Should robots lie?

chornedsnorkack.DeleteThis@hushmail.com wrote:
> Should robots be allowed to lie?
>
> The 3 laws were invented by humans to keep robots safe and to alleviate
> their Frankenstein complex. Also, robots were designed only to work in
> workshops and stations, where each and every human they might encounter
> would be authorized to order them around and could be held accountable
> by disciplinary action in case of misuse (quite apart from the robot's
> own dities to give priority to the orders of people most authorized to
> use it and to the First Law). In a sense, a robot is equivalent to a
> car with no keys, so anyone who reaches it can enter and drive.
>
> But when a robot has conflicting motives from sundry First Law
> possibilities of "harming" and "allowing to come to harm" humans and
> obeying/disobeying different orders, how can the situation be sorted
> out?
>
> For example, the Liar perceived human emotions directly - and lied. But
> obviously apart from telepathy, the robots might deduce emotional state
> of humans from analysis of their behaviour, just as other humans do.
> Should "emotional harm" be relevant to First Law?
>
> Of course, it is what makes for the job of Robot Psychologist, and for
> story plots. But is the problem intrinsic to the nature of logic? Or to
> the incentives for robot design?
>
> So, suppose that a robot is under conflicting orders, given by several
> humans, with several levels of urgency. Should a robot give a true
> account of the reasons it has for rejecting an order? Provided perhaps
> that the account is requested by someone high enough.
>
> And shouldn't robots be under laws to state that they withhold
> information, rather than lying?


I don't think Asimov's laws are feasible in real life, and partly
because of the vagueness of the "do no harm" law. Making fat people
exercise does some immediate harm and immediate increase in death rate,
as opposed to a longer term benefit.
Likewise, every living creature tries to "be fruitful and multiply" .
In doing so, every living creature is reducing its own life expectancy
in an effort to increase the fraction of its own genes passed on to the
next generation.- A. McIntire
 >> Stay informed about: Should robots lie? 

IsaacKuo wrote:

> The 3 laws of robotics were invented by Isaac Asimov so he
> could write interesting stories around them. They have not,
> AFAIK, ever been implemented in any meaningful way in any
> actual robot.

Somewhere amongst my books there is one where Asimov claim the
realization that his three laws of robotics are just a special case of
the four? laws of tools:

1st. A tool may not harm people. If possible a tool must prevent harm to
people.

2nd. A tool should perform the desired function, unless this would cause
harm to people.

3rd. A tool should not break, unless this is integral to its function or
breaking is necessary to prevent harm to people.

4th. I don't recall. Maybe there was no 4th?

> Indeed, a huge class of robots are specifically designed to
> violate the first and third laws--some of them on a truly
> apocalyptic scale.
>
> Isaac Kuo
>

Presumably these obey the n laws of weapons.
 >> Stay informed about: Should robots lie? 

"Jens E. Nyborg" <spamtrap DeleteThis @aabent.dk> writes:

[Schnipp]

>4th. I don't recall. Maybe there was no 4th?

This good be a good base for a story.

(IIRC somewhere in the later Foundation-episodes a
zero-th law was introduced)
--
Jan v/d Broek
balglaas DeleteThis @dds.nl
 >> Stay informed about: Should robots lie? 

Jens E. Nyborg wrote:
> IsaacKuo wrote:
>
> > The 3 laws of robotics were invented by Isaac Asimov

Really? Where is "Isaac Asimov" mentioned?

In the robot tales, e. g. Susan Calvin, a robot psychologist, is busy
troubleshooting. She talks behind the backs of the people who designed
the Lost Robot and relieved it of the second part of the First Law...
She does not talk with them or name their names. But they existed.

Is the inventor of positronic brain and Three Laws ever named?

> > so he
> > could write interesting stories around them. They have not,
> > AFAIK, ever been implemented in any meaningful way in any
> > actual robot.
>
> Somewhere amongst my books there is one where Asimov claim the
> realization that his three laws of robotics are just a special case of
> the four? laws of tools:
>
> 1st. A tool may not harm people. If possible a tool must prevent harm to
> people.
>
Not a general rule for tools.

