# Models & control

**URL:** <http://discourse.iapct.org/t/models-control/7145>\
**Category:** CSG1995\
**Created:** [October 20, 1995, 3:54pm UTC](http://discourse.iapct.org/t/models-control/7145 "1995-10-20T15:54:04Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![\_Avery\_Andrews](http://discourse.iapct.org/letter_avatar_proxy/v4/letter/_/848f3c/32.png) [@\_Avery\_Andrews](http://discourse.iapct.org/u/_Avery_Andrews)\
**Post date:** [October 20, 1995, 3:54pm UTC](http://discourse.iapct.org/t/models-control/7145/1 "1995-10-20T15:54:04Z")

</div>

[Avery Andrews 951020.1055 Eastern Oz Time]

Here are two instances from everyday life of what might be `model-based'  
control.

1. Walking with a group. You want to be near the other person/people,  
but can't spend all your time looking at them, so you rely, I believe,  
on assumptions such that if you keep walking in the same direction  
without much change in speed, they will tag along. Of course, every  
now and then this fails, you turn around & they're not there, and you  
experience that the peculiar disoriented feeling of a model falling apart.

2. Conversation. If I understand Martin Taylor more or less correctly,  
the speaker has a perceptual reference (my current attempted term  
for a reference level for a high-level perception) for the hearer to  
know something, so that when this reference is satisfied, the talk stops  
(to the extent that the purpose of the discourse is communication of  
factual information, of course there are lots of other perceptual  
references involved in discourse). But the speaker's perception of  
what the hearer knows will surely be a constructed model of some kind.  
So that if I perceive the hearer to know X, and say a sentence whose  
meaning M, together with X, implies Y, by reasoning which I believe  
the hearer to be smart enough to follow, then I will perceive the  
hearer to know Y, but if the hearer's subsequent behavior seems  
inconsistent with knowing Y, this perception will disappear.

So what's a model? I'd say that it is a perceptual circuit that  
takes a lot of inputs and produces outputs, some of which, at least some  
of the time, can be matched against the outputs of other perceptual  
circuits. And, when there's supposed to be a match, but there isn't, all  
of the outputs of the model are written off as wrong. E.g. when we  
turn around and don't see the person who we have thought was following  
us, it's the output of the model that we write off, not what our eyes  
tell us; that's what makes it a model.

Avery.Andrews@anu.edu.au

---

<div class="post-metadata">

**Author:** ![\_Martin\_Taylor](http://discourse.iapct.org/letter_avatar_proxy/v4/letter/_/a9adbd/32.png) [@\_Martin\_Taylor](http://discourse.iapct.org/u/_Martin_Taylor)\
**Post date:** [October 20, 1995, 5:26pm UTC](http://discourse.iapct.org/t/models-control/7145/2 "1995-10-20T17:26:14Z")

</div>

[Martin Taylor 951020 13:10]

> Avery Andrews 951020.1055 Eastern Oz Time

> 2. Conversation. If I understand Martin Taylor more or less correctly,  
> the speaker has a perceptual reference (my current attempted term  
> for a reference level for a high-level perception) for the hearer to  
> know something, so that when this reference is satisfied, the talk stops  
> (to the extent that the purpose of the discourse is communication of  
> factual information, of course there are lots of other perceptual  
> references involved in discourse).

In principle, you are right, but more generally in LP theory, the speaker  
has a reference to perceive some state in the listener, and that may  
involve action because the speaker currently perceives the listener to  
e in a state different from the speaker's reference state for the listener.

The controlled perception may be entirely or not at all related to what the  
listener knows. It may be simply to have the listener move in such a way  
as to block sunlight from falling on a page, accomplished by saying something  
like "You can hear a funny buzz if you stand over there and listen carefully  
for a while."

> But the speaker's perception of  
> what the hearer knows will surely be a constructed model of some kind.

Yes, that's a central aspect of LP theory.

> So that if I perceive the hearer to know X, and say a sentence whose  
> meaning M, together with X, implies Y, by reasoning which I believe  
> the hearer to be smart enough to follow, then I will perceive the  
> hearer to know Y, but if the hearer's subsequent behavior seems  
> inconsistent with knowing Y, this perception will disappear.

That's it, although I doubt we use that kind of logical approach much  
in practice. But we do, I think, model what the effect of our words  
might be, before committing to them. We control our imagined perception  
of the other before acting in a way that the other perceives. Normally,  
that's a very fast process, but sometimes one can observe it happening  
within onesself. When we have acted, we perceive the other's internal  
states in the same way we perceive any complex, by using our perceptual  
input functions on the obervable actions of the other. If the other's  
actions do not generate within us the perception that the other knows Y,  
then, as with Hans Blom's models, the external data force a recalibration  
of the model, and of our current perception.

I think that is ill-phrased, but my model of the audience suggests to  
me that some will understand what I mean, and some will not.

Martin