Slashdot Mirror
Human Brain seems to procceses image data serially
Tekmage writes "Ever wonder how the brain processes image/vision data? According to this research, it does so in a more serial manner than parallel. " This has been one of those on-going debates since the 1960s, with the advent of machine vision, with this being the latest round in the battle between the two sides.
Agreed. If they simply had an array of black blocks with one white block and said, "Find the white block," or, say, "Does this array of blocks contain two green ones," you'd be able to do that *instantly*, because, as I understand it, you would be able to find the patterns in the scene and discover the anomalies (the white block, for example) in much more of a parallel fashion, which allows you to do it nearly instantaneously, whereas examining a green and a red block for a small "nick" (which would require a detailed examination, thus movement of the eyeball itself), requires much more detail.
This is simply common sense. Wave your hand in front of a lamp (monitors don't work very well for this experiment). How do you see it? As one image after another; you don't see your hand in all possible positions at once.
Note that that's simply processing sensory data. The people who talk about spotting the red cube in a bunch of green ones are talking about something totally different: recognition. Even there, the brain picks the red cube out from the whole image; the reason the time to recognize the red cube doesn't depend on the number of total cubes is that you see the same size image, no matter how many cubes there are, and the red cube looks different enough from the green ones that it's easy to spot. It'd be like playing "Where's Waldo" in a situation where everyone else is wearing blue; no matter how many people there are you'll find Waldo in a second.
Still not convinced? Here's a simple experiment to try: play "Twinkle Twinkle Little Star" and Limp Bizkit's "Nookie" in your head simultaneously (those songs being chosen because they're totally different; feel free to substitute any other two songs that are sufficiently different), and try to concentrate on both at the same time (note: do this without actually saying the lyrics to either one; that's cheating). You can get pretty close, but I'll bet that you can't quite do it.
More than likely, the brain simply "multitasks" in a manner not unlike machines do today; it doesn't really run multiple processes at once but it can do a pretty convincing illusion. Since each area works somewhat independently of the others, you can get a bit of parallelism going. That's why you could sing the lyrics to one song while thinking of another; you've assigned a different area to each task. Put them in the same area (by not saying the lyrics to either one) and suddenly you can't do it.
So, cheer up. At least on this planet we're still top dog in terms of intelligence (your average U.S. politician notwithstanding).
What bothers me about the article is that it takes the stance of "The debate has always been which architecture is best. Now, since the human brain processes data serially, the debate is settled."
Since when was it established that the human way of processing things was the best for sure??? Very poor logic on their part (IMHO)
Werd.
To me, if the brain is really a large parallel machine, there's no reason why seperate threads of computation can't be going on in seperate parts of the brain - each taking up a small physical region of resources.
I agree, but I wouldn't consider these "threads" to be cognitive in nature. A person's immediate attention is always focused only on one item at a time. Try examining one object while describing another. Your mind has to switch back and forth to be able to do both "simultaneously."
In response to the "so this makes serial processers better" line of thought, it should be pointed out that it takes many parallel human braiin processors to get an accurate image descripion.
Think about trying to describe a thief you saw running down the street. You saw that he was tall and wearing a hat, someone else saw that he had a mustache, etc. Add more processors required to compensate for the uncertainty in the data from any single one, and you've got a system that doesn't look so serial anymore.
It's obvious that more than one human is required for an accurate description. They haven't proven anything in the serial vs. parallel debate!
-- First post (by a female living in a state that begins with M and does not end in a vowel with a birthday that falls
People have known for years that some visual processes occur in parallel, because they take constant time regardless of the amount of input. For example, if I ask you to pick one red square out of a scattering of many green squares, the time required does not depend on the number of squares. Other tasks require times proportional to the number of objects. For example, finding one red square in a scattering of red circles, green circles, and green squares, is a task requiring time proportional to the number of items you have to sort through. Everybody assumes that this is a serial process. All this has been known for years-- the description of the tasks that can be done in parallel, and hence the properties of the hardware that computes them, was pretty much settled in the late '80s.
