow. Bits are the new electrons. Analog is back, and its nature is to assume control.
> Say, for example, you build a system to map highway traffic in real time simply by giving cars access to the map in exchange for reporting their own speed and location at the time. The result is a fully decentralized control system. Nowhere is there any controlling model of the system except the system itself.
...
> Even in the age of all things digital, this cannot be defined in any strictly logical sense, because meaning, among humans, isn’t fundamentally logical. The best you can do, once you have collected all possible answers, is to invite well-defined questions and compile a pulse-frequency weighted map of how everything connects. Before you know it, your system will not only be observing and mapping the meaning of things, it will start constructing meaning as well. In time, it will control meaning, in the same way the traffic map starts to control the flow of traffic even though no one seems to be in control.
In these passages, "The Computer" that's running is the network, a meta-computer, made up of the individual "computing elements" if you will that are the physical pieces of hardware that most anyone would point to and call "a computer". And as you've recognized, that's Dyson's point: No matter what those little elements are made of, the overall picture is analog. If we want to understand how "Algorithms" are influencing the world, we can't think of them as "discrete computing algorithms". And finally, the emergent behaviour out of this can lead to large-scale control effected on our society without being designed in, and even if nobody is thinking to effect control.
Dyson seems to have only just worked out that evolution, culture, and politics are emergent.
Which is why the argument makes no sense. Either digital systems add something new and unique to the mix, or they don't.
If they don't, then what's the point of the piece?
If they do, blather about analog metazoans is irrelevant. The problem becomes one of understanding emergence without obfuscation and hand-waving.
In fact so-called emergent behaviour is not the problem. Not even slightly.
The biggest digital systems we have now have been consciously and deliberately designed to monitor and manipulate human behaviour.
There's nothing emergent, surprising, obscure, or quasi-mystical about either the outline goals or the specific techniques used to achieve them.
Instead of resorting to obscurantism, it would be far more useful to invent a new kind of "digital democracy" where the goals of large software corporations were open to democratic oversight.
Of course that's not going to happen while free market excuses and rhetoric are used to keep democratic oversight well away from these corporations.
I would argue that it is different, because it has computers in it.
> The biggest digital systems we have now have been consciously and deliberately designed to monitor and manipulate human behaviour.
They've been designed to make users click on this or that or watch or scroll more, but I disagree that all the consequences of those designs were or are understood. I don't think anyone intentionally designed a system to make kids watch creepy YouTube videos endlessly or not get vaccinated or think the earth is flat, they designed a system to "maximize engagement" and these things happened (maybe or not as a consequence).
[1] http://pespmc1.vub.ac.be/MST.html
[2] https://en.wikipedia.org/wiki/The_Case_of_Charles_Dexter_War...
I'm sorry but this runs counter the observed historical reality.
For reference, consider data transmission interfaces (USB, SCSI, memory busses, display interfaces, etc): "Is parallel better than serial? Is serial better than parallel?". The history clearly indicates that, every few generation of interfaces there's a big switch-over between parallel and serial, and back, and fore again. The approaches simply have different trade-offs, and at different point of evolution of our understanding and available technology it makes sense to pick one over another. Neither of them is inherently "ultimately better".
Likewise digital vs analog computing. It used to be[1] that analog (calculations and computations) ruled the day. In case we forgot already, there were not only analog (mechanical and electrical and electronic) programmable computers and fixed-function, complex calculators, but many an everyday object are analog electronic calculators in time and frequency domains. Cue the humble superheterodyne radio, and the analog color TV. Cybernetics used to be a thing, too.
Then came the digital electronic computation, which allowed us to translate lambda-calculus theory verbatim into discrete bits (sorry!) of control words and data. However digital is a leaky abstraction over switching electronics [2]; all the transistors aren't digital, and - Moore's law notwithstanding - digital view of analog semiconductors is modern equivalent of squaring a circle. Or describing astronomy with ever more epicycles. It works, but at the cost of the ever-growing complexity, and a bit of suspension of disbelief.
We already are having renaissance of continuous-valued computation: the present day crop of early quantum computers. Both the computing elements (quantum gates) and also the control programs and data are in the complex number domain.
On the other end of the spectrum, we're using ever more neural networks, which are analog abstractions over our digital computers. At some point somebody will start asking questions. Maybe even invent (or make feasible) the long-awaited memristor.
I'm sure we'll be shifting back to digital in a few decades though.
--
[1] glossing over discrete computation approach of the early pioneers like Charles Babbage
[2] and switching electronics, whether semiconducting or valve-based, is merely a leaky abstraction over high-gain analog amplification, but I digress
Machine learning is something that I believe can take advantage of analog computing. A machine learning algorithm does not need highly precise or accurate representations. Most current implementations of such processing units use fewer bits (usually 8).
However, even if we use fewer bits, the engineering effort (design, layout, lithography, etc.) that goes into making the processing unit still assumes that those few bits are error free. The manufacturing process treats it like any other digital circuit. It assumes data processing part should be fault free (e.g. treat MSB and LSB the same). Digital circuits also demand higher power compared to analog versions.
If an analog circuit can be designed for such algorithms, not only could it be much faster, it will probably consume far less power. With a super high bandwidth consuming little power, an analog processing chip may give us a much better playground to try advanced algorithms. The materials can then be optimized and we might end up with something like a brain.
Brains (all animals) process far more information for the power they consume.
Digital circuits give us low level reliability and so they are really good for simple control. Analog/biology don't give us that. But they can give us a high level reliability while delegating the low level reliability to digital counterparts.
Typically you can’t even use floating point representation: not accurate enough.
Sounds like someone's never shopped for guitar amps before.
I also like to share an instruction video on mechanical computers on every suitable occasion: https://www.youtube.com/watch?v=s1i-dnAH9Y4
This layman likes to refer to something about genetic algorithms and magnetic flux for reference on this.
I mean this: http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.50....
(Also, if this could be accomplished, the next step in human evolution might be proving that the entire universe is a giant Analog computer, but that's Sci-Fi at this point in time...)
old paper. 1996. Fragile.