[Martin Taylor 2008.03.23.10.09]
From Bill Powers (2008.03.23.0130 MDT)]
Martin Taylor 2008.03.22.23.03 --
The blurred handwriting is a case not covered by my "variability" hypothesis..... For conflict to arise, reading the message has to be important to the person looking at the blur. It wasn't important to me so I didn't even try.
Did you wait for the writing to become crisp, or did you just look at the starting blur? Or did the GIF animation not make it through your e-mail processing? My original was animated when I got it from CSGnet, but it wasn't when quoted in your message.
There is a stage in the animation when it is almost as easy to read as a properly handwritten text would be, and you wouldn't really "have to try", any more than you do when you glance at ordinary handwriting. Only when it becomes really crisp does it become clear it isn't handwritten text at all, because there are hardly any letters (just first letters of words).
That description of mechanism sounds quite reasonable. Would you therefore accept Bekesy audiometry as an experiment that demonstrates the control of uncertainty?
It does seems to me that control of uncertainty implies a reference level for uncertainty. In the Bekesy audiometry experiment, the reference level is suggested by the experimenter, who asks the participant to make the tone just detectable with 100% certainty, and just undetectable with 100% certainty of no tone.
Actually, not 100%. I don't know what levels individual subjects might choose for "do hear" and "don't hear", but I can assure you that for me, at least, I would judge the limits to be nearer 20% and 80%. That's not the point, though. The point of the Bekesy audiometry is that it is a tracking study, tracking the trace of maxiumum uncertainty as the tone changes frequency. There is nothing that it tracks except uncertainty, a simple scalar, like a compensatory track of a disturbed cursor.
The experimenter is asking the participant to be certain that the tone is there or not there, and inferring a region between those limits where there is uncertainty.
I suppose in a normal tracking study the experimenter is asking the subject to be certain that the cursor is to the left or right of the target and inferring a region between the limits of the tracking variation within which the subject is uncertain (assuming the action apparatus is sufficiently precise). The question arises there, too. Suppose the visible cursor is severely blurred in a standard tracking task, and instead of asking the subject to track 1 cm to the left of the target, you ask them to track so that they would bet 3:1 that the cursor is to the left, rather than the usual 50-50. Would that seem like a nonsensical task for most subjects? I doubt it.
I suppose the same could be done by adjusting the focus in the blurred handwriting example, so the participant would be asked to make the message just barely unreadable, and just barely readable. Then one could infer a region of intermediate focus where there is uncertainty. The "objective" uncertainty would be shown by correct and incorrect indentifications -- but that would reduce to my "variability" hypothesis.
The 'objective" uncertainty isn't really of much interest, in the same way as "real reality" isn't of much interest when we control an ordinary perception. The interesting question isn't whether someone is convinced (very certain) they will win the next lottery (objectively rather uncertain), but that there is even a perception of uncertainty associated with the perception of anything at all. It's the _association_ of a feeling of uncertainty with any specific perception that led me to my original question.
I think Rick is probably right in suggesting that there is no subjective uncertainty in the perception itself, whether it is pure or filled in from imagination. You don't say you hear a tone if you don't experience a tone.
Have you ever been a subject in a psychoacoustic experiment studying near-threshold effects? I don't think you would make that comment if you had. Most of the time, you simply don't feel sure whether you experienced a tone or not. That's quite different from correctly or incorrectly judging that a tone was really presented.
You may intermittently experience a tone, so you can perceive variations in its detectability, but when you're hearing it or not hearing it, you don't report the opposite. You say "Sometimes I hear it and sometimes I don't."
That's what you do if those are the choices you are allowed to make. Some studies don't limit the choices to "Yes" and "No", as I mentioned in an earlier message. They allow you to say, using a slider or a number from 1 to 10, how sure you are that you heard a tone.
The experimenter analyzing those data assumes that the subject is trying to tell the truth about the experience, based on the fact that the experimenter is usually a subject and does try to tell the truth, with results similar to the results from othe subjects. Then the experiment analyzes the data as though the subject would have said "yes" if the slider was above each possible position, knowing whether there actually was a tone on each occasion. The results uniformly show a smooth curve, which would be unlikely if the subject's self-report of their certainty was some kind of illusion.
Isn't all this the very thing that physicists argued about? We can say that on a small enough scale, measurements of position and momentum fluctuate unpredictably, so that given one measurement it is not possible to measure the other reliably. Each measurement, when made, gives a number that seems 100% certain, but the next measurement is different, also with 100% certainty. That shows us that we should not feel certain about any measurement even if we get a definite number from it, because there are variable influences at work that we can't predict or take into account, except statistically.
This all relates to the mapping of perception onto the real world, an exercise that often seems to be useful. Since we don't believe (there's that word from the subject line -- almost) that we can ever be assured of real reality, the more interesting question is how the perception is produced.
My question is simpler: what are the connections that create a perception of uncertainty and that allow that perception to maintain association with the perception that is felt to be uncertain. I suggested, for no reason other than that it had proved useful in the development of artificial neural nets of a structure similar to the perceptual side of the HPCT structure, that perhaps perceptions are better represented by complex numbers than by real numbers. I made no commitment to that, nor did I guess how it might be implemented biologically. What I am reasonably convinced of is that there must be some fairly tight coupling beteen the value of a perception and the perception of its uncertainty.
Martin