No doubt the research reported in this article is important for some reason, because I saw the technical paper it was based on in the most recent issue of Nature, which is a pretty major journal. Unfortunately I don't have it with me, so I can't read the paper and tell you why it is important. Certainly it's not just the fact that some kinds of visual perception are serial.
I start breakfast, and then take a shower while the water boils or whatnot. While taking a shower, I often think of what I'm going to code after breakfast. I would consider that to be "multi-tasking".
Now, here's another thing - how many times do you wake up in the morning with an answer to a complex coding problem? For me - it's *alot*. I find the answers just float in from dimension X into my head. That's parallel processing - part of my brain solved the problem while the other part handled something completely different without either part being aware of what the other was doing.
I think the debate is rather moot - we can do both. If you want to argue over the sematics, you can do so. But when I think of the brain, I think of it as a complex signals processor.
What I mean is, when you see something, it's translated into a signal, which is run through a series of filters and comparisons to tell you what you're seeing. This is also why you don't have an exact copy of what you saw - your brain only stores the "most significant bits" necessary to duplicate the signal. Some brains are better than others about reconstructing the signal. If you don't have all of the signal, your brain fudges it with values from similar experiences (or your values/beliefs). And if you have no signal at all, you post as an Anonymous Coward.
So my point is - it can be both. Infact, look at how society is structured - into clusters of people (brains?) that work in parallel on a project until completion (teamwork). Minimal communication. Why wouldn't your own brain be wired in a similar fashion - with dozens, if not hundreds, of semi-autonomous agents working towards the same goal?
--
As far as point 3, here's the relevant portion of the article:
The remaining items aren't delved into in the least, but it would certainly be nice if they were true.
It seems that by placing the blocks on opposite sides of the board (left and right), looking at the left block would elicit a higher amount of activity in the right side of the brain while examining the right block would fire up the left side. I believe these differences were what they were looking for. If the subject were able to examine both blocks in parallel, the two halves of the brain would work simultaneously. The experiment showed a 1/10th second or so difference that was always right -> left, indicating that they focused their attention on the left block followed by the right.
The article didn't really explain this, though, so this is just my educated guess.
We humans have developed organizational intelligence. Groups of human brains, hooked up with the appropriate networking, can themselves become an alien intelligence, as different from human intelligence as human behavior is from cellular behavior.
For a long time, this has been mostly the province of corporations and governments. Ever wonder why such entities often lack common sense? It's because they are made up of humans, but aren't human. Congress is a group of over 400 humans; it doesn't act as a human, but can be modeled as an intelligent, alien being.
Today, we have the Internet. On a smaller scale, we have Slashdot-style phenomena. These are virtually those "Beowulf clusters of human brains". It is just another alien intelligence.
The big difference between the Internet and government/corporate organizations is in the interhuman connectivity. In governments and corporations, the governing layers are codified into a bureaucracy. This causes specific people to act as chokepoints, and that in turn limits the number of people that can interact effectively. On the Internet, the governing layers are a lot less codified. This requires a lot more data filtering at the various nodes (humans)--spam and similar phenomena travel better across the Internet than through your office--and a lot more bandwidth. But the Internet is all about bandwidth.
Bureaucracies are alien intelligences made of humans. Internet communities are alien intelligences made of humans. They are different species of alien, and they are fighting each other.
Why are bureaucracies afraid of internet communities, and vice versa? The answer is easy to see if you stop thinking in terms of humans. The bureaucracies are seeing a brand new type of intelligence. The "Linux community" is a perfect example. Over the course of eight years, this thing has gotten Microsoft, one of the Lords of Bureaucracy, frightened. A race war of organizational intelligences is brewing, if not already being fought.
Is this the end of humanity and the beginning of organizational intelligence? Hardly. We have been living with bureaucracies since the Pharoahs, possibly before. But just the knowledge that there are inhuman intelligences out there helps you to better understand them, and to better interact with them.
--The basis of all love is respect
Right -- but that's just pattern recognition (something that is done in parallel), and not a cognitive analysis. You eventually just "know" that that pattern of 6 dots is, well, 6 dots.
