16 bits per second - the bandwidth of consciousness (2007) | Hacker News Reader

16 bits per second - the bandwidth of consciousness (2007)

fm.schmoller.net

27 pointsColinWright24 comments

That statistic comes from a book[0] that was written in 1998, and is based on little evidence. The science of consciousness has tremendously advanced in the last 14 years, and I'm not sure it is still seen as correct.

There is a preconscious reduction of bandwidth (lot's of pruning). The input is transformed, from lots of points in low dimensional spaces to fewer points in a higher dimensional one, and it is hypothesized that consciousness is isomorphic to the integration process[1] that performs that transformation.

That theory has the big advantage of being neither anthropo- nor neurocentric. It can be related to any physical system (biological or otherwise).

Stephen Hawking has been spotted at consciousness conferences recently.

[0] http://www.amazon.co.uk/s/ref=ntt_athr_dp_sr_1?_encoding=UTF...

[1] http://en.wikipedia.org/wiki/Integrated_Information_Theory

edit: corrected and refined.

I'm pretty sure that this integration process in my brain continues even when I'm unconscious. Or would this definition of unconscious preclude people ever being unconscious except when they're dead? I think that's sufficiently at odds with the everyday definition of consciousness that a new word would be called for.

> I'm pretty sure that this integration process in my brain continues even when I'm unconscious.

Nope. http://www.ncbi.nlm.nih.gov/pubmed/19818903

If not off, at least, it is greatly reduced, and memory encoding is turned off, anyway. You can even wake up for a few minutes and not remember it.

That is, in slow wave sleep. In REM sleep, you can be conscious, and probably are systematically (dreams) even though you don't always remember it.

>> That theory has the big advantage of being neither anthropo- nor neurocentric. It can be related to any physical system (biological or otherwise).

This advantage has a downside: the definition is so broad that any large, integrated system can be described as conscious. Of course, you can argue about what counts as "integration" and whether, say, Los Angeles is really a well-integrated system.

Reference: http://schwitzsplinters.blogspot.fr/2012/06/why-tononi-shoul...

Yes, it would mean that consciousness is pervasive in the whole universe, and, that, for example, your cells are probably conscious. See also http://en.wikipedia.org/wiki/Gaia_hypothesis.

The point is that the nature of the experience is isomorphic to the integration process, which is very different from system to system.

That's why IMO, for example, anthropomorphism in AI is a dead end, and the turing test is irrelevant.

The field may or may not have progressed dramatically. However, ironically your claim of a lack of evidence, is not evidential itself, and I suspect the dismissal is unwarranted. Tor Nørretranders[0] was expanding on the work of German engineer Karl Küpfmüller[1].

Furthermore Kupfmuller,Nørretranders are not alone in this regard:

"We can therefore conclude that the maximal information flow of the process of conscious sensory perception is about 40 bits/second -- many orders of magnitude below that taken in by receptors [nerve endings]."

-- Manfred Zimmermann. "Neurophysiology of Sensory Systems" in Robert F. Schmidt ed., Fundamentals of Sensory Physiology (1986) p.116

"... If we do not act in the way we think we do, the reason is partly to do with the bandwidth of consciousness - its ability to transmit information measured in terms of bits per second. [...] The bandwidth of consciousness is around eighteen bits."

--John N Gray (2002) [3][4]

First prize would be evidence for a refutation of the 'consciousness operates at a low information rate' hypothesis, as opposed to conjecture or speculation.

[0] http://en.wikipedia.org/wiki/Tor_N%C3%B8rretranders

[1] http://en.wikipedia.org/wiki/Karl_K%C3%BCpfm%C3%BCller

[2] http://www.amazon.com/Fundamentals-Sensory-Physiology-Spring...

[3] http://www.amazon.com/Straw-Dogs-Thoughts-Humans-Animals/dp/...

[4] http://books.google.com/books?id=k5PkZXJEQqoC&lpg=PA45&#...

I put the number in question because the people who made these estimations didn't have a proper grasp of what consciousness may be.

That being said, I didn't read the papers in question, and they may have a valid argument.

However, I don't understand how you can encode in 16 per second the evolution of your current experience out of all the possible states your brain may be in and go to.

There is a lot of compression due to expectations, but I still find it hard to believe that it is so low.

Edit: For serial tasks (say, reading, listening to someone speak), the 16-40 bps range makes some sense, but I don't think that visual experience can be reduced that much.

Edit 2: Take, say, music. You can encode the audible note tones on 6-7 bit. You can easily listen 4 notes chords played 4 times per second. We're already in the 100 bps ball park, and we haven't taken timbre and loudness into account.

I can't speak to the truth of that, but the conclusion the author tries to draw from it:

"and hence how designers of e-learning materials need to avoid creating cognitive overload."

is totally ridiculous. His conclusion might be true, but unless he wants to posit the number of these consciousness bits expended on parts of e-learning software it doesn't follow from the premise.

