AW: [aiethics] Thought experiment: Punishment of robots...

[Christian DroЯmann]

Alicebot AI Ethics Committee - http://www.alicebot.org


Chris wrote:

> > How should a robot that commits murder be punished then?
>
>   IMHO, your question *ought to be* considered absolutely absurd. The =
only
> thing that makes it not completely ridiculous is that we humans still
> subscribe, in general, to the idea that punishment (particularly the =
death
> penalty) is a good way to control behavior. The discussion of bot =
criminal
> justice hits at the core of our own criminal justics beliefs.

Yes...but experience shows that death penalty does not reduce the number =
of
crimes committed...

>   First, you're beginning with the assumption that a bot is a
> self-determined sentient being like we are, which of course it is
> not (yet).

That's why I called it a thought experiment :-)

> If Alice killed somebody, Dr. Wallace would face murder charges,
> not Alice.

I wouldn't say he'd face murder charges, but maybe charges for =
manslaughter
(I don't know whether this term exists in American English as well or if
that is called "second degree murder"...what I'm thinking of is =
accidentaly
killing somebody...may it be due to error or carelessness...does anyone =
of
the German speaking people here know a good translation for =
"fahrlaessige
Toetung" ?)

> But in the future, the degree of human responsibility for bot actions =
is
> bound to change, so let's discuss that:
>
>   Every weakness in criminal justice systems for human beings can
> be easily
> defeated by a bot:
>     1 - Fear of punishment is rarely on the mind of a hardened human
> criminal or a desperate person, anyway. This fear can be erased in a =
bot.

at least until bots do not only develop a conscience, but also
emotions...but maybe if death penalty would be applicable to robots as =
well
their desire to avoid their own destruction and thus avoiding a crime =
that
would lead to a death sentence might be interpreted as "fear of =
punishment"
And I think that death penalty's aim to scare off potential murderers =
would
be most effective with robots, as due to absence of emotions that could =
lead
to ignoring the fact that they will face death penalty if they kill =
someone,
like it sometimes happens with humans, they would simply consider =
killing as
illogical...not because they think that killing is wrong, but simply =
because
it would mean their own desctruction...UNLESS the robot finds a way to
guarantee he will not get caught...

>     2 - Prison sentences and forced labor are meant to force a human =
think
> about their crimes. Bots don't need to think about anything for
> any duration
> they want.

Doing something that will lead to imprisonment should seem illogical to =
a
robot as well for the reasons I mentioned above...

>     3 - Humans experience pain of punishment differently. Bots
> don't need to
> experience it at all.

The aim of punishment (at least in great parts of the western world) is =
not
to cause pain...look at the discussions about the electric chair as a =
means
of "cruel and unusual punishment"

>     4 - The "victims rights" movement is often dissatisfied with even =
the
> harshest sentences - they will hardly get their 'revenge'
> satisfaction on a
> bot.
>     etc...

at least the victim (or, in case of murder, his/her relatives) can be =
sure
the robot won't do it again...

>  In all, the argument of 'punishment versus rehabilitation' for humans =
is
> given another dimension when you discuss bots. Ultimately, much of the =
way
> criminal justice is perceived is based on a human belief that our
> brains are
> NOT just machines, that there is a "soul" of some sort, evil or good, =
that
> controls our behavior. In this belief system, rehabilitation of
> the brain is
> besides the point - revenge and penitence are spiritual
> activities that act
> upon the souls of criminals and victims alike, and punishment is the =
sole
> tool of that action.
>
>  It's obvious that a mass-murdering homicidal bot could be easily
> reprogrammed to become a brain surgeon or a babysitter, but it's not =
so
> obvious (or even remotely ethical!) to reprogram humans! A prison =
sentence
> or a public whipping are examples of the kind of blunt
> instruments we use to
> induce self-reprogramming in humans. In some (probably rare)
> cases this kind
> of behavioral modification works I suppose, but it's clearly a
> crude method.

I agree with you here...killing bots would simply considered =
malfunctioning
machines that need to be repaired...

>  Now, the death penalty: Executing a human works to the extent that it
> really does totally guarantee that he/she won't commit a crime ever =
again
> But such elimination is impossible with a bot: You can delete the
> HD, smash
> the exoskeleton, or beam the mainframe into the sun - the bot could =
still
> back itself up thousands of times across a network, even without =
wires.
> Elimination of a bot is going to be a problem, but it's clear that =
even
> nailing down a bot to reprogram it will be a problem. Clearly our
> traditional way of holding a human "responsible" for a "crime" will =
not be
> adequate.

there would be the need for a function that prevents the bot from =
backing
itself up to just anywhere...
it should be made technically impossible for the bot to backup himself =
to a
place other than a certain preset place (whateve that maybe...maybe a
"federal center for robot backups"...

> Here are some other kinds of malicious bots in the future:
> 1) Viruses - From scripts that clog your email program, to programs =
that
> embezzle money, to massive trojan horses that cut off life support in
> hospitals or disable a nation's military infrastructure, viruses
> will always
> be with us and will usually be the responsibility of their creators =
(human
> or not!).

that is true because those programs are _intentionally_ evil...

> 2) Social Manipulators - Bots that slander or libel people to
> enable its own
> social ambitions, whose crimes are neither physical nor financial, but
> rather sociological. Imagine a Joe McCarthy robot.

again, if the bot was _purposely_ programmed to behave this way, then =
the
creator should be held responsible...

> 3) The Borg - what happens when a group of bots have some degree of
> individuality, but also have a collective identity, or even a =
heirarchy of
> collective identities? A decision to commit a crime may be made by an
> individual, a subgroup, or by the whole group. Which entitity
> would you hold
> responsible for a single criminal act? Conspiracy, racketeering, and =
war
> crime laws address such issues for humans, but will they work for
> bots? For
> example, Nazi war criminals were held responsible for their crimes as
> individuals and for the actions of their subordinates. But in a =
machine
> collective, bots might be held responsible for actions of their =
'innocent'
> peers if their decisions were made as a group (which can, of course, =
never
> be proven of human war criminals).

That is indeed a rather  difficult problem...
I think the only way one could judge this is to determine whether the
individual COULD have behaved in a different way...

Christian