I suspect that the article is talking about very high level cognitive processes. But it's clear that heck of a lot of parallel preprocessing has to happen upfront! Before you can shift your attention from one object to another, you have to recognize that object regardless of how it is oriented, what the lighting conditions are, what the background is and so on. They aren't saying that this is all serial.
But to me, those higher processes have less to do with vision and more to do with reasoning. I might experience one thought after another concerning some object, but I still see all the objects in front of me.
The raw image data is of course handled in an extremely parallel fashion, but the cognitive process involved, identifying patterns and discriminating between one object and another, is serial.
This really shouldn't be that big of a surprise. Try watching two or more moving objects simultaneously, and pay attention to how you do it. Your attention ends up being focused on one item at a time, albeit relatively quickly (depending on how fast you think and how much caffeine you've had).
Though I basically agree with their findings, I'm not too thrilled about how this experiment was set up. They basically *forced* the participants to think serially by placing both of the suspect blocks on opposite ends of the board (yes, I know that's really the only way they could reliably determine which item was being focused on and when). The eye ball itself isn't capable of doing a detailed analysis of imagery except in the very small area in the direct center of its field of view. It's only logical for the participant to immediately identify the different colors peripherally (and perhaps even in parallel -- the experiment never delved into this part) and then concentrate a detailed glance first on one block, then on the other. Biologically, it had to happen that way. Their eyes couldn't have efficiently made the same analysis in a parallel fashion.
"We are the first research group to show definitively that the human brain processes images serially-paying attention to only one object at a time and shifting rapidly from object to object"
now my question is, might this not have to do with the human's eyes and focusing on one object at a time and switching between multiple images quickly to try to bring them into focus as simultaneously and seamlessly as possible?
"It was important that we knew the order in which they paid attention to the colored objects, because the N2PC works by correlating the brain waves coming from each side of the brain over many statistical trials, so we had to always have them search in the same order"
He acknowledges that the brain is paying attention to certain objects based on color in a certain order, but attributes this to the brain and not to the input device. I'm going to make a crude analogy which will probably get shot down, but if you can think of a better one, please post it. It's like taking a mono VCR hooked up to mono speakers VS a mono VCR hooked up to a surround sound speakers. You know it's able to process the info better, but it can't because of the input device's shortcomings.
It seems totally intuitive. The only news here is that they've got documented data to back up intuition.
... And sage, and rosemary... "
:)
We can only focus on one thing at one time, therefore we can only handle one visual input. I'd venture the guess that all our I/O is serial - with quite a bit of DMA capability thrown in.
We can tune in on a single conversation in a room full of people, and switch focus from one to another, but it's real hard to keep track of more than that. We remember music sequencially, but unless we're well trained in music, we can not correctly conceptualize chord structures.
We become completely oblivious to the goings-on when we watch (and listen to) TV. We have a difficult time separating olfactory inputs - so we process those serially as well. "What is that? Lemon?
The only sense that seems parallel to me is the tactile. Though, since tactile input is the summation of very many single (bit) neurons, the parallelism we experience is probably the result of a lot of preprocessing of stimuli in the sensory nervous system and the spinal chord.
The neat thing is when we tune all the senses into the same stream of data. Remember last Christmas? The scent of the cooking goose, the sound of the Grandma Got Run Over By A Raindeer, the blinking of those damned lights and the itchy wool sweater..
With all of the senses delivering a variety of data that shares the same conceptual context, the imprint of the event is more powerful than if the serial stimuli from the different senses were reporting on events that we know are not related. This is probably why we remember better those times when all our senses are firing in parallel on the same concepts.
I'd venture the guess that as this research progresses, we will learn that we manage some pseudo-parallelism in our input processing through a similar mechanizm that we rely on for memory. Chunking, was it?
For example, if shown a group of objects, we can visually process them based on similarity (i.e. they're all read, square, whatever) so we notice more than if they were all distinctly different. Then we get lost in the volume of data that we have to take it.
