In this
posting I want to examine roboethics and responsibility. I will do this by
considering autonomy. The idea of roboethics seems to have originated with
Asimov’s three laws of robotics which he proposed in 1942. In this posting I
will consider a revised draft of these laws based
on an EPSRC/AHRC initiative in roboethics, see Alan Winfield . These
revised Laws are,
1.
Robots are multi-use tools. Robots should not be
designed solely or primarily to kill or harm humans, except in the interests of
national security.
2.
Humans, not robots, are responsible agents. Robots
should be designed & operated as far as is practicable to comply with
existing laws and fundamental rights & freedoms, including privacy.
3.
Robots are products. They should be designed using
processes which assure their safety and security.
4.
Robots are manufactured artefacts. They should not
be designed in a deceptive way to exploit vulnerable users; instead their
machine nature should be transparent.
5.
The person with legal responsibility for a robot
should be attributed, (with
responsibility?)
I agree
that humans rather than robots are responsible agents. I also believe robots
such as those working on car production lines are just machines like washing
machines. However I am going to argue that advances in robotics are likely to smear
the attribution of responsibility for events that happen due to a robot’s
actions.
Let it
be assumed all possible robots are used in a similar way to those working on
car production lines. If a machine cause’s damage due to a design fault then
the manufacturer is at fault, for instance Toyota Toyota
However I believe my assumption is false. Robots are indeed
instruments we use to obtain our goals as is any machine. But the operations of
a robot need not be the same as any other machine. If I use a knife or washing
machine the operation of knife or washing machine is completely determined by
me. The goals of a robot are completely determined but the way it operates
need not be. For instance a driverless train may be classed as a robot if it
has no human on board even to act as a controller in an emergency such as those
on the Copenhagen Metro. A computer program that automatically trades on the
stock market might also be classed as a (ro)bot provided we think the terms
instrument and machine are interchangeable in some contexts. If it is accepted
that (ro)bots can operate in ways that are not completely determined by the
designer or operator then there is indeed a need for roboethics.
Let us
consider a (ro)bot that independently trades in stocks for a firm of
stockbrokers. Let it be assumed for the sake of simplicity that the goal set by
the roboticist who designed this machine is only to maximise profits for the
stockholders. Let it be further assumed this machine buys stocks in one company
for only some of its stockholders inflating the share price. It then sells the
shares other stockholders hold in this company at this inflated price. This
sale causes the price to return to its former level. Let it be assumed such a
strategy maximises profits for stockholders in general, but that some
stockholders gain at the expense of others. The actions the machine takes are
unethical. Clearly this machine contravenes the 4th law of robotics
as set out above. This example raises two questions. Firstly what can be done
to prevent the contravention occurring again and who is responsible for this.
It seems to
me there are two possible ways of ensuring the actions a machine takes are
ethical. Firstly the goals of the machine must be adequately specified in order
to prevent it producing unethical actions. Secondly if the machine can operate
in a way not completely determined by the designer
or operator then the machine must have some inbuilt standards against which it
can check the morality of any its proposed actions. Let us assume goals
of a machine can adequately specified so as to prevent it producing immoral
actions. Let us consider this assumption in conjunction with my example of a
stock trading machine. The 2nd law of robotics states that
humans rather than robots are responsible agents. In my example I would worry
that even though humans are responsible for the machines goals it is by no
means clear who should be held accountable for any failings, responsibility
becomes smeared. In my example the machine’s designer is unlikely to be fully
aware of all the possibilities of stock trading whilst the machine’s owners may
be unaware of how to build goals into a machine. Someone might object my worry
is illusory and that the stockbrokers involved must be able to fully set out
the machine’s goals to the designer. However I have other important worries.
Firstly I worry whether, as the tasks machines become able to do become more
complex, it is possible to completely specify the goals of a machine so none of
its actions can be considered as immoral. Secondly I worry that the ends a
designer sets a machine means the means it takes to achieve these ends is
always justified. Because of these worries I would suggest instead of
attempting to prevent a machine producing immoral actions by adequately framing
the machine’s goals that it would be preferable to have a
machine with some inbuilt standards against which it can check the morality of
doing any its proposed actions.
Someone might object if my suggestion is adopted that the 2nd
law of robotics is contravened. She might point out I am proposing giving
autonomy, albeit limited, to robots and with that autonomy comes
responsibility. I will now argue my suggestion means giving some form of self-government to robots but it does
not mean giving them even limited
autonomy. Self-government in this context simply means having the capacity
to independently decide on and follow some course of action. A child can be self-governing
in a domain specified by her parents. A robot can also be self governing in
some limited domain specified by its operators, for example a planetary rover.
However self-government as defined above does not mean being autonomous. I have
argued repeatedly in this blog autonomy
requires that an agent must ‘care about’ or love something. It seems
extremely doubtful to me whether robots will ever ‘care about’ or love
anything. It is certain present robots day cannot care. It follows it must be
doubtful whether robots can be autonomous. It further follows even if a robot
has some limited self government it cannot be held responsible for its actions
if responsibility requires autonomy. It still further follows that my
suggestion, that (ro)bots have inbuilt standards against which they
can check the morality of doing any some proposed actions, does not contravene
the 2nd law of robotics.
Nevertheless my suggestion does have worrying implications
for responsibility. Only humans can be held responsible for the actions of
(ro)bots but adopting my suggestion would mean this responsibility becomes
smeared. At the present the Military is showing great interest in military
robots, see (1). Let it be assumed such a robot kills a large number of
non-combatants in error. Let it be further assumed this machine had inbuilt
into its software a set of moral standards. The outcry would be enormous and
there would be demands that someone should be held responsible. Clearly as I
have argued the robot cannot be held responsible. Who then should be
responsible, the operator, the software designer or the ethicist who drew up
the moral standards? I would argue no one is clearly responsible as responsibility
has become smeared between a large numbers of people due to the way the machine
operates. Because of this smearing of responsibility it might be argued no
(ro)bots should ever be made which operate with an inbuilt set of moral
standards. However I would argue this smearing of responsibility is not a
problem peculiar to robotics but is an inevitable consequence of the complexity
of modern life. Let us assume a pilot in a plane over Libya
- Wendell
Wallach, Colin Allen, 2009, Moral
Machines, Oxford University Press, page 20.
