In an earlier essay ("The History of Thinking"), I presented the thesis that brain function was initially a holistic pattern-recognizing process, but with mammals it made a leap to sequential processing. This may surprise those who think of "linear thinking" as intrinsically inferior to "holistic thinking". Indeed, the fact that holistic, pattern-recognizing thinking preceeded sequential thinking clearly demonstrates its superior utility. But the fact remains that sequential thinking allows us to tackle problems that cannot be handled by pattern-recognizing thinking.
Before I get rolling with my primary thesis in this essay, I shall present an example of what I mean by the two types of thinking. My exemplars are none other than Dick Duck and Penelope Pig, two denizens of Habitat Crawfordat, and the behaviors that I shall consider are part of their daily routines.
Every evening, Dick Duck and his girlfriend Jane must retire from the ducky comforts of their pond to the safety of their racoon-proof duck cage. Of course, being ducks, they do not think so far ahead, but I do that thinking for them. My task is to entice them into their duck cage. The trick lies in establishing and maintaining a simple pattern that their dim ducky brains can readily recognize. The first element of this pattern is for me to stand near their duck cage in the late afternoon sun. They respond to this visual pattern by exiting the pond and approaching the duck cage. Now, however, I must maneuver them into the cage itself, a problem because I myself constitute a negative pattern to them (ducks don't like humans to get too close). So I cannot pull them into the cage, I must push them in with the force of my animal repulsion. In doing so, I have learned that ducks maneuver by a simple algorithm in response to my movements. So long as I don't move too quickly or threateningly, their algorithm is to move directly away from me, and also to avoid obstacles or corners that could become traps. I therefore take a semi-circular path from the opening of the cage, around the ducks, to a position that places them between me and the cage opening, at which point I begin moving directly towards them. The process is always complicated by the meanderings of Penelope Pig, Binky Burro, and Galahad Goat, each of whom constitutes a duck threat of differing radius. But, having learned the values of these duck threat radii, I can always maneuver the ducks into the cage by assuming that Dick will follow the Duck Standard Threat Avoidance Algorithm. What's especially significant about this algorithm is that, with the exception of panic considerations, it is utterly devoid of historical considerations. If Dick happens into a specific corner, he will always exit that corner in the same direction, even if he has already done the same thing several times in preceeding 60 seconds. He doesn't seem to remember anything that I did just a moment ago. He has no sense of history or sequence. That makes him easy to herd.
Penelope Pig is a tougher customer. She has memory and uses it. If she makes a break for freedom, and I have to herd her back home, she can be maddeningly adroit at anticipating my feints. She too follows a Threat Avoidance Algorithm, but that's only a guideline. She pays careful heed to my motions and stays well out of traps and corners. She can anticipate a trap several steps ahead, and is too smart to fall into any trap that has worked on her once. Two years ago, I had to medicate her eyes, something she considers as tantamount to sooey-cide. I therefore trapped her in the duck cage to do my dirty work. To this day, Penelope avoids the duck cage whenever I am anywhere nearby. She's got sequential memory.
But what is the mechanism that permits Penelope's sequential memory? How can she recognize sequences of events where Dick cannot? To answer this question, I'll jump over to digital electronics. I recall the good old days of building digital circuits out of TTL chips, and there were some valuable lessons from those days. Stable pattern recognition was fairly simple to do. For example, suppose that I have eight wires coming in from some input device and I need to recognize a particular pattern on those eight wires. For example, suppose that the pattern 11001100 on those eight wires represents the pattern I want to trigger on. It is a simple matter to build a pattern recognizer for this pattern using TTL chips. I simply hook each of the wires that should be zero into an inverter, and then feed their output, along with the outputs of the wires that should be ones, into an eight-wide AND-gate. If I don't have an eight-wide AND gate, I can assemble one out of a bunch of two-wide AND-gates. When I'm done, I have a little circuit that will light up when the desired pattern comes down the wire. Note also that this little circuit requires just four inverters and one eight-wide AND gate. That's not many chips.
(An aside: who cares about recognizing patterns? What has any of this to do with thinking? The answer is that just about all thinking could be characterized as pattern recognition. I'm not saying that pattern recognition is exactly the same thing as thinking -- nobody really knows what thinking is. Instead, I'm saying that a great many processes that we call "thinking" can also be described as pattern recognition. When I yank my foot back from the precipice, it is because my visual processing system recognized the visual pattern of distant valley floor as an intrinsically dangerous pattern. There can be millions of different images of "valley floor way, way down there", but they all add up to the same thing: don't fall! Recognizing that pattern, reducing the image to the warning, is an act of thinking.)