A large amount of tools are made solely to harm people.

For much of the rest of tools, the abilty to harm humans is a byproduct
of the tool's ability to do the desired function. It is accepted that
it is the responsibility of the human users of the tool to avoid
harming humans.

> 2nd. A tool should perform the desired function, unless this would cause
> harm to people.
>
> 3rd. A tool should not break, unless this is integral to its function or
> breaking is necessary to prevent harm to people.
>
> 4th. I don't recall. Maybe there was no 4th?
>
> > Indeed, a huge class of robots are specifically designed to
> > violate the first and third laws--some of them on a truly
> > apocalyptic scale.
> >
> > Isaac Kuo
> >
>
> Presumably these obey the n laws of weapons.

So. Why are robots emphatically NOT weapons?

Presumably in order to alleviate Frankenstein Complex. If someone is
building weapons, this is the pretext to stop this activity. Making
robots was allowed because they had First Law and therefore could not
become weapons, even at the hands of their makers.

But were the Three Laws well designed to meet their goals?
 >> Stay informed about: Should robots lie? 

On Thu, 8 Sep 2005, Jens E. Nyborg wrote:

>
> Presumably these obey the n laws of weapons.
>

Well, (semi)invert the tools laws:
1) A weapon should not break unless unless this is integral to its
function or is necessary to harm to people.
2) 2nd. A tool should perform the desired function, unless this would
prevent harm to people.
3) A weapon must harm people. If possible a tool must cause
addition damage such as will cause harm to people.


Tighting up a bit:
1) A weapon must function when used*.
2) When used against an enemy, a weapon must inflict maximum harm on that
enemy.
3) A weapon must not harm friendly forces unless this will reduce harm to
the enemy or otherwise prevent funtioning.

[And the zeroth law, a weapon will make a -large- amount of money for the
defense contractor]

3ch

*Though possibly not used directly against an emeny --e.g. covering fire,
show of force, etc.
 >> Stay informed about: Should robots lie? 

alanmc95210.RemoveThis@yahoo.com wrote:
> chornedsnorkack.RemoveThis@hushmail.com wrote:
>> Should robots be allowed to lie?
>>
>> The 3 laws were invented by humans to keep robots safe and to
>> alleviate their Frankenstein complex. Also, robots were designed
>> only to work in workshops and stations, where each and every human
>> they might encounter would be authorized to order them around and
>> could be held accountable by disciplinary action in case of misuse
>> (quite apart from the robot's own dities to give priority to the
>> orders of people most authorized to use it and to the First Law). In
>> a sense, a robot is equivalent to a car with no keys, so anyone who
>> reaches it can enter and drive.
>>
>> But when a robot has conflicting motives from sundry First Law
>> possibilities of "harming" and "allowing to come to harm" humans and
>> obeying/disobeying different orders, how can the situation be sorted
>> out?
>>
>> For example, the Liar perceived human emotions directly - and lied.
>> But obviously apart from telepathy, the robots might deduce
>> emotional state of humans from analysis of their behaviour, just as
>> other humans do. Should "emotional harm" be relevant to First Law?
>>
>> Of course, it is what makes for the job of Robot Psychologist, and
>> for story plots. But is the problem intrinsic to the nature of
>> logic? Or to the incentives for robot design?
>>
>> So, suppose that a robot is under conflicting orders, given by
>> several humans, with several levels of urgency. Should a robot give
>> a true account of the reasons it has for rejecting an order?
>> Provided perhaps that the account is requested by someone high
>> enough.
>>
>> And shouldn't robots be under laws to state that they withhold
>> information, rather than lying?
>
>
> I don't think Asimov's laws are feasible in real life, and partly
> because of the vagueness of the "do no harm" law. Making fat people
> exercise does some immediate harm and immediate increase in death
> rate, as opposed to a longer term benefit.
> Likewise, every living creature tries to "be fruitful and multiply" .
> In doing so, every living creature is reducing its own life expectancy
> in an effort to increase the fraction of its own genes passed on to
> the next generation.- A. McIntire

Isn't that covered by the "Zeroeth" law ?
 >> Stay informed about: Should robots lie? 

chornedsnorkack.DeleteThis@hushmail.com wrote:
> But were the Three Laws well designed to meet their goals?

