Here's a bit from another section in Koch ("The Neurobiology of Consciousness", citations omitted), a passage that deals both with the distinction and relationship between consciousness and attention and with the question of what consciousness is, in functional terms. (Emphasis added)
Many actions in response to sensory inputs are rapid, transient, stereotyped, and unconscious. They can be thought of as cortical reflexes and are sometimes called zombie behaviors. A slower, all-purpose conscious mode deals with broader, less stereotyped, and more complex aspects of the sensory input (or a reflection of these, as in imagery) and takes time to decide on appropriate responses. A consciousness mode is needed because otherwise a vast number of different zombie modes would be required to react to unusual events. The conscious system may interfere somewhat with the concurrent zombie systems: focusing consciousness onto the smooth execution of a complex, multi-component, and highly trained sensorimotor task -- dribbling a soccer ball, to give one example -- can interfere with its smooth execution, something well known to athletes and their trainers. Having both a zombie mode that responds in a well-rehearsed and stereotyped manner and a slower system that allows time for planning more complex behavior is a great evolutionary discovery. This latter aspect, planning, may be one of the principal functions of consciousness.
It seems possible that visual zombie modes in the cortex mainly use the dorsal stream in the parietal region. However, parietal activity can affect consciousness by producing attentional effects on the ventral stream, at least under some circumstances. The basis of this inference is clinical case studies and fMRI experiments in normal subjects. The conscious mode for vision depends largely on the ventral "what" stream.
Seemingly complex visual processing (such as detecting animals in natural, cluttered images) can be accomplished by cortex within 130-150 ms, too fast for consciousness to occur. It is plausible that such behaviors are mediated by a purely feedforward moving wave of spiking activity that passes from the retina through V1, into V4, IT, and prefrontal cortex, until it affects motor neurons in the spinal cord that control the finger press (as in a typical laboratory experiment). The hypothesis that the basic processing of information is feedforward is supported most directly by the short times required for a selective response to appear in IT cells. Indeed, Hung and colleagues were able to decode from the spiking activity 100 ms after image onset from a couple of hundred neurons in monkey IT the identity of a single image flashed onto the retina of the fixating animal. Coupled with a suitable motor output, such a feedforward network implements a zombie behavior -- rapidly and efficiently subserving a binary categorization task in the absence of any conscious experience.
Conscious perception is believed to require more sustained, reverberatory neural activty, most likely by way of cortico-cortical feedback from other neocorical regions. These feedback loops would explain why in backward masking a second stimulus, flashed 80-100 ms after onset of a first image, can still interfere (mask) with the percept of the first image. The reverberatory activity builds up over time until it exceeds a critical threshold. At this point, the sustained neural activity rapidly propagates to parietal, prefrontal, and anterior cingulate cortical regions, thalamus, claustrum, and related structures that support short-term memory, multimodality integration, planning, speech, and other processes intimately related to consciousness. Competition prevents more than one or a very small number of percepts to be simultaneously and actively represented. This is the hypothesis at the heart of the global workspace model of consciousness.
Just in case there are any lurkers who don't have the time or inclination to learn textbook jargon, this is basically saying that it's possible for people to do things that we normally would think would have to involve conscious awareness -- like clicking a button when you recognize an animal hiding in a forest in a picture -- much more quickly than would be possible if consciousness were actually involved.
This is because there are "feedforward" (i.e. one-way) pathways through our brains that allow us to respond to recognition of certain kinds of things. Basically, from eyes straight through the brain and out to our muscles.
You'll notice that the study that was mentioned involved recognizing animals in landscapes. We're pretty good at that, having evolved to do it. There's a reason they didn't choose, say, picking out 3-digit numbers that are multiples of seven from among random arrays of 2-4 digit numbers.
So our "zombie" modules in the brain are limited. They deal with things we're evolved to know how to do, and things we're very well trained how to do, such as dribble a soccer ball.
So if that's feedforward mode, what about consciousness?
Koch describes a feedback system within the neocortex (our most recently evolved brain tissue) which reaches a threshold before "rapidly [propagating] to parietal, prefrontal, and anterior cingulate cortical regions, thalamus, claustrum, and related structures that support short-term memory, multimodality integration, planning, speech, and other processes intimately related to consciousness".
That would explain why a "masking" image flashed
after exposure to a previous image can prevent you from ever being aware of the first image -- the first image didn't have time to reach the feedback threshold that would make it perceptible to consciousness, even though it's perceptible to certain zombie modules.
So here we have a framework for understanding how one attentional system can serve those brain structures and processes responsible for generating conscious experience and at the same time serve those brain structures and processes that do other things.
Gazzaniga has some very interesting things to say about Jeff Hawkins' view of those feedback systems, but I'm not convinced that those comments belong on this thread. Perhaps I can pare them down....