Control of Behavior

[From Rick Marken (2017.04.03.1750)]

How about a PCT analysis of this:

https://nyti.ms/2nPvprZ

BestÂ

Rick

···

Richard S. MarkenÂ

"Perfection is achieved not when you have nothing more to add, but when you
have nothing left to take away.�
                --Antoine de Saint-Exupery

[Martin Taylor 2017.04.05.11.57]

[From Rick Marken (2017.04.03.1750)]

How about a PCT analysis of this:

https://nyti.ms/2nPvprZ

Or this

Title: :

the first line of which is: “.” Later in the first paragraph, there is this: “”
The article on Uber would require a pretty complex analysis, I
think. So many perceptions to be guessed at controlled by so many
people. So much network dynamics to be analyzed. I don’t know which
would be the more difficult area to start with, but both would have
to be resolved in a full analysis. I wonder whether there are useful
simplifications that would make sense?
Martin

···

http://www.nature.com/articles/srep45722

  •  Control strategy of hand movement depends on target
    

redundancy*

  •  When we move our hand to some
    

specific target, we plan how to reach it before performing a hand
movement** when
we grab a vertical pole, we have to control the hand movements in
the plane orthogonal to the pole, but we need not control the
movements along the pole. Such a point-to-bar reaching par- adigm
(manifold reaching paradigm4,5) contains a problem in which the
CNS has to deal with the target redundancy, i.e., where should we
move our hand? By comparing point-to-point reaching with
point-to-bar reaching, we can gain an insight into how the CNS
deals with a target redundancy in movement planning.*

[From Rick Marken (2017.04.07.1130)]

···

Martin Taylor (2017.04.05.11.57)

RM: How about a PCT analysis of this:

https://nyti.ms/2nPvprZ

MT: Or this

[http://www.nature.com/articles/srep45722](http://www.nature.com/articles/srep45722)

RM: Sure, but I was interested in getting a PCT analysis of the Times article “How Uber Uses Psychological Tricks to Push Its Drivers’ Buttons” because it was specifically about how Uber and Lyft are using “behavioral science” to control the behavior (the title of this thread) of its drivers.

Â

MT: The article on Uber would require a pretty complex analysis, I

think. So many perceptions to be guessed at controlled by so many
people.

RM: I think it is possible to provide a pretty straightforward PCT analysis of the behavior control programs described in the article. I will wait for a couple days and see if anyone else wants to take a crack at it before I do.Â

Best

Rick

So much network dynamics to be analyzed. I don't know which

would be the more difficult area to start with, but both would have
to be resolved in a full analysis. I wonder whether there are useful
simplifications that would make sense?


Richard S. MarkenÂ

"Perfection is achieved not when you have nothing more to add, but when you
have nothing left to take away.�
                --Antoine de Saint-Exupery

[Martin Taylor 2017.04.07.16.40]

[From Rick Marken (2017.04.07.1130)]

I didn't think the point was to do what is, as you say, pretty

trivial. My thought in offering the link was to see whether it can
be done in a way that Scientific Reports would publish. It might
also be a nice exercise for the newer-to-PCT readers of CSGnet.

Martin
···

Martin Taylor (2017.04.05.11.57)

RM: How about a PCT analysis of this:

https://nyti.ms/2nPvprZ

MT: Or this

            [http://www.nature.com/articles/srep45722](http://www.nature.com/articles/srep45722)
          RM: Sure, but I was interested in getting a PCT

analysis of the Times article “How Uber Uses Psychological Tricks to Push Its Drivers’ Buttons” because it was specifically about how Uber and Lyft are using “behavioral science” to control the behavior (the title of this thread) of its drivers.

            MT: The article on Uber would

require a pretty complex analysis, I think. So many
perceptions to be guessed at controlled by so many
people.

          RM: I think it is possible to provide a pretty

straightforward PCT analysis of the behavior control
programs described in the article. I will wait for a
couple days and see if anyone else wants to take a crack
at it before I do.

