Face It, Your Brain Is a Computer
nytimes.com
nytimes.com
My brain might work like a computer (computation et al), but the word computer evokes actual existing computers to a layman -- and my brain doesn't have a von neumann architecture, nor does it run something similar to our software.
A neural network does computation of course, but we wouldn't call it a "computer" in the sense we use the word when dealing with, well, computers.
We can simulate the universe with astounding accuracy on many levels (though not all) with computers. Furthermore, when studying the universe deeply we see astoundingly computer-like traits[1]. But to express this by saying the universe is a computer is very misleading.
Similarly, artificial neural networks simulated by computers achieve huge strides in AI, accomplishing things many said would always be impossible (e.g. art[2]). Also, we see many computer-like traits in the brain. But to express this by saying the brain is a computer is very misleading.
[1] http://lmgtfy.com/?q=the+universe+is+a+computer+simulation
[2] http://googleresearch.blogspot.ca/2015/06/inceptionism-going...
"Computer" doesn't automatically equal "von neuman architecture".
> nor does it run something similar to our software
A lack of software-equivalent is a pretty good criticism, though.
No, but that's what most people are familiar with (if they are familiar with computer architectures in the first place), so it pretty much equals that for them.
They're not. There's really only two kinds of people here—those who are unfamiliar with computer architecture (99% of the population), and those who are familiar with Von Neumann architecture, and know that it's not the only game in town.
However, this is the problem. Computers are designed to interpret or translate the messy analog world where nothing is exactly 1 or 0, into a precise, deterministic digital world where everything is either 1 or 0. And I think it's highly likely that the brain does not work in this manner whatsoever, and that in fact, in the translation of reality into 1s and 0s, you've lost far too much to properly "compute consciousness"
Would it be possible to build a synthetic brain out of some form of "analog computer", capable of consciousness? Almost certainly. But since we have no idea what sort of architecture or paradigm that would use, or how we would program it or whether it would even "execute instructions" in a way remotely similar to digital computer software, such a declaration becomes meaningless.
But neurons totally do work by encoding information with discrete signals.
https://en.wikipedia.org/wiki/Action_potential#Process_in_a_...
Action potentials are discrete events, and just like hi or lo in a silicon computer actually correspond to a range of voltages, the exact voltage-vs-time curve of an action potential likely has considerable flexibility.
"The frequency of action potentials (APs) generated by neurons is correlated with different events; however, it does not mean that such events are ‘encoded’ in the brain using the firing rate or any other ‘temporal code ’ (e.g. interspike interval)…
…Simultaneous firing of action potentials (synchrony) in a neuronal ensemble increases the interaction of many- body systems and implicitly boosts the computational power [7]. The well accepted model of ‘activation function’ that describes artificial neurons simulated on Turing Machines does not approximate the process of ‘solving equations’. Therefore, the popular claim that current artificial neurons and neural networks are ‘realistic’ models of their biological counterparts is misconstrued and untrue."[0]
However, the relevance of Turing model is questioned even in case of present-day computing [33] [34]. Indeed, any computing machine that follows a Turing model would be highly inefficient to simulate the activity of biological neurons and experience an increased slowdown. Since the super-Turing computing power of the brain has its origins in these ‘strong’ interactions that occur inside neurons, current models have missed the most important part. Simply, Nature doesn’t care if the N-body problem has analytical solutions [36] or can be simulated in real time on a Turing machine [37].
...
While previous models have attempted to represent Hamiltonians using Turing machines [35] the paper [1] shows that the Hamiltonian model of interaction can represent itself a far more powerful model of computation. Turing made an important step forward; however, there is no need to limit natural models of computation to Turing models. In this sense, the new framework of computation using interaction is universal in nature and provides a more general description of computation than the formal Turing model. In other words God was unaware of Turing's work and has put forward a better model for physical computation in the brain.
The amount of money being wasted trying to make special FMRI's and custom EEG caps is lost on me when attempting to solve equations along variable boundary conditions seems like it should come first…
Here's the paper the author wrote which that paper is branched from: http://www.researchgate.net/profile/Dorian_Aur/publication/2...
