the road dial and the akrasia horizon

Here, finally, is our write-up on the akrasia horizon and road dial:
blog.beeminder.com/dial

http://dreev.es – search://"Daniel Reeves"

I understand the value of simple, comprehensible user interfaces,
but speaking purely theoretically (and therefore impractically),
the “one week” aspect is obviously not a theoretically-motivated
parameter.

Does anyone have ideas of what the “dial” would look like if we cast
simplicity to the winds,
and try to get a theoretically-motivated number?
For example, we might assume that the user is acting based on
hyperbolic discounting,
prefers to go off the road rather than stay on the road today,
and prefers to stay on the road forever rather than go off the road
forever.

Can we use that, and some algebra and get, for example,
that the horizon ought to be some specific function of the degree of

Johnicholas

I like one week. It seems about right for the designated purposes. And I’m one of the people who complained most loudly that I wanted to be able to change my slope once my metabolism slowed down and I didn’t realize how much it would suck to be hungry enough to stay on my YBR.

If we were to derive something theoretically, it would totally depend on the parameters you chose for your hyperbolic discounting function (How much do you discount tomorrow versus today, over and above how much you discount the day after versus tomorrow? And also, what is the actual length of “today,” for akratic purposes? The right timescale might be a minute or an hour or a week rather than a day.) I see no reason why we know these “fundamental” parameters (or, indeed, that they truly mean anything more fundamental, as opposed to being a convenient theoretical modeling device) than the parameter of a week that’s been chosen here.

David

On Sep 7, 2011, at 1:04 PM, Johnicholas wrote:

I understand the value of simple, comprehensible user interfaces,
but speaking purely theoretically (and therefore impractically),
the “one week” aspect is obviously not a theoretically-motivated
parameter.

Does anyone have ideas of what the “dial” would look like if we cast
simplicity to the winds,
and try to get a theoretically-motivated number?
For example, we might assume that the user is acting based on
hyperbolic discounting,
prefers to go off the road rather than stay on the road today,
and prefers to stay on the road forever rather than go off the road
forever.

Can we use that, and some algebra and get, for example,
that the horizon ought to be some specific function of the degree of

Johnicholas

“Don’t believe everything you read on the Internet.” – Abraham Lincoln

I second that.

One week (or anything between 5 and 10 days) works fine and is well
enough, to my mind.

And I think that the akratic horizon is about more than hyperbolic
discounting. Wait, “think” is not the right term. - I feel it. It’s a
kinesthetic reference. (If that doesn’t make sense to you, never mind.

Alex

Wednesday, September 7, 2011, 10:17:01 PM, you wrote:

I like one week. It seems about right for the designated purposes.
And I’m one of the people who complained most loudly that I wanted
to be able to change my slope once my metabolism slowed down and I
didn’t realize how much it would suck to be hungry enough to stay on my YBR.

If we were to derive something theoretically, it would totally
depend on the parameters you chose for your hyperbolic discounting
function (How much do you discount tomorrow versus today, over and
above how much you discount the day after versus tomorrow? And
also, what is the actual length of “today,” for akratic purposes?
The right timescale might be a minute or an hour or a week rather
than a day.) I see no reason why we know these “fundamental”
parameters (or, indeed, that they truly mean anything more
fundamental, as opposed to being a convenient theoretical modeling
device) than the parameter of a week that’s been chosen here.

David

On Sep 7, 2011, at 1:04 PM, Johnicholas wrote:

I understand the value of simple, comprehensible user interfaces,
but speaking purely theoretically (and therefore impractically),
the “one week” aspect is obviously not a theoretically-motivated
parameter.

Does anyone have ideas of what the “dial” would look like if we cast
simplicity to the winds,
and try to get a theoretically-motivated number?
For example, we might assume that the user is acting based on
hyperbolic discounting,
prefers to go off the road rather than stay on the road today,
and prefers to stay on the road forever rather than go off the road
forever.

Can we use that, and some algebra and get, for example,
that the horizon ought to be some specific function of the degree of

Johnicholas

“Don’t believe everything you read on the Internet.” – Abraham Lincoln

Best regards,
Alexander mailto:miro23@gmail.com

One week is good for decreasing the slope but I think it would be good
for slope increases to happen instantly (with the ability to undo
should you make a mistake entering the new value).