[From Rick Marken (2017.04.09.1420)]

···

[Martin Taylor 2017.04.07.16.40]

MT: I didn't think the point was to do what is, as you say, pretty

trivial.Â

RM: It’s uncanny how you are able to consistently find me saying things that I don’t think I said. What I said was that a PCT analysis of the programs described in the article (Â https://nyti.ms/2nPvprZÂ ) would be straightforward, not trivial. Here’s a passage from the article that, I think, lends itself to a straightforward PCT interpretation:

In 2013, the company hired a consulting firm to figure out
how to encourage more driving during the platform’s busiest hours.
Â
At the time, Lyft drivers could voluntarily sign up in
advance for shifts. The consultants devised an experiment in which the company
showed one group of inexperienced drivers how much more they would make by
moving from a slow period like Tuesday morning to a busy time like Friday night
— about $15 more per hour.
Â
For another grroup, Lyft reversed the calculation, displaying
how much drivers were losing by sticking with Tuesdays.
Â
The latter had a more significant effect on increasing the
hours drivers scheduled during busy periods.

RM: The consultants were devising a way to control the times when the drivers would sign up for their shifts. They wanted the drivers to sign up for the busiest hours. They tested doing this by disturbing possible controlled variables that could only be kept under control by signing up for shifts during the busier periods. The two possible controlled variables were 1) how much they would make and 2) how much they would lose by signing up for shifts during the busier periods. The greater increase in sign ups for busy period shifts with a disturbance to possible controlled variable 2 suggests that most drivers are controlling for not losing money or the drivers who do control for not losing money control with higher gain than those who control for making money.Â

RM: A better definition of the hypothetical controlled variables might be “the concept of making money” and “the concept of losing money” since the drivers are not shown a real time display of how much money they are making (or losing) relative to how much they could be making (or losing). Also, these results are based on aggregate data so we can’t know what each driver was actually controlling for. But it is clear that aggregate behavior can be controlled fairly well by applying disturbances to hypothetical controlled variables – well enough to make a difference for the company’s bottom line.Â

RM: This apparently non-coercive approach to controlling aggregate behavior  (non-coercive in the sense that drivers are not forced to sign up during busy hours) does have its potential problems. The main one I see is that it can lead to possible conflicts withing the driver. This is because the outputs (signing up during busy periods) that are used to correct the disturbance to “not losing money” may conflict with control of other things that the driver is controlling for, such as being home with a loved one during what turn out to be the busy hours that the company wants covered.Â

MT: My thought in offering the link was to see whether it can be done in a way that Scientific Reports would publish. It might also be a nice exercise for the newer-to-PCT readers of CSGnet.

RM: OK, I downloaded that article. But since you suggested it, I’d be interested in seeing your PCT interpretation of that article before I try to give mine.Â

Best

Rick

Â


Richard S. MarkenÂ

"Perfection is achieved not when you have nothing more to add, but when you
have nothing left to take away.�
                --Antoine de Saint-Exupery

          RM: I think it is possible to provide a pretty

straightforward PCT analysis of the behavior control
programs described in the article. I will wait for a
couple days and see if anyone else wants to take a crack
at it before I do.Â

[Martin Taylor 2017.04.09.17.29]

[From Rick Marken (2017.04.09.1420)]

Yes, and its equally uncanny how you consistently have me saying

things I don’t think I said. The only thing I said about the Uber
article was that I thought a full analysis would be quite complex.
“Complex” is a poor synonym for “trivial”. “Straightforward” is a
pretty good one.

Then  I suggested that the PCT analysis of the article <[http://www.nature.com/articles/srep45722](http://www.nature.com/articles/srep45722)    >

might be a good exercise some of the readers less familiar with PCT
modelling, and in a second message (the one to which you were
responding), I said that this was why I posted the link in the first
place.

If you can do a thorough PCT analysis of the Uber article, go for

it. I still think the interactions among the many different
perceptions controlled by the many different people are pretty
complex, and the scheduling of driverless vehicles wouldn’t be
trivial in itself, even without taking the different perceptions
controlled by the different drivers into account.