Yes. Their goals were:

1) Be short enough to be usable for short story purposes (where the laws
must be restated at the start of each story for new readers, without
annoying old readers with a long infodump).

2) Sound plausible enough for willing suspension of disbelief.

3) Be sufficiently vague and loophole-rich to lead to lots of
story-usable trouble.

Empirically, they did a fine job of meeting those goals.

--
"Always look on the bright side of life."
To reply by email, replace no.spam with my last name.
 >> Stay informed about: Should robots lie? 

In the book _Good Omens_, Gaiman & Pratchett don't resist telling of a
couple of the dirty tricks their resident devil has done. We chuckle
and enjoy them. But there's no corresponding joyful story about what
his angelic counterpart has done. This is a comedy after all, and
what's so funny about good deeds?

But more so - it's not always easy to recognize what angels would do to
spread good.

Campbell wrote Asimov's three laws of robotics with the understanding
that it could be innately be obvious to any robot what "harm" is.
After working with this for a while, Asimov combines two of his future
histories by adding the zeroeth law which recognizes this.

Men of good will spend lots of time arguing about what is good, proper,
helpful, and what is evil, improper, and harmful, with considerable
disagreement. How is someone going to program into the basic
positronic brain design a basis where these definitions are immediately
obvious? Everything we do is a trade-off, and so many things we do
have unforeseen consequences.

(I'm sure everybody in this thread has already though of "Liar!")

So moving to the topic of this thread. It's easier to approach telling the
truth. But where it matters, people put lots of qualifications on their
statements so that we can judge for ourselves how true it is. PI is 3 to
one significant digit, but I've read people claim that a biblical reference
to something 1 unit across and 3 around proves the Bible is lying. Can we
have every statement full of qualifications so that you can tell how true it
is? Because very few things are absolute.
 >> Stay informed about: Should robots lie? 

On 9-Sep-2005, throopw.RemoveThis@sheol.org (Wayne Throop) wrote:

> : In the book _Good Omens_, Gaiman & Pratchett don't resist telling of a
> : couple of the dirty tricks their resident devil has done. We chuckle
> : and enjoy them. But there's no corresponding joyful story about what
> : his angelic counterpart has done. This is a comedy after all, and
> : what's so funny about good deeds?
> : But more so - it's not always easy to recognize what angels would do to
> : spread good.
>
> You mean... if you are God, when you do things right
> people won't be sure you've done anything at all?

The ending of that book has a similar thought, but it's a bit of a spoiler.
 >> Stay informed about: Should robots lie? 

In alt.books.isaac-asimov chornedsnorkack DeleteThis @hushmail.com wrote:
: Should robots be allowed to lie?

: For example, the Liar perceived human emotions directly - and lied. But
: obviously apart from telepathy, the robots might deduce emotional state
: of humans from analysis of their behaviour, just as other humans do.
: Should "emotional harm" be relevant to First Law?

For any robot advanced enough to understand the concept, the answer is
clearly "yes," at least according to Asimov. The issue comes up in
more than one story -- e.g. Tony's "making love" to his mistress in
"Satisfaction Guaranteed," or Andrew's realization in "The Bicentennial Man"
that what looks like harm in the short term can be good in the long term.

-----
Richard Schultz schultr DeleteThis @mail.biu.ac.il
Department of Chemistry, Bar-Ilan University, Ramat-Gan, Israel
Opinions expressed are mine alone, and not those of Bar-Ilan University
-----
"Logic is a wreath of pretty flowers which smell bad."
 >> Stay informed about: Should robots lie? 

