[Aleader-dev] Re: comments on paper

[Joshua N Pritikin]

Here is today's big email.

> WLJ okay, here is what i wrote on my first read.  When i really get
> serious about commenting on this there will likely be many many more
> edits.  I'm sorta testing the waters right now.  Okee dokee?

Sure, right.  No point in going too far too soon.

For my replies, I've tried to sort your questions by order importance
(in case you run out of time half-way through reading this long email).

> have you heard of emotional intelligence (EI)?  I even think
> there is EQ or EIQ (emotional intelligence quotient).  EI was kind of a
> fad, I think it is fading now.  Daniel Goleman was one its its big
> popularizers.  He tried to make business interested in it, arguing that
> hi EQ is what leads to successful leadership abilities in the business
> world.  Salovey and Mayer are more academic types who have put a lot
> of energy into the area.

First of all, I consider "Emotional Intelligence" as a particular
sub-type of "Personality Development".  Is this a reasonable
classification?

Here's what I wrote in the Aleader manual.  (I know the writing style
is informal.  Just read it for the general idea.)

+++

Personality development generally focuses on the student's reaction to a
given situation.  Certainly choosing the correct reaction @emph{is}
important.  However, deciding on a reaction depends on assessing the
situation accurately.  If my AQ is below average then attending a
personality development course will do more harm that good.  I can learn
new, better reactions but I will often use them at the wrong time and it
will seem artificial.

What if we could objectively measure AQ?  If my assessment quotient is
below average then i can invest time in raising my AQ, clearing the way
for constructive personality development.

+++

In short, I argue that AQ is a pre-requisite for effective personality
development.

Next, I'll tackle the crucially important question of "why film?"

> Written stories are good enough.  If an implemented theory
> can handle distinguishing happy/indifferent/sad reactions in Sally in
> response to...
>
> "Mommy gives Sally a [apple, slap on the face, chunk of coal,
> piece of celery, blank stare]"
>
> ...We'll be doing great.  Thus, it is premature to worry about extremely
> rich scenarios such as Star Wars quite yet....Thus, although it would be
> fun to watch films and rate emotions seen via the Aleader thing, I am
> not sure how it would be a efficient way of building an AI that has a clue
> inferring human emotional reactions.  In short, why not focus on simple
> textual stories.  Once those work, then tackle film.
>
> ...
>
> WLJ as I said in previous email, Why not start with a written story?
>
> Nonetheless, I am not convinced that at this level of AI development, that
> we need film. 

It's not a question of the our current level of AI development.  There
are at least two reasons to use film.  The first reason is
_repeatability_.  A written story relies on imagination too much.  For
example,

  "Mommy gives Sally a slap on the face."

Questions arise in the mind.  Why?  How hard was the slap?  Perhaps it
was a relatively gentle slap?  Did Sally take it as a joke or gravely?
Etc.

Digression is a real danger here.  I argue that writing a completely
unambiguous story is a difficult task.  I wouldn't want to read such a
story either.  However, without eliminating the ambiguity, we don't
achieve repeatability.  Two people can read the same story and come up
with different interpretations.  You "explain" this with generativity,
but to me it just sounds like hand-waving.  OK, you get the idea.

The other reason to use film is more subtle.  As you know, cognitive
psychology posits that there is a conceptual process prior to emotion.
I want to point out that there is a perceptual process prior to
conception, like this:

  Perceptual -> Conceptual -> Emotional

The challenge is to isolate each of these steps as much as possible so
we can reflect the exact role that each step plays in our life
process.  (I know a good reference book for this if you want to read
more.)

Film is perceptual, an experience, while a written story is
conceptual, made of words.  With a written story, it is easy to get
mixed up between the conceptual story and our conceptual analysis of
that story.  With film, a clean separation between the experience and
the conceptual analysis is more tractable.

The other point I want to mention is that I have already done the hard
work of making working with film technically feasible.  This is new in
the sense that prior to 2003, only very expensive software could make
film so easy to handle.  Once you play with the Aleader CD then you
will see how easy I've made it.  If you want more encouragement, take
a look at this comment:

  http://mail.nongnu.org/archive/html/aleader-dev/2003-06/msg00005.html

> ... Film contains confounds.
> E.g. if someone is wearing a red shirt, that might heighten the
> overall sense of anger or arousal in the film.  But a Cyc-like
> system (i.e. strictly inferencing AI system with e.g. no vision
> system) won't really have a good rep capacity for such things.  
>
> [and]
>
> WLJ Connecting images of faces to emotions is a lot different than
> connecting situation to emotions.  A Cyc-like system is much better
> with situations.  Faces are better with s.t. more ineffable like a
> neural net or a genetic algorithm.