Martin
···

[Martin Taylor 2017.04.07.16.40]

            MT: I didn't think the point was to do what is,

as you say, pretty trivial.

          RM: It's uncanny how you are able to consistently find

me saying things that I don’t think I said. What I said
was that a PCT analysis of the programs described in the
article ( https://nyti.ms/2nPvprZ ) would be
straightforward, not trivial.

                        RM: I think it is possible to provide a

pretty straightforward PCT analysis of the
behavior control programs described in the
article. I will wait for a couple days and
see if anyone else wants to take a crack at
it before I do.

[From Rick Marken (2017.04.09.1715)]

···

Martin Taylor (2017.04.09.17.29)–

MT: If you can do a thorough PCT analysis of the Uber article, go for

it.

RM: I did that in my previous post. Feel free to critique it and/or give you own analysis of the article you suggested (http://www.nature.com/articles/srep45722).Â

BestÂ

Rick

Â

I still think the interactions among the many different

perceptions controlled by the many different people are pretty
complex, and the scheduling of driverless vehicles wouldn’t be
trivial in itself, even without taking the different perceptions
controlled by the different drivers into account.

Martin


Richard S. MarkenÂ

"Perfection is achieved not when you have nothing more to add, but when you
have nothing left to take away.�
                --Antoine de Saint-Exupery

[Martin Taylor 2017.04.11.17.42]

[From Rick Marken (2017.04.09.1715)]

I read a description of a PCT view of a small part of the Uber

problem. I have no argument with what you actually wrote, but it
wasn’t anything like “a thorough PCT analysis of the Uber article”.
I thought you were intending to do an analysis of the complexities,
such as the network of cross influences among the drivers and
between drivers and the central algorithms and (as your squib
mentions) the Uber owners, and how they relate to scheduling not
just shifts, but pickups – just to mention one of the issues I
thought would be rather tricky.

Martin
···

Martin Taylor (2017.04.09.17.29)–

            MT: If you can do a thorough PCT analysis of the Uber

article, go for it.

          RM: I did that in my previous post. Feel free to

critique it and/or give you own analysis of the article
you suggested (http://www.nature.com/articles/srep45722).

[From Rick Marken (2017.04.11.1915)]

···

Martin Taylor (2017.04.11.17.42)–

MT: I read a description of a PCT view of a small part of the Uber

problem. I have no argument with what you actually wrote, but it
wasn’t anything like “a thorough PCT analysis of the Uber article”.

RM: It was as thorough as I could make it, given the fairly limited information about the research that was available in the Times article. I only presented it as a pretty good example of “control by manipulation” as was demonstrated using the rubber band demo in the “Experimental Methods” chapter of B:CP (p. 245 of the 2nd Edition). The Times article describes control of the drivers’ behavior – signing up for shifts – by disturbing a controlled variable – loss of income–in a manner that is exactly analogous to control of S’s behavior – Â finger position-- by disturbing a controlled variable – the position of the knot. The behavioral researchers described in the Times article wouldn’t describe it that way but they were clearly looking for a variable that most drivers were controlling for by signing up for shifts.Â

Â

MT: I thought you were intending to do an analysis of the complexities,

such as the network of cross influences among the drivers and
between drivers and the central algorithms and (as your squib
mentions) the Uber owners, and how they relate to scheduling not
just shifts, but pickups – just to mention one of the issues I
thought would be rather tricky.

RM: There wasn’t really that much in the article on which to base an analysis of these complexities. The only reason I brought up the article was that it described methods for controlling the behavior of the drivers that were exactly analogous to the methods for controlling behavior described in B:CP. I thoughtthe Times article was interesting because it showed how the simple demonstrations that are used to illustrate control phenomena – and how those phenomena work – can be used to analyze real world behavior.Â

            MT: If you can do a thorough PCT analysis of the Uber

article, go for it.

