As for the topic of “gender detecting”, a “receptionist” for an office is usually female, for a hotel is usually male; a “nurse” is usually female, a “taxi driver” is usually male, etc. So if the bot is wrong in making that assumption, it’s actually very human-like!

I would love to see you say that in a room full of women. Please wait until I’ve got the camera.

Jan Bogaerts - Feb 6, 2012:

As for the topic of “gender detecting”, a “receptionist” for an office is usually female, for a hotel is usually male; a “nurse” is usually female, a “taxi driver” is usually male, etc. So if the bot is wrong in making that assumption, it’s actually very human-like!

I would love to see you say that in a room full of women. Please wait until I’ve got the camera.

Good one Jan.

rule of thumb #1. humans, in general wear pajamas more so than elephants.
rule of thumb #2. females, in general are nurses more than males.

So these probabilities change over time.

Rule #2 held for decades, up to say the 70’s perhaps.  Now rule #2 is not so much.  The bot should not perhaps assume this rule now-a-days.

However, I think Rule#1 is at a quite different level of assurance

It is pretty easy to know which rules can be assumed and which not.

For example, whether it is going to rain 3 weeks from today is a bit uncertain, but I think I can assume tomorrow the sun will still rise in the east and set in the west (Rule #3).

Common sense pretty much tells us which rules our bots can assume, and which not.

There are some, like Rule 3, which perhaps won’t be able to be assumed fifty million years from now, and others like Rule 1, which can be assumed for the next thousand, until elephants perhaps evolve and start thinking like humans.

But then other categories, like Rule 2, which like Jan pointed out, is pretty much not a rule anymore.

I hardly think though, that we need to ‘calculate’ the relative weight of these rules.  I think pretty much anyone could put these rules into categories of how long they are likely to be in effect… and simply enter them into a chat bot, along with there relative “time to live” approximation (whether it be days, years, decades, millennium or millions of years… ‘ball park’ estimate is good enough)

Again, these rules are just that.. rules of thumb, and if the bot makes an error, doesn’t matter, people also do make the same errors.

You simply want to minimize the number of errors made… and these rules of thumb would help in that way.

Some of these rules would also be affected by context.  For example, in Iran, women aren’t allowed to do many things that men are allowed… so the rules would be adjusted if the context of the conversation was Iran. You get the idea.

Perhaps this Time-To-Live factor, associated with each category, could be such that we would have categories of : a) physics b) human employment trends, c) animal versus human behaviors ,etc.

Jan Bogaerts - Feb 6, 2012:

As for the topic of “gender detecting”, a “receptionist” for an office is usually female, for a hotel is usually male; a “nurse” is usually female, a “taxi driver” is usually male, etc. So if the bot is wrong in making that assumption, it’s actually very human-like!

I would love to see you say that in a room full of women. Please wait until I’ve got the camera.

  but *usually* it is true. maybe the nurse one is bad but how about the taxi driver? or captain of a submarine? or a chief cook? you know better… 

Victor Shulist - Feb 6, 2012:

Jan Bogaerts - Feb 6, 2012:

As for the topic of “gender detecting”, a “receptionist” for an office is usually female, for a hotel is usually male; a “nurse” is usually female, a “taxi driver” is usually male, etc. So if the bot is wrong in making that assumption, it’s actually very human-like!

I would love to see you say that in a room full of women. Please wait until I’ve got the camera.

Good one Jan.

rule of thumb #1. humans, in general wear pajamas more so than elephants.
rule of thumb #2. females, in general are nurses more than males.

So these probabilities change over time.

Rule #2 held for decades, up to say the 70’s perhaps.  Now rule #2 is not so much.  The bot should not perhaps assume this rule now-a-days.

However, I think Rule#1 is at a quite different level of assurance

It is pretty easy to know which rules can be assumed and which not.

For example, whether it is going to rain 3 weeks from today is a bit uncertain, but I think I can assume tomorrow the sun will still rise in the east and set in the west (Rule #3).

Common sense pretty much tells us which rules our bots can assume, and which not.

There are some, like Rule 3, which perhaps won’t be able to be assumed fifty million years from now, and others like Rule 1, which can be assumed for the next thousand, until elephants perhaps evolve and start thinking like humans.

But then other categories, like Rule 2, which like Jan pointed out, is pretty much not a rule anymore.