As such it is transmitting a fourier transform of the signal it is transmitting between the cells (but with different basis functions, not cosines). When a pulse is received, this will result in a sudden jump of the action potential, and then slowly die down. If a second pulse comes while the first is still in action it gets multiplied by the current value of (this particular) input potential. Rapid pulsing (3-4 ms between pulses) means something along the lines of "multiply membrane potential by 10 for every pulse", whereas a pulse width of ~350ms means "keep the membrane potential constant". Any one of these pulses may push a neuron "over the edge" and have it transmit a pulse of it's own (or a series of pulses as pulsing the axon lowers the membrane potential by a factor. This may or may not cause it to go below the trigger level, and so sometimes the axon needs to pulse multiple times to get the membrane potential back below the trigger).
This is a beautiful way to encode simplifications of analog signals. It's log scale in the sense that 2 rapid pulses encode that the function value should be an order of magnitude larger than 2 slow pulses. It is extremely sparse (pulsing speeds rarely exceed 1hz, and only for short times) so it is quite energy efficient. Your brain quite literally does nothing at all unless there's actual information to process. I find it a bit misleading that wikipedia says that the frequency of pulses encodes the info : it's true, but makes one think this frequency is constant, when in fact it is changing rapidly. Does the word frequency really make sense when any 2 measurements of pulse frequency will be different ?
Normally one would call pulses digital, because they are 1 or 0. Values in between are interpreted as being either 1 or 0, definitely not analog. But the time spacing between the pulses is what matters, and this is measured to astonishing accuracy by a chemical reaction. A measurement like that, one would normally call analog. So it doesn't really match either signal type ... It is very different from any analog or digital signal I know about.
One might say your neurons contain an encoded, but uncompressed mp3 of the signal they're transferring between neuron membranes : it's transmitting in the frequency domain, but without the compression that reduces the zeroes, so it's sending very long series of zeroes, interspersed with the very occasional one. This works well because transmitting a zero is done by doing nothing at all.
The question is whether the brain offers a more complete framework than all the models of computation we know today (which are equivalent to one another). To put it another way : is there a function that the brain can compute that a Turing Machine cannot. We know of no such function now, but have ruled out yet that there is none, although most computer scientist agree with the Church-Turing hypothesis.
Unless I've missed something, it's not actually been shown that the universe itself is Turing-computable, therefore there's no reason to assume everything within the universe could be equivalently simulated on a Turing machine.
It could also be that in theory, consciousness is Turing computable, but not in practice. For example just computing a protein fold for 1 millisecond takes hundreds of ASICs days of compute time. This might give some clue that complex physical wetware processes are not able to be efficiently simulated in real time on any realistic Turing machine.
You act as though the only datatype usable by computers is bool. That's ridiculous. Yes, computers are binary, but they can compute analog things just fine via floating or fixed point arithmetic to arbitrarily high precision, among many other things. To claim that the brain has a fundamental advantage over a Turing machine is to claim that there's some magical quantum (or otherwise) advantage in the brain's physical system that is un-simulate-able by a Turing machine.
You can't even argue that analog provides superior precision to digital representation of continuous values, because the analog part of the brain is represented with discrete units (accumulation of neurotransmitter molecules in various parts of the neuron)! In other words, our brain actually uses fixed point math (granted, incredibly complicated fixed point math :))
Again, please point to any research you know of indicating that the universe itself is fully Turing-computable.
Of course you can always see quantum phenomena affecting outcome, you just have to rule those out as neligible / much lower than noize threshold.
https://en.wikipedia.org/wiki/Computable_function#Uncomputab...
Citation needed.
Instead, it processes information as it comes, this information comes during the processing of previous information and the output partly loops back to the input. Short-circuits everywhere, no clock, etc. I'm unconvinced but I could not refute the argument formally.
Note that this does not contradict the Church-Turing Hypothesis : if you restrict the brain to one input, some processing steps with a pen and paper without arrival of new information, and an output, then it is very possible that the brain can only work out Turing-computable problems.
Computers do this all the time. Simplest example is "cancel" button everywhere.
Therefore, we lack formal way of satisfactorily compare the brain to a Turing Machine.
Information comes to brain via neurons, and those have finite speed, it's not as they had quantum-like properties. Signal in; signal out, with delay and altered magnitude.
I, too, think that the brain is 'a computer'. But the formal definitions of computation we have today require the computer to be left alone for some time between the moment the input is given and the moment the output is due. The brain does not work like that, ergo we lack a formal model of how the brain operates, and thus the definition of 'computable function' -which is based on our model of computation- is less general than what the brain can do.
e.g. We have no proof that the only way neurons communicate with one another is via the synapses.
I fail to capture the difference.
And we don't, and worse, we may not ever have access to these concepts, as Hofstadter and others have argued. It may not be possible to represent the idea of ideas in an idea.
