Hacker News Reader

Comment by burnte | Hacker News Reader

I did read the summary and see what you saw, but that still leads me to believe the headline of the article is clickbait and that the authors don't understand that action signalling doesn't require the same bandwidth as information processing, but even then it's way more than 10b/s.

Look at a backhoe. It has a few levers and a couple pedals. EXTREMELY simple interface. Each lever and pedal is basically 1 bit for engage/disengage, but the operator has to process orders of magnitude more sensory info to operate it properly. You could use an arduino to control a backhoe, but you'd need quite a powerful computer to know what to tell the arduino to do. This shouldn't surprise anyone. Knowing how to use the tool well is always far more complicated than simply knowing how the tool operates.

View on HN

Here's a simple reaction test: https://humanbenchmark.com/tests/reactiontime

It's 1 bit (discriminate red vs. green) but I doubt many here can manage 100 ms, which is what it would take to implement 10 of those decisions per second.

If you extend your arm straight out to point for example, that's significantly more than ten commands sent at once to your arm muscles. It's just non-conscious. That test measures a bunch of things all together too. Visual processing, color recognition, and conscious decision to act. I got just under 240, and considering everything I had to do, that's pretty neat.

This measures latency not processing power. You can have 10 b/s or 1 Gb/s of processing power with the same latency.

yes, but the greater the latency, the greater the number of bits per period required to get the same throughput.

I come in around 250 ms on the latency, which means that to get up to 10 bps I'd need to actually be able to discriminate between 6 choices each time.

To get up to 100 bps?

Can you point me to the book that says that "Booleans" are encoded in the human brain just like in a binary computer and they take exactly 1 bit to store? I mean, why not assume error correction, while we are at it, or some kind of compression algorithm that reduces the size further?

If that sounds ridiculous to you, you are beginning to get it. Every single brain cell works on electric potential (a continous value). Not a single one of them can be said to be in the "zero" or "one" state (a discrete value).

"bits of information" doesn't mean 1's and 0's. "bits" is a measure of entropy.

If you send an analog signal over a wire and the receiver on the other end can distinguish between 1024 different voltages reliably, then you sent 10 bits of information. Even though you sent neither a 0 nor a 1, but an analog voltage.

It's about the "information" as an abstract concept, not about how that information is encoded as data sent over some medium. I can send you the same thing thousands of times. I would have sent you lots of data, but 1000 copies of the same thing still only contains as much information as a single copy does.

But on the other hand, a "fast" debator from a high school debate team can process 260 words per second while compressing that down to notes, while simultaneously queuing up meaningful responses, while evaluating which/what to go with for a particular audience/judge, while listening for mistakes from their opponent. If you distill bitrate down to serialized responses to a canned psychometric task, sure, but why do we think that's the total throughput? It isn't, trivially, if we inspect what people do every day.

260 words per SECOND? Are you sure about that? I can only assume you mean for skim/speed reading.

that's a funny typo, I apologize.

No need to apologize! We're all human!

Can the combination of mind and body only process reactive tasks serially, or can it process them with some amount of parallelism?

And is 100ms (I struggled to get below 180ms, myself) only the time needed to process, or does that include IO latency?

If it includes IO latency, then is there a buffer?

According to psychology research on human perception there is indeed something that you could call buffered IO in the human brain.

It's used in theories to explain why you can hear your name in a short sentence in a crowded room and then know what the sentence is. Including one or two words before your name. While if your name wasn't said, no attention is diverted to that sound and the "buffer" contents get dropped without retrieving the information.

Yes, and there's different buffers with different capacity for different types of perception according to these theories. Look up Baddeley & Hitch for details. I would guess there are similar buffers for other senses that didn't make it in the lectures in college.

This might be the worst possible way to measure the net throughput of conscious decision making I've ever seen. A wildly ecologically invalid, serialized discrimination task.

Now take someone navigating a social situation with three simultaneous participants, actively listening, coordinating body language, interjecting, responding to questions... and you have a human being operating at a "bitrate" of information that is so many orders of magnitude removed from this bullshit psychometry task as to prompt the question: "what the actual are they talking about".

Using a computer analogy to try to understand the brain functions is ok as long as we remember that the brain is not actually one of them. There have been different technological analogies throughout history including hydraulic systems, telegraphic and telephonic lines, all of them might seem comical today but they were the best explanation at the time, same will be in the future with ours, comparing something like the brain (or rather the whole living being) to a mathematical processing device is more than short sighted. We try to explain everything from the engineering point of view by isolating things but that’s not how life or the universe works.

> or can it process them with some amount of parallelism

I guess someone with two boxes handy could set them up next to each other and run two copies of this test to see if their reaction time holds up or if it lengthens?

EDIT: mine suffers greatly on dual wield: 225 -> 320 ms

I'll have a good think about whether we can process things in parallel or only in series the next time I decide to walk and chew gum while bouncing a ball, not tripping over things, and staying out of traffic.

none of those activities need involve conscious decisions, however

Having a think is not a conscious activity?

that, yes, sorry.

I meant walking, chewing gum, dribbling a basketball, and avoiding traffic.

Mh, balance that ball on your finger while keeping it in spin or dribble it between your legs while spinning your body around and see how many conscious decisions you make while not stepping into canine feces.

Depends on expertise. Consider a toddler learning to walk, while eating a snack, carrying a toy and avoiding the cat.

Driving stick is a good example, as shifting gears requires conscious thought at first and becomes mostly automated with experience.

There is a difference between latency and throughput.

True, maybe you can find a test which asks, instead of a single button click, to click one of 4 possibilities for a reaction?

EDIT: see appendix A.3 Perception

You mean like typing? Typing we click buttons accurately in sequence, people can type way faster than 10 bits per second.

Then consider something more complex like drawing, you draw way way way more than 10 bits a second.

not like typing, which is largely automatic; conscious decisions per second.

EDIT: not the expanded bit rate, the optimally compressed bit rate, if that makes sense?

EDIT2: typing at 90 wpm is 1,5 wps, or <10 bps according to Shannon (~1 bit per letter, 5 letters per word)

> EDIT2: typing at 90 wpm is 1,5 wps, or <10 bps according to Shannon (~1 bit per letter, 5 letters per word)

People can type way faster than 90 WPM, we talk about bitrate of the hardware here its the same even if you haven't practiced typing. And typing is still not what our consciousness is made to do, we have way higher bitrate than that when doing more native tasks such as running in the woods. You can't run in the woods without consciously thinking about each step and deciding where to put your foot next to not hurt your ankle and not collide with trees, that has a massive bitrate.

Running in the woods supports TFA: 180 steps per minute * 16 possible locations per step = 12 bits per second; maybe there are fewer or more possible footfall spots but no matter what it's more than 1 bit per second and less than 100. (and way less than 1e9)

> not like typing, which is largely automatic; conscious decisions per second.

That’s kind of a cop out - at what point do very quick conscious decisions become “automatic.”

Any FPS is this but with about a million possibilities (depends on the screen resolution). People can become crazy good at it.

Before you say it does not require decisions - you can have friendly fire on.

IO is the problem not processing power.

Reply on news.ycombinator.com