It is currently a big problem that when you reset your goal you need
to wait 1 week before the YBR starts to slope at all.

Robbie

On 08/09/2011, at 6:35 AM, Alexander Schwarz miro23@gmail.com wrote:

I second that.

One week (or anything between 5 and 10 days) works fine and is well
enough, to my mind.

And I think that the akratic horizon is about more than hyperbolic
discounting. Wait, “think” is not the right term. - I feel it. It’s a
kinesthetic reference. (If that doesn’t make sense to you, never mind.

Alex

Wednesday, September 7, 2011, 10:17:01 PM, you wrote:

I like one week. It seems about right for the designated purposes.
And I’m one of the people who complained most loudly that I wanted
to be able to change my slope once my metabolism slowed down and I
didn’t realize how much it would suck to be hungry enough to stay on my YBR.

If we were to derive something theoretically, it would totally
depend on the parameters you chose for your hyperbolic discounting
function (How much do you discount tomorrow versus today, over and
above how much you discount the day after versus tomorrow? And
also, what is the actual length of “today,” for akratic purposes?
The right timescale might be a minute or an hour or a week rather
than a day.) I see no reason why we know these “fundamental”
parameters (or, indeed, that they truly mean anything more
fundamental, as opposed to being a convenient theoretical modeling
device) than the parameter of a week that’s been chosen here.

David

On Sep 7, 2011, at 1:04 PM, Johnicholas wrote:

I understand the value of simple, comprehensible user interfaces,
but speaking purely theoretically (and therefore impractically),
the “one week” aspect is obviously not a theoretically-motivated
parameter.

Does anyone have ideas of what the “dial” would look like if we cast
simplicity to the winds,
and try to get a theoretically-motivated number?
For example, we might assume that the user is acting based on
hyperbolic discounting,
prefers to go off the road rather than stay on the road today,
and prefers to stay on the road forever rather than go off the road
forever.

Can we use that, and some algebra and get, for example,
that the horizon ought to be some specific function of the degree of

Johnicholas

“Don’t believe everything you read on the Internet.” – Abraham Lincoln

Best regards,
Alexander mailto:miro23@gmail.com

Yeah, we recently decided that imposing that week of flat road is no good.

I like your proposal that you can make it harder immediately but
making it easier has the one-week delay.
That’s actually what we originally envisioned but thought a universal
one-week delay would be simpler (don’t have to deal with the undo
problem, for one thing – see the footnote in the above link). Maybe
you’re right though…

We’ll hurry and get this version deployed where the initial flat spot
is not imposed on you and then decide about the more general solution.

Johnicholas: I think I agree with David Reiley but still very
interested in seeing what a more principled model would suggest (even
if that involves equally arbitrary parameter choices).

On Wed, Sep 7, 2011 at 17:32, Robbie Clarken robbie.clarken@gmail.com wrote:

One week is good for decreasing the slope but I think it would be good
for slope increases to happen instantly (with the ability to undo
should you make a mistake entering the new value).

It is currently a big problem that when you reset your goal you need
to wait 1 week before the YBR starts to slope at all.

Robbie

On 08/09/2011, at 6:35 AM, Alexander Schwarz miro23@gmail.com wrote:

I second that.

One week (or anything between 5 and 10 days) works fine and is well
enough, to my mind.

And I think that the akratic horizon is about more than hyperbolic
discounting. Wait, “think” is not the right term. - I feel it. It’s a
kinesthetic reference. (If that doesn’t make sense to you, never mind.

Alex

Wednesday, September 7, 2011, 10:17:01 PM, you wrote:

I like one week. It seems about right for the designated purposes.
And I’m one of the people who complained most loudly that I wanted
to be able to change my slope once my metabolism slowed down and I
didn’t realize how much it would suck to be hungry enough to stay on my YBR.

If we were to derive something theoretically, it would totally
depend on the parameters you chose for your hyperbolic discounting
function (How much do you discount tomorrow versus today, over and
above how much you discount the day after versus tomorrow? And
also, what is the actual length of “today,” for akratic purposes?
The right timescale might be a minute or an hour or a week rather
than a day.) I see no reason why we know these “fundamental”
parameters (or, indeed, that they truly mean anything more
fundamental, as opposed to being a convenient theoretical modeling
device) than the parameter of a week that’s been chosen here.