howard.TakeThisOut@brazee.net wrote:
> In the book _Good Omens_, Gaiman & Pratchett don't resist telling of a
> couple of the dirty tricks their resident devil has done. We chuckle
> and enjoy them. But there's no corresponding joyful story about what
> his angelic counterpart has done. This is a comedy after all, and
> what's so funny about good deeds?
>
> But more so - it's not always easy to recognize what angels would do to
> spread good.
>
> Campbell wrote Asimov's three laws of robotics with the understanding
> that it could be innately be obvious to any robot what "harm" is.
> After working with this for a while, Asimov combines two of his future
> histories by adding the zeroeth law which recognizes this.
>
> Men of good will spend lots of time arguing about what is good, proper,
> helpful, and what is evil, improper, and harmful, with considerable
> disagreement. How is someone going to program into the basic
> positronic brain design a basis where these definitions are immediately
> obvious? Everything we do is a trade-off, and so many things we do
> have unforeseen consequences.
>
> (I'm sure everybody in this thread has already though of "Liar!")
>
> So moving to the topic of this thread. It's easier to approach telling the
> truth. But where it matters, people put lots of qualifications on their
> statements so that we can judge for ourselves how true it is. PI is 3 to
> one significant digit, but I've read people claim that a biblical reference
> to something 1 unit across and 3 around proves the Bible is lying. Can we
> have every statement full of qualifications so that you can tell how true it
> is? Because very few things are absolute.

True, a robot must also be able to sum up the information at its
disposal to be short enough, and accurate for its purposes, to be
delivered to humans. Or to be stored in preference to other
information.

But how can a robot observe the First Law and give it any preferential
status if it already does not have a policy of ascertaining, evaluating
and memorizing "Truth"?

A robot must have understanding of its surroundings, effects its
actions may have on humans AND possibilities for humans to come to harm
that it might acvoid by its actions. And it must place that above all
Second Law command. It must not assume that humans giving it orders
have taken into account the possible harm to humans resulting from
those orders.

Obviously, a robot can harm humans if it fails to understand that its
action or inaction causes harm to humans. Therefore, a robot must have
best possible way of understanding the relevance of all the information
it has so as to decide which actions lead to harm to humans.

Also, a robot can harm humans by following Second Law orders if it
lacks input information. For example, if a robot is told to drive a car
along a certain road in a hurry, and assured that the road will be
empty of all other traffic, it would assume this does not violate the
First Law. But in fact, the human giving those orders might have
specific knowledge that there would be a human there, and that if the
robot hurries as ordered, it will not have time to prevent killing the
human. Or he might have forgotten the fact even though he should have
remembered.

If a robot has the ability to evaluate the truthfulness and relevance
of inormation for First Law purposes, why not in order to tell truth to
humans?

And of course, a robot might have to delete potentially vital
information just to manage its available memory. It could notice that a
certain road is weakened and would collapse under a vehicle, killing
humans. Et cetera. But if it does not anticipate returning to the road
to carry humans there or being asked about the road, it eventually
would have to forget about it. Again, a robot has to evaluate what it
chooses to forget.

So. Can a robot deliberately make itself believe that a certain action
does not ham humans if it simultaneously has reached the conclusion
that it will?
 >> Stay informed about: Should robots lie? 

"the Swampster" <spamtrap RemoveThis @balglaas.dds.nl> wrote in message
news:43212d61$0$2428$a473aa58@reader01...
> "Jens E. Nyborg" <spamtrap RemoveThis @aabent.dk> writes:
>
> [Schnipp]
>
>>4th. I don't recall. Maybe there was no 4th?
>
> This good be a good base for a story.
>

Wasn't it something about OCP? Ah... no, wait..
 >> Stay informed about: Should robots lie? 

IsaacKuo wrote:

> chornedsnorkack.TakeThisOut@hushmail.com wrote:
> >Should robots be allowed to lie?
>
> >The 3 laws were invented by humans to keep robots safe and to alleviate
> >their Frankenstein complex.
>
> The 3 laws of robotics were invented by Isaac Asimov so he
> could write interesting stories around them. They have not,
> AFAIK, ever been implemented in any meaningful way in any
> actual robot.
>
> Indeed, a huge class of robots are specifically designed to
> violate the first and third laws--some of them on a truly
> apocalyptic scale.

Robo Soldiers!!!!
 >> Stay informed about: Should robots lie?