I hardly think though, that we need to ‘calculate’ the relative weight of these rules.  I think pretty much anyone could put these rules into categories of how long they are likely to be in effect… and simply enter them into a chat bot, along with there relative “time to live” approximation (whether it be days, years, decades, millennium or millions of years… ‘ball park’ estimate is good enough)

Again, these rules are just that.. rules of thumb, and if the bot makes an error, doesn’t matter, people also do make the same errors.

You simply want to minimize the number of errors made… and these rules of thumb would help in that way.

Some of these rules would also be affected by context.  For example, in Iran, women aren’t allowed to do many things that men are allowed… so the rules would be adjusted if the context of the conversation was Iran. You get the idea.

Perhaps this Time-To-Live factor, associated with each category, could be such that we would have categories of : a) physics b) human employment trends, c) animal versus human behaviors ,etc.

if the bot is processing an article from a newspaper in 1969, is it reasonable to assume “a nurse” is a female?  Then how about the newspaper from 1970, 1971, ... and when is the turning point of stoping this assumption? or better to use a decreasing ratio, eg 90% in 1970,  89% in 1971? ...

John Li - Feb 7, 2012:

if the bot is processing an article from a newspaper in 1969, is it reasonable to assume “a nurse” is a female?

 

sure, I’d say yes, if it has no other hints (nothing said so far in the conversation that would change that).

John Li - Feb 7, 2012:

Then how about the newspaper from 1970, 1971, ... and when is the turning point of stoping this assumption? or better to use a decreasing ratio, eg 90% in 1970,  89% in 1971? ...

Yes, perhaps something like that, which is what i was getting at with the “Time To Live” idea.

John Li - Feb 7, 2012:

  but *usually* it is true. maybe the nurse one is bad but how about the taxi driver? or captain of a submarine? or a chief cook? you know better… 

 

Having driven cab in two different cities for over 7 years, I can say from personal experience that there are just as many female taxi drivers as there are men. Saying that *any* particular profession is predominantly one gender or another is a form of stereotyping. Even certain “Adult Entertainment” professions are not as one-sided, gender-wise, as you may think. I would cite examples, but this is a family friendly forum, and I’m certain that you can figure out for yourself what I’m referring to.

But this is straying from the topic at hand, so let’s drop this line of discussion for now, and concentrate on other possibilities, shall we?

if the bot is processing an article from a newspaper in 1969, is it reasonable to assume “a nurse” is a female?  Then how about the newspaper from 1970, 1971, ... and when is the turning point of stoping this assumption? or better to use a decreasing ratio, eg 90% in 1970,  89% in 1971?

This reminds me of a logic riddle: if you have a million grains of sands, that’s a pile of sand, if you remove 1 grain, it still remains a pile, but if you remove 999.999 grains, you no longer have a pile.When did the pile of grain stop to be a pile?

Yes, the threshold problem.  When people have the same issue ‘drawing that line’, we shouldn’t worry that bots will have also.  If there is no way of determining where that line should really be, then the bot picking its own is not a problem.  I think below zero Celsius is cold, but if you ask my wife, anything below 25 C is cold !  If it is subjective, everyone is entitled to their opinion.  But when it is far removed from the ‘grey area’, the number of disagreements would drop dramatically.  So you’ll have much less arguments about whether an elephant wears pajamas than you would about the usual sex of a ‘nurse’. 

In fact, for the most part, the only time you will have someone put up a fuss about the possibility of an elephant wearing pajamas, is when someone is trying to argue there point of view about AI or chatbots.

That’s just a general comment John (I wasn’t poking at you! )

Brings up an interesting question—Can a Chatbot be allowed to have an opinion?!!  If we’re in the ‘grey area’ (like ‘how cold is cold’), and humans are allowed to have their own definition, can’t a chatbot also?  We tend to think… oh… gotcha !  your bot has a fixed value for what it thinks ‘cold’ is… but how it it arrive at that?  oh? it doesn’t have a complex algorithm or something.. it just thinks X degrees is cold?  bad bad… can’t do that… but yet humans are allowed to do that !! why is that?!

Subjectivity require fuzzy logic type approaches.

Think about the word “cool” in the following lines:

That was cool!
My coffee was cool.
In the summer my air conditioner keeps me cool.
My air conditioner was only cool.

The good or badness of each “cool” is relative.

Regarding the “elephant” in the room:
Resolving who was wearing pajamas could be deferred (and may not be required at all). It could be resolved by clues later in the conversation.

What may be harder (and possibly more important) is that shooting something while you (or it) is in pajamas, is an unusual event. It may/should require additional follow up statements/questions based on the “surprise” factor.

Good observation Merlin.