David

On Sep 7, 2011, at 1:04 PM, Johnicholas wrote:

I understand the value of simple, comprehensible user interfaces,
but speaking purely theoretically (and therefore impractically),
the “one week” aspect is obviously not a theoretically-motivated
parameter.

Does anyone have ideas of what the “dial” would look like if we cast
simplicity to the winds,
and try to get a theoretically-motivated number?
For example, we might assume that the user is acting based on
hyperbolic discounting,
prefers to go off the road rather than stay on the road today,
and prefers to stay on the road forever rather than go off the road
forever.

Can we use that, and some algebra and get, for example,
that the horizon ought to be some specific function of the degree of

Johnicholas

“Don’t believe everything you read on the Internet.” – Abraham Lincoln

Best regards,
Alexander mailto:miro23@gmail.com

http://dreev.es – search://“Daniel Reeves”

I think getting rid of the imposed flat week will probably fix Robbie’s problem to first order. I’ll be curious to see if it doesn’t.

David

On Sep 7, 2011, at 4:19 PM, Daniel Reeves wrote:

Yeah, we recently decided that imposing that week of flat road is no good.

I like your proposal that you can make it harder immediately but
making it easier has the one-week delay.
That’s actually what we originally envisioned but thought a universal
one-week delay would be simpler (don’t have to deal with the undo
problem, for one thing – see the footnote in the above link). Maybe
you’re right though…

We’ll hurry and get this version deployed where the initial flat spot
is not imposed on you and then decide about the more general solution.

Johnicholas: I think I agree with David Reiley but still very
interested in seeing what a more principled model would suggest (even
if that involves equally arbitrary parameter choices).

On Wed, Sep 7, 2011 at 17:32, Robbie Clarken robbie.clarken@gmail.com wrote:

One week is good for decreasing the slope but I think it would be good
for slope increases to happen instantly (with the ability to undo
should you make a mistake entering the new value).

It is currently a big problem that when you reset your goal you need
to wait 1 week before the YBR starts to slope at all.

Robbie

On 08/09/2011, at 6:35 AM, Alexander Schwarz miro23@gmail.com wrote:

I second that.

One week (or anything between 5 and 10 days) works fine and is well
enough, to my mind.

And I think that the akratic horizon is about more than hyperbolic
discounting. Wait, “think” is not the right term. - I feel it. It’s a
kinesthetic reference. (If that doesn’t make sense to you, never mind.

Alex

Wednesday, September 7, 2011, 10:17:01 PM, you wrote:

I like one week. It seems about right for the designated purposes.
And I’m one of the people who complained most loudly that I wanted
to be able to change my slope once my metabolism slowed down and I
didn’t realize how much it would suck to be hungry enough to stay on my YBR.

If we were to derive something theoretically, it would totally
depend on the parameters you chose for your hyperbolic discounting
function (How much do you discount tomorrow versus today, over and
above how much you discount the day after versus tomorrow? And
also, what is the actual length of “today,” for akratic purposes?
The right timescale might be a minute or an hour or a week rather
than a day.) I see no reason why we know these “fundamental”
parameters (or, indeed, that they truly mean anything more
fundamental, as opposed to being a convenient theoretical modeling
device) than the parameter of a week that’s been chosen here.

David

On Sep 7, 2011, at 1:04 PM, Johnicholas wrote:

I understand the value of simple, comprehensible user interfaces,
but speaking purely theoretically (and therefore impractically),
the “one week” aspect is obviously not a theoretically-motivated
parameter.

Does anyone have ideas of what the “dial” would look like if we cast
simplicity to the winds,
and try to get a theoretically-motivated number?
For example, we might assume that the user is acting based on
hyperbolic discounting,
prefers to go off the road rather than stay on the road today,
and prefers to stay on the road forever rather than go off the road
forever.

Can we use that, and some algebra and get, for example,
that the horizon ought to be some specific function of the degree of

Johnicholas

“Don’t believe everything you read on the Internet.” – Abraham Lincoln

Best regards,
Alexander mailto:miro23@gmail.com

http://dreev.es – search://“Daniel Reeves”