Now suppose that I want to do exactly the same thing (recognize the pattern 11001100), except that this pattern will not come down eight separate wires; instead, it will come down a single wire as a sequence of bits. You would not believe how much more difficult this makes the problem! At any given instant, I have only a single bit coming down the wire, so there's no static circuit that can recognize the pattern. Instead, I have to set up a much more complex circuit. There are four major innovations that must be added to my simple TTL circuit to enable me to recognize that sequential pattern:
1. Memory. Clearly, once a new bit has come in, the old bit is gone -- but it had better not be forgotten. We'll need some kind of memory circuit to hold those old, now-dead bits. Otherwise there won't be a pattern to recognize. Now, a memory circuit requires a handful of gates -- let's say, four gates for every bit of memory. So we're going to need 32 gates just to remember the pattern as it comes in. Remember, our previous circuit used only five gates in total.
2. Clocking. This is even trickier. If the bits are going to come in one by one, then there's a timing problem: how long do we wait for one bit to come in? When do we decide that it's time for another bit? It's easy if we get a "one", and then a "zero", and then a "one", and so forth. But what if a "one" is followed by another "one"? Is it one long "one", or two "one"s in sequence? How can we be sure of the difference? The answer lies in something called "clocking". We have to add a clock to our circuit that tells us how long to wait for a single bit. That clock itself adds lots of gates to the design; let's say 20 gates for the clock. Then we have to send the clock signal -- a kind of digital "tick, tock" -- out to the gates that receive the signal. All this gets even more complicated -- and adds more gates to the circuit. Let's say 10 gates for this task.
3. Routing. Suppose that we are about to receive the 5th bit in the sequence of 8 bits. If we're going to remember each bit, then we have to store the 5th bit into the 5th slot. That means that we have to route the first bit to the first slot, the second bit to the second slot, and so forth. Guess what? This adds even more complexity to our circuit. I'd guess it at maybe ten gates.
4. Termination. Lastly, we have to know when to stop receiving data and start recognizing it. After the 8th bit straggles in, we have to stop the process of clocking bits into memory and start the process of recognizing them. This requires us to add a counter to our circuit, something that tells us how many bits have come in. When the counter reaches 8, we know that it's time to stop counting. This in itself is a pattern recognition problem, requiring four gates.
Once we have clocked all the bits into the memory locations and decided that it's time to stop loading bits and start recognizing patterns, the job is identical to the simple task: we have eight bits lined up in eight wires, and we must recognize a pattern. But look how much trouble we had to go to to get to this point. One way to estimate the magnitude of the task is to count gates. Remember, the simple pattern recognition task requires exactly five gates (four inverters and one big AND-gate). The sequentialized version of the same task requires at least 70 gates.
Now you can see why sequential thinking came so late in evolution: it's much more expensive than direct pattern recognition. Which leads to the question: why bother? The answer, of course, is demonstrated by Penelope Pig and Dick Duck. Dick's a sucker for herding, but Penelope is nobody's fool. In terms of survival, it doesn't take much in the way of brains to outwit Dick and catch him. Penelope is harder to catch.
But there's another factor I'd like to digress on: the association between pattern recognition and visual processing on the one hand, versus sequential thinking and auditory processing on the other hand.
Visual processing is almost entirely static pattern recognition. You see an image, and your brain converts that image into a recognizable pattern. Put a snake into the visual field of a primate, and the primate will react with fear. Ducklings can recognize a raptor's characteristic shape immediately after hatching. All sorts of other visual tasks are simple matters of recognizing particular shapes, colors, textures, or combinations thereof.
There is, of course, some visual processing that takes temporal or sequential factors into account. For example, we react instantly to the image of a rapidly approaching object. Yet much of this processing, it turns out, takes place not in the brain but right in the retina, and it tends to be highly specialized in nature, sort of "wired in" to the rods themselves. I think it highly significant that the only sequential visual processing we have had to be custom-built right into the retina, rather than handled inside the brain.
But consider that auditory perception is inherantly sequential in nature. That is, a visual field is a big bunch of pixels all in parallel; the task of figuring out what's in a visual field is almost exclusively one of large parallel pattern recognition. But the auditory signal is one-dimensional: sound is a sequential stream of information, not a broad field of parallel information. Thus, processing sound is intrinsically a sequential process.
This suggests an interesting hypothesis: that creatures with smaller brains should rely less on auditory processing and more on visual processing. In particular, I would speculate that reptiles would rely primarily on visual and olfactory sensing, using auditory sensing only for simple processes. This in turn would suggest that reptiles (and also amphibians, insects, and fish) would utilize only single-sound calls. I don't know if this is the case, but what little I do know about calling behavior suggests that there is indeed some truth to my speculation. My turtle Toby warns me to back off with a ferocious hiss. Rattlesnakes make their single rattling sound.