I certainly don't want to get side-tracked into image recognition,
extracting features from an image automagically.  Rather, I see the
research in automated facial expression or body language recognition
as a guideline for what we can reasonably objectively infer.

For example, if I see a particular type of facial expression then I
want to know whether I can objectively assert that "this person's face
looks angry."  Or if there is research that says "if someone is
wearing a red shirt, that might heighten the overall sense of anger or
arousal in the film" then I have a justification for entering that
assertion if I see a red shirt in a film.

In all cases, this is a a fully manual process.  I (or whoever is
analyzing the film) is making the judgment about is what the facial
expression, what is the color of the shirt, etc and hand entering the
assertions into the KR along with a frame timestamp.

Yes this is tedious, but we don't need to model _everything_.  We just
need to make a good faith effort to model all material evidence and
avoid accusation of cherry picking or manipulating the evidence to
produce the results we want to achieve.  If we can model enough of a
film to compute 3-5 situations then it may be sufficiently impressive.

> WLJ sorry if you want to publish you have got to say stuff like what
> was your sample, how many subjects actually said "YEs, I get it."

Yes, of course.  I put a big warning in the abstract that this stuff
is missing.  Maybe you skipped the abstract.  ;-)

I have been trying to devise a plan for gathering statistics.  The
challenge to keep it fun, while making it scale up to such a big
model.  I think I have something that will work.  I call it the
"Guessing Game."  Take a look at:

  http://savannah.nongnu.org/download/aleader/htdocs/empathy/guess1.html

Now click "Correct Analysis" at the bottom of the page.

Three of the situations classify to the same emotion.  However, one of
the situations is different.  You are challenged to pick out the
differing situation.  I speculate that I can convince people to go
through about 40 slides (maybe requiring 1-2 hours).

I expect that some pairs of emotions will be hard to distinguish while
other pairs will be easier.  From the results, we should be able to
show that people have trouble distinguishing certain facets of a
situation.  OK, something like that.

Here's the big picture:

[1] The Guessing Game will statistically prove that even "normal" people
have difficulty assessing situations.

[2] A formal KR model will demonstrate the existence of a completely
objective way to assess situations.

Accomplishing just these two studies might be enough to publish a
research report.  I'm not sure about the next steps:

[3] The subjective believability question remains: do our emotion
categories actually feel like a good fit?  To answer this, we can't
just use average people because average people aren't very good at
assessing situations.[1] First, we have to train people how to assess
situations. Then we can ask them about the emotional reasonableness or
believability.  Your empirical methods apply.  However, I'm not sure
how to argue that the training won't interfere with the subjective
judgment.  (I can expand on this if it is not clear.)

[4] Copy Goleman: Try to find a statistical link between career
success and situation assessment ability.  This kind of study would
require quite a lot of effort.  I think we would need the support of
an institution.  At least, there is no way I could do this on my own.
On the other hand, this is a target at which to aim.  There is a lot
more demand for general career success as compared to clinical
treatment of autism.  That is not to say that abnormal psychological
conditions such as autism could not find benefit from Aleader.

> WLJ What is the definition of a basic emotion?  Gallons of ink have
> been spilled in the literature atttempting to decide 1) are there
> basic emotions 2) definitions of basic emotions 3) lists of basic
> emotions.  Typically they include stuff like happy, sad, mad,
> afraid.

Yikes.  I want to avoid this question.  I will strike the words from
the article.

Isn't it enough to offer statistics from the Guessing Game and a
formal KR analysis of the Aleader classification system?

> WLJ RECURRING QUESTION: what is the overarching goal here --
> teaching people to increase their AQ or building an AI that can
> reason about other's emotional states?  

My overarching goal is to improve society.  I believe that teaching
people to how to increase their AQ is part of that.  Given that you
are pursuing a clinical program (as opposed to military or
pseudo-military game with artifical intelligence), I expect that you also
aim to improve society.  I think you said as much in a previous email:

> I am in a Counseling PSychology PhD program.  There is quite a
> large distance between clinical psych and ai.  Autism seems like on of the
> few possible bridging areas in the nearer term.  In the longer term, the
> bridging territory is huge.  But right now, the gap between the two
> disciplines is painfully large.

Am I mistaken about your overarching interests?

