4 ms·

I didn't read the whole thing, but I did find parts of it surprising. Some things that jumped out at me: > during conscious states, the electrodynamics of the

by abeppu 5y ago

I didn't read the whole thing, but I did find parts of it surprising. Some things that jumped out at me: > during conscious states, the electrodynamics of the cortex are poised near a critical point or phase transition and that this near-critical behavior supports the vast flow of information through cortical networks during conscious states But we've had a lot of success building artificial systems that support a "vast flow of information" which I would expect are mostly nowhere near such a boundary. Stuff moves around within a periodic structure, and moving more info has corresponded to moving more information per period. Why would we not expect a pile of neurons with a ginormous number of connections to be able to move a lot of information even through largely periodic behavior? > we found that the Lempel–Ziv complexity of the model’s simulated electrodynamics (with noise inputs) was maximal when the deterministic component of its dynamics were poised near this critical onset of chaos I guess if the measure of "information richness" is based in compressibility of electrical activity ... is it surprising that it isn't higher in the chaotic phase? Why wouldn't one expect the chaotic system to produce as much or more random output?

derefr 5y ago
> Why wouldn't one expect the chaotic system to produce as much or more random output? Because neurons as systems don't like random input. Noisy random input doesn't make (most) neurons fire. Rather the opposite, actually. Neurons aren't (usually) simple threshold comparators of a single input; they're arbitrarily-complex gate-like recognizers, that fire in response to detecting signals in their combined inputs, or even upon matching temporal firing patterns within the signals of their combined inputs. In that sense, if you tune every component in the system to produce more noise (chaos), you decrease the signal-to-noise ratio, and so the components that react to "important recognizable changes in the signal" within that noise have a harder time recognizing the thing they're looking for, and so they recognize it less, and so switch less. --- By trite analogy: suppose you're a security guard. Your job is to recognize+report "real crimes in progress", while not reporting "false alarms." (Pretend that the police charge for false dispatches in your town, so your boss wants to optimize shrinkage vs. call budget.) Also, your boss has decided to "keep you on your toes" by introducing "noise" into your computerized security system. Sometimes it'll show a burglary in progress when there isn't one. Sometimes it'll show that everything's fine, when actually the boss has secretly sent in a white-hat burglar of his own. How willing to believe the security cameras should you be, to optimize your output signal to downstream demand? As it turns out, you should probably try to figure out the false-positive vs. false-negative rate you're seeing reported by the downstream, and then calibrate your confidence up/down until you're making a balanced number of Type 1 vs. Type 2 errors. That level of confidence is the "critical point" — where the noise in the system is equally likely to help your detection (by "correcting" a false negative/positive, turning it into a true positive/negative) as it is to harm. Moving away from that critical point, in either direction, makes your output less useful as a downstream signal.
scythe 5y ago
>I guess if the measure of "information richness" is based in compressibility of electrical activity ... is it surprising that it isn't higher in the chaotic phase? Why wouldn't one expect the chaotic system to produce as much or more random output? Chaos can be highly regular. For example, the limiting case of the logistic map, x -> 4x - 4x^2, is chaotic, but also has an exact solution x[n] = sin^2(arcsin(sqrt(x[0])) * 2^n). It's not hard to see why this still obeys the usual properties of Lyapunov doubling etc, but a simple double exponential doesn't seem like the sort of thing that leads to consciousness! From an information-theoretic point of view, the nth bit of x[0] becomes relevant at roughly step n, which means there is a steady "trickle" of information into the system instead of the more complicated dynamics we expect of a "computational" system -- in fact, if we consider the binary expansion of (arcsin(sqrt(x[0]))/pi) we find that only a finite number of bits have an effect on the system's measurable properties at any time, and the effect is rather like scrolling steadily along an infinite tape of random bits.

by abeppu 5y ago

derefr 5y ago

> Why wouldn't one expect the chaotic system to produce as much or more random output? Because neurons as systems don't like random input. Noisy random input doesn't make (most) neurons fire. Rather the opposite, actually. Neurons aren't (usually) simple threshold comparators of a single input; they're arbitrarily-complex gate-like recognizers, that fire in response to detecting signals in their combined inputs, or even upon matching temporal firing patterns within the signals of their combined inputs. In that sense, if you tune every component in the system to produce more noise (chaos), you decrease the signal-to-noise ratio, and so the components that react to "important recognizable changes in the signal" within that noise have a harder time recognizing the thing they're looking for, and so they recognize it less, and so switch less. --- By trite analogy: suppose you're a security guard. Your job is to recognize+report "real crimes in progress", while not reporting "false alarms." (Pretend that the police charge for false dispatches in your town, so your boss wants to optimize shrinkage vs. call budget.) Also, your boss has decided to "keep you on your toes" by introducing "noise" into your computerized security system. Sometimes it'll show a burglary in progress when there isn't one. Sometimes it'll show that everything's fine, when actually the boss has secretly sent in a white-hat burglar of his own. How willing to believe the security cameras should you be, to optimize your output signal to downstream demand? As it turns out, you should probably try to figure out the false-positive vs. false-negative rate you're seeing reported by the downstream, and then calibrate your confidence up/down until you're making a balanced number of Type 1 vs. Type 2 errors. That level of confidence is the "critical point" — where the noise in the system is equally likely to help your detection (by "correcting" a false negative/positive, turning it into a true positive/negative) as it is to harm. Moving away from that critical point, in either direction, makes your output less useful as a downstream signal.

scythe 5y ago

>I guess if the measure of "information richness" is based in compressibility of electrical activity ... is it surprising that it isn't higher in the chaotic phase? Why wouldn't one expect the chaotic system to produce as much or more random output? Chaos can be highly regular. For example, the limiting case of the logistic map, x -> 4x - 4x^2, is chaotic, but also has an exact solution x[n] = sin^2(arcsin(sqrt(x[0])) * 2^n). It's not hard to see why this still obeys the usual properties of Lyapunov doubling etc, but a simple double exponential doesn't seem like the sort of thing that leads to consciousness! From an information-theoretic point of view, the nth bit of x[0] becomes relevant at roughly step n, which means there is a steady "trickle" of information into the system instead of the more complicated dynamics we expect of a "computational" system -- in fact, if we consider the binary expansion of (arcsin(sqrt(x[0]))/pi) we find that only a finite number of bits have an effect on the system's measurable properties at any time, and the effect is rather like scrolling steadily along an infinite tape of random bits.