But birds engage in more extensive calling behaviors. They sing multi-note songs. Interestingly enough, my duck Dick doesn't have a multi-note repetoire: he just says "quack". Maybe that's another reason why he's so easy to catch: weak sequential processing.
OK, so here I've set up this implicit hierarchy of sequential processing, with those dumb reptiles, fish, insects, and amphibians at the bottom with their stupid single-sound processing, then birds a little further up the scale with their multi-note songs. Obviously this puts us humans way up in the stratosphere with our complicated multi-note spoken languages. Isn't it nice when we can prove that we stand at the pinnacle of creation? Let's all pause for a moment to admire ourselves.
Back to work. Let's consider the significance of multi-syllable auditory processing. (I'm going to use the term "syllable" here to denote any "auditory unit of expression". A single note of birdsong would then fit my loose use of the term.) Obviously, more syllables are more difficult to process than fewer syllables. We can therefore use the "typical syllable length" of a creature's call as a rough indicator of the complexity of its sequential-processing neural circuitry. Thus, Dick the Duck comes out looking pretty dumb with his single-syllable quack. The cuckoo, if I am not mistaken, looks better with his two-syllable "cuckoo". And the best and brightest of the lot -- if my hypothesis be true -- would be the songbirds with songs up to a dozen notes long.
By the way, I'm not imputing general intelligence to multi-syllabic processing. This would make songbirds smarter than ducks. Now, ducks are pretty dumb, but I have long since given up ascribing any meaning to the term "intelligence". Songbirds can process more notes than ducks can -- that means that they have greater abilities to process notes. It doesn't mean that a songbird would beat a duck in a chess match.
So now let's apply this reasoning to ourselves. We too have "songs" composed of sequences of syllables, but we call them "sentences". Our sentences, of course, are more complex than birdsong because we have more consonants and vowels than they have. (Birdsong, I believe, is basically a bunch of vowels with no consonants at all. The same thing seems to be true of whalesong.) So our songs have a much more complex set of basic components. For example, how many notes would you say go into the sound of "ch"? How many into "sh"? Those are complex waveforms, carrying far more information than a single note.
But what really interests me is the length of our typical songs. Have you ever noticed that sentences seem to have a maximum upper length? I'm not talking about those long-winded written sentences that march across the page like some grand parade and are almost always the result of the author trying to show off to the reader just how smart he is, even though the whole thing could be said in a much shorter sentence; this kind of thing is particularly evident in the writings of the Victorians, although the Germans the world title for death-march sentences with their particularly cruel trick of putting the verb at the end of the sentence, hold.
Anyway, real, nonpretentious sentences -- the kind that we speak rather than the kind that we write down -- tend to be less than fifteen syllables long. Again, I'm not referring to the hoity-toity sentences that we spray all over each other when we're at work, but rather the workaday sentences we use on our family when we get home. If you push your listener, you might get 30 syllables into a sentence and still be understandable, but once you get up to 70 syllables, it's a real strain to keep track of. For example, try reading the previous sentence aloud to somebody and see if they can follow it. It comes in at 52 syllables -- and it is a mouthful.
This "standard length" suggests that the sequential-to-parallel conversion mechanism in the brain suffers from some physical limitations. Several explanations for this limitation are possible: 1) the termination process is based on an internal timer. This would also explain why a slowly spoken sentence is so exasperating; 2) the buffer that stores the incoming sentence isn't very big; 3) the wavefront processing system that takes over once the sentence is inside the brain can't handle too wide a wavefront.
Now at long last I can start to bring my point home: the existence of such tight constraints on sequential processing strongly suggest that the human brain is not a general-purpose sequential processor (like a Turing machine). Instead, it's a parallel processor that fakes sequential processing by converting it into parallel form before serious processing begins.
The implications of all this are staggering. There is no Spock-ish war between logic (sequential processing) and emotion (parallel processing) inside our souls. Instead, logic is a layer on top of emotion, a front end for emotion. Our exploitation of logic has yielded civilization, technology, and Nintendo -- but differentiating logic from emotion inside our heads is as deceptively hopeless a task as removing a tree. You can cut it off at the trunk but you'll never get all the roots out.
Believe it or not, all this intellectualizing really does bear on the subject of play. I'm not sure how I'll get there, but I aim to demonstrate that play is a necessary function in the operation of layered sequential/parallel brains. Bear with me -- I'm working on it.