In my mind, building the AI part is just a formality to convince
people that AQ is a viable theory.  On the other hand, it's a very,
very important formality because it forces us to spell everything out
in excruciating detail, which makes a convincing proof.

> > Once evocative examples are classified then we must subjectively judge
> > whether the classification matches our empathy sense.
> > Does each smiling face really evoke a happy feeling?
>
> WLJ I think evoke is a misleading term here.  What you mean is "If
> someone is smiling does that really mean they are happy?"  Your use
> of evoke suggests that the viewer is caused to feel happy by the
> actors smiling.  What you are interested in is "mindreading"
> (inferring) others emotions, not evocation.

I am still in the process of refining terminology.

We probably need to discuss this again, later.  I need to read up on
the academic literature -- see what terms people are using.  You
probably need to try reading the Philosophy chapter in the Aleader
manual, at least the part about attention and the 1st, 2nd & 3rd
person perspectives.

> WLJ Where are you interested in publishing this?

I don't have any pre-defined direction.  Aleader was developed outside
the academic community.  I've only recently (2-3 weeks ago) started
looking back to academia for parallel projects or similar research
topics.  To make any more progress, I need to gain a lot of
credibility and visibility.  I'd welcome any advice you could provide
in this regard.  Finding someone like you could be (is?) a real
breakthrough.

Another problem is due to where I live.  Here in Maharashtra, most
people don't speak English very well.  The pool of canidates available
for experimentation, typically undergrad psychology students, is
severly limited.  On the other hand, I recall that you conducted your
experiments via email.  So maybe it boils down to a matter of
credibility and visibility.

> Question: Who is on address@hidden  Just you or more than you?

There are two people who are just watching.  They hardly ever post
anything.  There are about 30 people on aleader-announce.  Almost none
of them are active either.  However, there are a few exceptions:

+ Sagar Behere is a recently graduated mechanical engineer here in
Nashik (my home town in Maharashtra).  He has been very helpful --
reading and commenting on most of the stuff I have done.  He said he
will introduce me to a psychologist friend next week.  He thinks
Aleader will be useful as another psychology test.  On the other hand,
his primary interest is robotics & free software.  He isn't interested
in deep theory or getting much more involved himself.

+ My mom has made some helpful comments occationally.  She promised me
that she would try out the recent stuff I've done with the CD soon.

+ My wife's support has been limited to, "is it done yet?"  ;-)

Recently, a Ben Yogman sent email and said he was a former Cyclist and
was interested.  However, I found out that he's more interested in OCR
(optical character recognition), at least in the short term.  He said
something about needing money.

> WLJ I have significant student loans.
> ...
>
> Who knows, maybe a business could be created out of this work.  Well,
> starting a business is a hugely risky proposition, especially when one
> has loan payments to make.

Yah, don't even think about it.

If I somehow get a business running around AQ education then you are
the kind of person I would look to hire.  But I am still a long way
from starting a business.  It's not even on the map yet.

> Any comments/questions/reactions?

Have you thought about working for a business that you don't care
about much but pays well?  It's a waste of talent, I know, but money
is a practical necessity.

> Google just taught me sorta what an ISO is - a file containing an 
> image of a linux distro...Do you know what ISO stands for? (that will
> help me remember what ISO is).

ISO is an abbrev of ISO-9660 which is the main standards document
describing how to record a computer file system on a CD (as opposed to
audio).  Does that help?  ;-)

> Well, I would like to try it out but I don't have a CD burner.  Maybe
> I should use my student loan money to buy one, they seem pretty useful
> pretty leveraging.

If you do purchase a CD-Writer, be sure to get one with Burn-Proof or
Burn-Free protection.  However, considering your tight financial
situation, I can't recommend it.

Do you have any friends who are computer science majors?  CD-Writers
are so cheap now, I can't imagine that a CompSci major would be
without one.

If you really can't find a CD-Writer then you can probably find a
Linux User Group in your area.  Here's a list:

  http://www.linux.org/groups/usa/texas.html

Contact them and I'm sure someone will help you out.  Aleader runs on
Linux, after all.

And now for an answer to your least important question:

> > ...  Do you know how to use CVS?
>
> I've used it a little when I was at Cycorp.  It's been 2 years since
> I've been there.  Be forewarned -- is it okay if I ask you some dumb
> newbie questions?

Yah, no problem.  :-)

-- 
Victory to the Divine Mother!!         after all,
  http://sahajayoga.org                  http://why-compete.org