As with the chunking that takes place when trying to remember more that the 7 (avg) simple items, finding commonality among the items we try to process sensually, makes it possible for us to more more data through our inputs. Sort of a lossy compression really.
-- What you do today will cost you a day of your life.
Redmond, Wash. Wed. September 8 1999.
Microsoft today announced Windows for Neurones, the brand new Microsoft operating system for life critical operations. No release date has been set yet, but Microsoft hope to have a release version on the shelves by the fall of 2000.
It is thought that Microsoft have been working on this product for several years, early alpha versions of which can still apparently be seen in institutions around the US. "We had problems with the initial cooperative multitasking that we tried. Processes would sometimes end up in a loop, and not release the processor for other tasks." an insider said. "The results of these early tests can be seen as high up as ex-president Ronald Reagan. He was an early alpha tester, but developed problems. Unfortunately, the uninstall wasn't available then."
Microsoft site several advantages to using the OS:
1. Your brain is no longer dependant on old proprietry systems, some of them as old as several million years! We've learnt a lot in all those years. Windows for Neurones (sometimes referred to as Windows Neurones Technology, or just WinNT,) uses such modern features as pre-emtive multitasking, and virtual memory.
2. Your brain can now use cheap, off the shelve productivity software. Studies have shown that a lot of people have to have productivity tools (calanders, addressbooks etc.) as external programs or peripherals. WinNT has all this built in. It is also easy to use, "it's as if it knows what you are thinking," an insider said.
Some people have expressed concerns over the scalability of the new WinNT. While older systems (such as AT&T Metabolism Control and HP Coordination) have exploited the natural parallism in the typical brain, WinNT's new visual system appears to process data in a serial fashion, limiting the ability to exploit the brain's parallel capabilities.
"Rubbish," said a MS insider, "It has been shown in independent studies that our approach is upto 300% faster in processing visual data, for example." he said, quoting a recent study by Mindcraft Inc., a service-oriented, independent test lab. The visual aspects of the OS, what the person sees, has been controversial in recent discussions.
Existing OS providers in this critical industry also slam WinNT's reliability, based on test observations. "We have systems with a mean time before critical failure of 100+ years. I don't understand why anyone would want to upgrade. While brains running our OS consume ~20% of the bodies metabolic rate, we estimate WinNT brains to use upto 30% of the bodies metabolic rate, as it has no power saving facilities. Existing systems have the ability to sleep, saving power, but I've heard WinNT can keep you up all night. This can cause real problems." said a rival brain OS provider. Even if people think the visual aspects are better, which is debatable, a nice visual interface is a waste of time if your heart stops beating! Some things are simply more important than good visuals.
Microsoft refused to release licensing details, but it is said not be following the recent trend of open source software, and open API's and protocols. It is said to include a new licensing agent, called 'Paranoia', which prevents third parties from getting too close and examining it's workings, or 'reverse engineering' as it is known.
I seem to remember from long ago that the switching time for neurons is the same as that for mechanical relays .. on the order of milliseconds. Such low switching time makes it impossible for vision to be operating in anything other than a massively parallel manner.
Given this, the article will have to do better than just state `vision is serial' w/o specifying how that is possible when using slow neurons.
Joe
A straight quote from the article:
"Luck was able to use N2PC to identify whether a person was processing visual signals one at a time or simultaneously. "
Bummer of a name for a probability doctor, eh?
+&x
I agree. Simply because the human has arisen to become the dominant form of life on the planet, what does this have to bear on the overall scheme of intellegince?
Let's see, out of how many billions of species over a few billion years have we come to dominate so totally (unless the Ants have nukes we don't know about)? My guess it has something to do with our brains, and how they work. Hands are pretty cool (read my thoughts) but I have to assume (unless you live in Kansas) that your brain helped them along to their current level of dexterity at some point (perhaps your parents chose white collar jobs?).
If machines can and do someday become intellligent, and do indeed surpass human intelligence,
it'll be 'cause we want them too. I'll leave it at that.
Read my sig, and you'll see where I fall on the debate.
+&x