          RM: I did that in my previous post. Feel free to

critique it and/or give you own analysis of the article
you suggested (http://www.nature.com/articles/srep45722).Â

RM: I thought the article was also interesting because the “behavioral scientists” mentioned in the article clearly had the right intuition about “how driver’s work”; that is, they knew that the drivers were signing up for shifts to control for some perception and they did something like the TCV to figure out what that perception was (for most drivers, anyway). What the researchers didn’t realize was that the drivers control for many variables and that their “non-coercive” control by manipulation technique can lead to conflict, as was demonstrated in the demo on p. 245 of B:CP, 2nd edition.Â

BestÂ

Rick


Richard S. MarkenÂ

"Perfection is achieved not when you have nothing more to add, but when you
have nothing left to take away.�
                --Antoine de Saint-Exupery

[From Rick Marken (2002.09.25.1630)]

Bill Powers (2002.09.25.1336 MDT)--

Something here still nags at me -- perhaps you can fix it. When I speak of
a causal connection, I mean a reliable way to make something happen. A
thermometer reading is pretty reliably caused by the temperature of the
medium it's put in. Turning a car's steering wheel to the right pretty
reliably caused the car to turn right. In general, "x causes y" means y =
f(x), at least as I understand the term.

Yes. That's what I mean, too: x causes y means y = f(x). But in your examples,
particularly the steering wheel example, y = f(x1) + f(x2) + ... The direction
the car goes (y) depends on the direction in which you turn the steering wheel
(x1) and on other factors (x2...), too. So if you're comfortable with the idea
that the direction of turn of the steering wheel causes the direction of the car
then you should be comfortable with the idea that disturbances cause behavior.

I can also see bringing in multiple causation: y = f(x1, x2, ...), as in
your .qo = d1 - d2. But in that case, I could not predict y unless I knew
beforehand, or could set, the values of x1, x2, and so on. If y = f(x1,
x2) and I can control only x1, then I have no way of knowing what value of
y will result because I don't know what x2 will be. The best I could do
would be to predict the _most likely_ value of y given x1 by computing the
distribution of values of x2 and using the mean value in the prediction. In
any one trial, the prediction would be off by some amount ranging from
small to large.

Sure, that's true of prediction. But this fact about prediction is true for all
systems, control or causal. If the output, qo, of a system is multiply
determined (qo = r - d for control and qo = d1 - d2 for causal) then you can't
predict qo if you know the value of only one of the independent variables.

I'm not used to thinking of a variable being "caused" by another variable
when (a) the supposed causal variable is not the only one that can
contribute to the effect,

But you think of car turning this way even though the steering wheel is not the
only variable that affects the direction of turn.

and (b) the other variables are not predictable.

The direction of the wind is not predictable but it doesn't stop you from
thinking that the direction of the car is caused by the steering wheel and the
wind.

So what do we make of this?

What I make of it is what I said in my previous post:

what is unique about control systems [in terms of causality] is _not_
the absence of a causal connection between independent (d) and
dependent (qo) variable nor is it that secular variations in reference
specifications (r) nullify this connection in some way. What is
unique about control systems is that the causal connection between
independent and dependent variable goes through the environment (via
the feedback function connecting dependent and controlled variable); it
does not go though the system that produces variations in the dependent
variable (as appears to be the case with living organisms).

In other words, the important thing about causality when behavior is looked at
through control theory glasses is the _behavioral illusion_. It's not that
behavior is not caused by external circumstances; behavior (the actions that
affect that state of a controlled variable) is caused by disturbances to
controlled variables. Nor is it that behavior is uncontrollable; behavior can be
controlled despite autonomous variations in the reference for the controlled
variable. What we see when we look through control theory classes is that the
function relating stimulus (disturbance) to response (output) in a control
organization is a property of the environment, not the organism. That's what I
make of it.

Best regards

Rick

···

--
Richard S. Marken, Ph.D.
The RAND Corporation
PO Box 2138
1700 Main Street
Santa Monica, CA 90407-2138
Tel: 310-393-0411 x7971
Fax: 310-451-7018
E-mail: rmarken@rand.org