"only a minuscule proportion of the sensory data processed by the unconscious mind (capable of processing approximately 11 million bits per second) is referred to the conscious mind (capable of processing approximately 50 bits per second)."

By that measure, a two-hour movie can be compressed down to 22.5 kilobytes... :-)

Automated sub-sub conscious processors in the brain are mostly responsible for how little information the higher mind has to deal with. If there is a tree in a movie, the visual cortex will say "It's a tree, what next?". If there is an abstract tree symbol, the visual cortex will say "What is this giant solid color block sitting in the middle of my field of view?" and send it up to your conscious mind for processing, resulting in a more taxing experience for the audience. Of course, if we knew how the brain worked, we could maybe somehow send signals directly to whatever ball of neurons handles abstract concepts...

>By that measure, a two-hour movie can be compressed down to 22.5 kilobytes... :-)

I can't speak for other's but for myself that sort of compression ratio seems at least possible.

Most of the conscious mind is dealing with a highly compressed stream of symbols (compressed abstracted versions of the actual audio-visual stream), and how they relate to one another in time and space.

More concretely, imagine yourself watching a movie in the theatre:

"That actor's accent is a bit off" "Why do her teeth look so bright?" "I forgot who directed this, hmm was it that Swedish guy.. the guy who directed 'Heaven', wait... was he Swedish? Aargh I wish I could go on IMDB now and check" ".. ugh that kid in the front row's phone is distracting" "The visual style reminds me of this other movie". "The music is getting ominous, something bad is about to happen." "Oh crap the main character is screwed, how he is going to escape this trap?" "I wonder if my gf is enjoying the movie?" etc

For each viewer, that seems likely to me. For viewers collectively, there is more information; some viewer's girlfriend is going to be someone who studies trees, and another viewer is going to think some character looks kind of like the President. This effect of viewer diversity might justify the video file being more than a megabyte!

16/40 bits per second? Then why is the quality of my vision so precise and I'm still able to hear the sounds of my environment? I've heard 2 bits/sec before as well.

Is this referring to conscious sensations recorded in memory, then maybe I can be warmed up to the idea. Long term memory is definitely spotty, although why can I recall my favorite latest dance track with such precision? The lead, bass, kick, clap and atmospherics themselves must compose more than 16 bits/sec.

What about what is processed by the unconscious, to be remembered later? Like when you didn't notice something at the time, but in the future something sparks your memory to remember it. Does the unconscious (that can become conscious) operate on a separate port with its own bandwidth?

But consciously? I strictly, abhorrently refuse. This music is FLAC, baby.

On the other hand, if it is true, then maybe our brain's compression algorithms are insane. Maybe we process sensations in relation to all our past sensation - I hear only the unique parts of this epic bassline, but the data contains pointers to all the basslines I've heard before, turning my consciousness into a complex amalgam of current and past. Then a two-hour movie in 22.5 kilobytes could make sense.

I believe the idea is that sensory input happens far lower than our conscious mind (I.e. thoughts we perceive in our head). Our reaction to input: what we notice about a picture, what we hear- can be compressed to 16/40bits per second.

I'm curious. How do they quantify the amount of sensory information our brain receives vs. the amount that we consciously perceive?

The input bandwidth can be estimated by the amount of afferent nerve fibers, and the neuronal firing rate.

For an example of the estimation of the conscious bandwidth while listening to music, see the second edit of this comment: http://news.ycombinator.com/item?id=4829147

What is the definition of 'consciousness'? I have read papers where it seems that the definition of 'consciousness' is verbal report, or report that corresponds to almost 100% detectability of a signal.

Consciousness is defined as the ability to experience things subjectively.

There are active paradigms to detect it in non-communicating subjects (playing tennis in one's head while in an FMRI, for example), and research is under way to find passive paradigms.

See for example: Preserved feedforward but impaired top-down processes in the vegetative state. http://www.ncbi.nlm.nih.gov/pubmed/21566197 , or this: http://blogs.nature.com/news/2012/10/a-tool-to-quantify-cons...

so there should be an audio compression algorithm that can reduce 5min tracks to 5k files (and still sound equivalent to uncompressed tracks)? that seems a little optimistic to me.

No. If we accept the article's claims at face value (which is another question) all we can say is that there should be a different audio compression algorithm for each person that could reduce 5 min tracks to 5k files. But the algorithm tailored to your auditory processing system would not work for me, because I might be more or less sensitive to certain types of changes than you.

I wonder how much space a midi of the same song (minus the vocals) usually comes to.

A few tens of KB usually, and can be compressed to just a few KB.

Way more information than is actually remembered : What must you remember of the song to be able sing it in the shower after one hearing? Maybe add the air guitar and other instrumentals after a few more hearings...

Quick : Think of that internet song with the Korean guy doing the horsey dancing. Heard it a few times? What's the third verse like?

We're not talking anything like lossless compression : Kind of the opposite.

what's the amount of information you remember got to do with the amount you are conscious of?

Reply on news.ycombinator.com