That of which we cannot speak that thereof we must remain silent, or something.
I'm confused... is the counter-argument AGAINST the brain being BEYOND a Turing machine (if I am to interpret your sentence literally)?
Or by this double negative do you actually mean the counter argument is AGAINST the brain being ONLY a Turing machine? (Because that's what the rest of your post sounds like.)
> and the output partly loops back to the input.
So? Turing machines have no issues with feedback loops; what would lead you to think they do?
I think he said it partly to tweak his MIT AI Lab audience (this must have been about 1982). But he had a point: a lot goes on in the brain that is not digital.
Computers are designed to emulate brains. More specifically, to perform a certain category of brain activity (calculation).
To date, a brain is still needed to make a computer do anything. No brain, no software; no software, all we have is an inert lump of metals and plastics.
We should be quite wary of reversing metaphors in this fashion. 19th century psychology contained a number of metaphors derived ultimately from the steam engine, for human consciousness, and for reality. All we've done is update the machine; the comparison is still backwards.
The distinction between analog and digital computers doesn’t seem to be so important to me. Any computation that an analog computer can do, can also be done with digital computers and vice versa. Digital computers can process continuous-time signals as precise as analog computers, and analog computers can do logic. Which kind of computer you use depends on convenience with respect to the resources and computational problem in hand.
The “brains don’t download software” part also isn’t convincing. Brains are capable of manipulating symbols, but the rule set needs to be learned in the case of arithmetic, or selected, in the case of natural language according to inputs. Also, software doesn’t have to be like “1. Do this, 2. Do that, 3. End”, like the article implies. The brain can be running on a game loop, planning future actions according to some stochastic utility function.
Also, I find it useful when thinking about brains to separate the functions that are done unconsciously from those that are not. To me, unconscious functions of the brain seem more like using specialised hardware. Of course, those parts are almost certainly doing computation to full their functions, but it is “not abstract symbol manipulation according to some learnable rules” that people associate with consciousness.
But that is only one level. There can be additional layers of function, which can be influenced by sub-atomic entities, like elementary particles. We still don't know how deep this goes. This is, what most likely, provides depth of our consciousness.
Our biological hardware is built out of finite atoms. Consciousness unfolds out of "electrochemical circuits". Electricity == consciousness.
We have what is called "human energy" - emotions are one example of expressions out of it.
For AI, to be human like, it has to simulate physical behaviour of human energy.
I see, that out of this energy behaviour simulations can emerge what I call "computational psychology". That is, computer could analyze and decode human behaviour and "understand" like psychologist does this.
Thing is, that many people don't see these subtle patterns - they are learned over lifetime with different success rates. Once they are aware of these patterns, they can function more efficiently as a human. Such "psychologist as app in your pocket" potentially can raise overall quality of society.
http://www.amazon.com/On-Intelligence-Jeff-Hawkins/dp/080507...
I think to postulate that there's a yet undiscovered subatomic elementary particle that gives rise to consciousness is hogwash. "Consciousness" or awareness of self has been shown in other animals. Humans are distinct in their ability to couple self-awareness and toolmaking.
There is no "human energy" it's the same material as in all other animals, we just have more of it. The "human energy" could be classified as distinct fuzzy algorithms found in humans which aid pattern recognition.
Yes, animals have that same substance, but at lesser scale.
I make distinction between pure intellect, which computer can do fine, and other component, which I doubt that it can be accessed by computer without connecting silicon with biological material.
(2) A universal computer is a computer that can simulate any other computer with some bounded slowdown in performance (the actual important thing about such computers is that they can simulate better computers).
So, while the brain seems to have a universal computer (since it can design and run universal computers), it’s not obvious that a universal computer can simulate a brain without infinite slowdown. Some evidence against brains being equivalent to Universal Turing Machines is that some brains can perceive real numbers. Thus, it’s conceivable that even if the mind (the thing that is realized by a brain) is a partially ordered set of states, it’s possible that the states themselves are not computable.
Now, I am not claiming that we cannot eventually build a brain, just suggesting that maybe the only thing that can simulate a brain is another brain.
In this case that might mean (a) providing enough information to explain why escalating the question is necessary to the discussion, as opposed to a closet insult, and (b) using impersonal language. For example, you already changed your wording from "do you know what science is" to "what is science". That's a big difference.
[1] http://news.nationalgeographic.com/news/2014/11/141114-autis...