MaCro Philosophy

Revisiting Introspection: The Reverse Playing Card Experiment and Dennettian Predictive Processing

I previously discussed the playing card experiment and how it can go against our intuitions about the expected clarity of peripheral vision. I concluded that, while it shows our intuitions are wrong about introspection in this setting, we are not wrong about our everyday experience. This is because everyday experience also includes top-down predictions and the capability to interact (to check) filling in the details. I think I was wrong on both counts. Hopefully here I can rectify that.

The reverse playing card experiment

In the original playing card experiment you start with an unknown card at arms length to the side of you, and, while fixating on a point directly ahead, slowly bring the card around in a circle in front of you. You need to bring it surprsingly close to directly in front of you before the details on it become clear.

I previously used this video to argue that top-down predictions (that there is a ball) can fill in the gaps left by the peripheral vision. In the video the player is directly over the penalty spot, his peripheral inputs are of a blurry white circle, which combined with the prior knowledge that a ball should be there, is enough to convince him to kick. I previously claimed:

"It makes more sense to me to say that his conscious experience is of a ball (what he believes he is seeing due to high-level predictions), than of an indeterminate white blob (the information he has when only considering his low-level visual system). ... In the card experiment, there is no way to make a good high-level prediction so we (perhaps for the first time) notice the inaccuracy of peripheral vision by itself."
This claim leads to an easily testable hypothesis:
Hypothesis 1: The lack of clarity observed in the playing card experiment will disappear with top-down information about the card.
It is also easily disproved. Even after studying the card beforehand, the card experiment still works.
Reverse playing card experiment: Draw a card from a normal deck and study it until you know every aspect of its appearance. Hold the card at arm's length directly in front of you. Without moving your eyes, slowly rotate the card away from the center of your visual field. How far from the center must you bring it before the image of the card loses clarity?
For me, the result was even more surprising than the original. I expected the card to remain clear much longer than it did, but it quickly faded, changed, became distorted, and lost or even changed colour. It does seem that you can get further in this direction, but there is a clear drop-off in clarity very early on.

So much for top-down information filling in all the details. You can even glance over at it and then back away again and the low detailed version reappears. Perhaps its not that then, but the possibility to interact that is most important. I previously also claimed:

It's the strong constraint of no eye movement ... that causes the startling result.
If our experience of the world is based on the possibilities for interaction, then of course strange things happen when we purposefully prevent those interactions. Again this is testable (although less easily than in the previous case):
Hypothesis 2:The results in both playing card experiments will disappear when artificial constraints on interaction with the card are removed.
I already mentioned that glancing at the card and back doesn't improve clarity, but this isn't quite enough as it requires returning to the constraint of looking forwards. Fortunately, another loose end from a previous post can help here.

DES Results

In Direct Experience Sampling a subject is asked to report their experiences at random times of the day. I used my random alerter for a week and attempted to recall and write down my experience every time it alerted me. My first few attempts at writing down the contents of experience were the previously reported vague descriptions of what I was doing, alongside a surprising feeling of clarity of some very specific current focus of my attention. I expected that over time, as I grew more familiar with the task, my descriptions would improve. That is not what happened. Instead, I became less and less confident at my ability to accurately recall any details about the contents of my experience. I quickly stopped writing down anything at all, as I became less and less sure of the accuracy of anything I could write. Originally I had the intention to start writing again as soon as I improved. Later I realised that there was just no point.

In this experiment there was no constraints placed on the allowed actions (at the time of the inspected experience) yet there was a similar surprising lack of clarity in the reportable contents. It is well known that the contents of consciousness are not easily translated into words. But this is more than that, I couldn't even be sure that I was recalling them correctly pre-translation. At first it felt frustrating, but later it became more interesting. Could I really have been wrong about my own experience, something I've lived with and made my goal to understand, for so many years?

Revisiting the Question

Relating this back to the video again, we have been considering two seemingly competing possibilities:

  1. At the time of kicking the imaginary ball the player's experience contains a high-definition experiential image of the ball.
  2. At the time of kicking the imaginary ball the player's experience contains (if anything) a low-definition experiential image version of a white circle (that is believed to be the ball).
In both cases the player must believe that there is a ball otherwise we cannot explain his actions. The high-definition answer is tempting because it feels like the experiential image coincides nicely with the belief. It also coincides better with our pre playing card experiment intuitions. The low-definition answer coincides better with the post playing card results.

What actually is the difference between the two answers? Both of them explain the player's behaviour. Both are unverifiable. It is hard to imagine a conversation with the player that would be illuminating.

Silly memes and imaginary conversations aside, I think the question actually is just wrong. Perhaps it doesn't make any sense to ask about aspects of conscious experience considered by themselves (out of the context of the rest of experience and removed from objectively verifiable aspects), because they don't really exist. Furthermore, current understandings of how the brain function seem to allow us to talk about the examples above without making this mistake.

Rapid Real-time Construction and Comparison

Close your eyes and attempt to reconstruct, in as much detail as possible, an internal model of the world around you. This type of construction is what the brain is doing all the time, only, in real-time, with almost no room for error. It's true that the brain gets feedback from the senses to help, but this is in the form of a massive, constantly changing set of electro-chemical signals. The feedback is not in the form of people, objects, colours, or even directions. All of these have to be constructed. The only way that it can be called feedback at all (and not just inputs) is if we consider the brain as constantly trying to predict its inputs, that way it can use any mistakes as feedback to improve its model. This process is rapid, real-time construction (specifically not re-construction). The construction goes all the way up from the inputs to high-level concepts, and, then, in order to get some feedback, is compared all the way back down.

This up and down process is happening rapidly and in real-time. There is no time to wait for signals to go up before they come back down. At all points there is upwards flow meeting downward flow. It is only in considering this as a whole process that we can start to talk about experience.

Dennettian Predictive Processing

The brain does what it can to keep up with the constantly changing inputs, cutting corners when necessary. For example, it has learnt that objects stay the same as they enter the periphery, but not having the need, time, or resources to actually check, it just assumes this is the case. Objects in the periphery have some feedback through the senses, so they form part of what we call our visual image. However, it does not make sense to consider the visual aspect of them on its own, it only makes sense to consider it as a relation to the whole. For objects in the periphery, depending on how much focus is placed on different parts of the model, it makes sense to say they are either blurry or clear in different situations. Neither is wrong (or right).

The Right Question

The right question to ask (or the thing to translate questions about experience to so that they actually make sense) is: What are the qualities of the combination of predictions and feedback that forms your combined model of the world? When, like in the playing card experiment, we focus our attention, we can increase the relative importance of specific aspects of the model. If we attend to our visual field then the feedback through the eyes gains more importance, so we report something more closely related to the capabilities of the senses. In everyday life however, we are not attending specifically to the visual field, so the higher-level conceptions are relatively more important.

Obviously a lot more needs to be said, and there are some especially tricky self-referential questions about the capacity to attend to aspects of the model. But this is at least a better account than I gave before. It explains why our pre-experiment expectations may be wrong, and the results of all the experiments.