Hack #102 : Alter Input With Expectations

This is a hack which never made it into the book, but we thought it worth sharing. At this point, to get the most out of this hack, look at this figure (in a pop-up window) quickly before reading on. It’s not important to try and work out what it is, but have a good look. Seen it? Now, without further ado…

Hack #102: Alter Input With Expectation


The balance between feed-forward and feed-back connections in the brain gives a clue to the balance between raw sensation and expectations in constructing experience.

Feedback is ubiquitous in the brain. The brain is not just massively parallel [Hack #52], it is also massively interconnected- an awesomely complex cybernetic system.

Individual neurons only transmit information one way, but all regions of the central nervous system both send and receive signals. About 45% of connections in the brain are feedback connections. Although we talk about hierarchies in processing (for example in hack#13), this is because of the sophistication of the information that is represented there, not because there is a flow of information to the top without signals being sent back down the hierarchy.

The sensory systems are part of a feedback loop with the motor systems. As we move, we adjust our movements according to the new sensations we receive. But as fundamental as this, there is also feedback within each system. Sensation itself is adjusted by feedback from the cortex.

For example, pathways ascend from the eye to the visual cortex (via the lateral geniculate nucleus [Hack #13]), but they also descent from the visual cortex to the eye (again via the lateral geniculate nucleus). Eyes are not just dumb sensors. From the very beginning our senses are fully integrated into the whole system that is the brain.

The functional corollary of feedback connections is top-down processing on perception. Top-down is internal information – what you know and what you expect. Bottom-up is external – the information that’s actually coming in and which your brain has to make sense of. The existence of so many feedback connections suggests that information coming into our brains is altered by top-down processing from the very beginning. Those wires must exist for a reason. It looks like we are wired up to have what we perceive influenced by what we already know and what we expect to happen next.

In Action

This Figure (pop-up window) shows a classic example of how our perception is affected by what we know [1].

If you haven’t seen this before you might not spot at first that it is a dog in the snow. Once you’ve seen it, you can’t un-see it. It’s not a conscious decision on your part; rather, your perception is now dominated by the top-down information about how this stimulus is organized. If you were just using bottom-up information, the picture would remain a picture-without-any-interpretation-a collection of light and dark splodges. But because you now know that there’s a dog in the picture, you brain imposes that knowledge on our perception and you can’t help but read the picture that way.

How it Works

So we’re always trying to fit what we’re sensing to what we know. If we just used bottom-up information only – trying to deduce what we are seeing without any expectations and assumptions – it’d be too ambiguous and too slow. But if we just used top-down information we’d only see what we already know-there’d be no surprises and we’d get caught out whenever things differed from our expectations.

You can see your top-down processes at work best in situations where the bottom-up processes are weak. With vision this might be in the dark, or where you only glimpse something or someone for a fraction of a second. In hearing this might be where background noise is loud. Poor resolution, brief or noisy information tips the balance in favor of top-down information. What we see comes to reflect more of what we already know and what we expect. Hence we see things in the dark: our brains fill in what is most likely there, what might be there, or what we fear could be there, based on small clues from what actually is perceivable there.

There’s a balance between experiencing the world just as it is without any interpretation and experiencing the world just using expectations. Neither extreme is possible; it always has to be a compromise. Looking at the extent of physical feedback looks in the brain, it turns out the balance between working out the universe from first principles every time (bottom up) and imposing our expectations on what we perceive (top down) is ever so slightly in favor of bottom up. In terms of connections, it’s 55% vs 45%.

See Also

These ideas are expanded upon and explored with far more profundity and depth than here in these two papers by Karl Friston of the Functional Imaging Laboratory at UCL, London (www.fil.ion.ucl.ac.uk).

  • Friston, K. (2002). Functional integration and inference in the brain. Progress in Neurobiology, 68(2), 113-43.
  • Friston, K. (2002). Beyond Phrenology: What can neuroimaging tell us about distributed circuitry? Annual Review of Neuroscience, 25, 221-50.
  • End Note

    1. Gregory, R., L. (1970). The Intelligent Eye. London: Weidenfeld & Nicolson. (This figure is also found on many psychology sites on the Web.)

    2 thoughts on “Hack #102 : Alter Input With Expectations”

    1. Robert Anton Wilson talks a lot about this phenomenon and its relationship to mysticism. The mechanism described above is the answer to the koan “Who is the master who makes the grass green?” Light reflected off a blade of grass enters our eyes, information is transferred to our brains which organize the information and produce for us an image, which is the green-ness of the grass. Our experience of the grass is not the grass alone, nor is it our nervous system alone, but a combination of the two. The word yoga comes from the same root as yolk and essentially means “to bind two things together”. Hence, our experience is the yoga of the objective world and our nervous system.

    2. A friend who read this post wrote to say to me
      Just a quick correction on your latest post about feedback. You state:
      “For example, pathways ascend from the eye to the visual cortex (via the lateral geniculate nucleus [Hack #13]), but they also descent from the visual cortex to the eye (again via the lateral geniculate nucleus)”
      As far as I know, in mammals, there is no pathway from the visual cortex or LGN to the Retina. There is a pathway from the visual cortex to the LGN however.
      This is probably correct – there is no direct feedback from visual cortex to the retina, but there is feedback from the rest of the brain down the optic nerve towards the retina (see http://webvision.med.utah.edu/fbloops.html, especially section 5 which deals with mammals). So maybe no feedback from visual cortex, but from hypothalamus and maybe other regions. Feedback connections seems generally ill-understood and less celebrated than feed-forward connections. One neuroscientist told me at a conference that to invoke feedback connections in your explanation of a phenomenon was “the last refuge of a scoundrel”

    Leave a Reply

    Fill in your details below or click an icon to log in:

    WordPress.com Logo

    You are commenting using your WordPress.com account. Log Out /  Change )

    Facebook photo

    You are commenting using your Facebook account. Log Out /  Change )

    Connecting to %s

    %d bloggers like this: