aises_9_3

<h1 id="moral-considerations">A.3 Moral Considerations</h1>
<p>How can we determine whether an action is right or wrong? What are
the kinds of principles and values that should guide our moral
decisions? There are many factors to consider. Here, we’ll focus on a
few—-goodness, constraints, special obligations, and options-—that very
commonly enter into moral decision making.</p>
<h2 id="the-goodness-of-actions-and-their-consequences">A.3.1 The “Goodness”
of Actions and Their Consequences</h2>
<p>Moral decision making often involves considering the values, or
“goods,” that are at stake. These may be intrinsic goods or instrumental
goods.</p>
<p><strong>Intrinsic goods are things that are valuable for their own
sake.</strong> Philosophers disagree about what, if anything, is
intrinsically good, but many argue for the intrinsic value of things
like happiness, love, and knowledge. We value such things simply because
they are valuable—not because they necessarily lead to anything
else.</p>
<p><strong>Instrumental goods are things that are valuable because of
the benefits they provide or the outcomes they achieve.</strong> We
pursue instrumental goods as a means to an end, but not for their own
sake. Money, power, and education are examples of instrumental goods. We
value them because they can lead to other things we value, like
security, influence, career opportunities, or intrinsic goods.</p>
<p><strong>Intrinsically good things are not necessarily instrumentally
good.</strong> Sometimes, intrinsically bad things can be instrumentally
good and intrinsic goods can be instrumentally bad. For instance, many
people believe that honesty is intrinsically good. However, it’s easy to
imagine cases in which honesty can lead to bad outcomes, like hurt
feelings. Suppose a friend has confided in you that they are staying at
a shelter to hide from an abusive partner. If that abusive partner asks
you for your friend’s location, you may think that that honesty is
intrinsically good. However, revealing your friend’s location would be
instrumentally bad, as it may lead to further violence and perhaps even
a risk to your friend’s life. On the other hand, consider medical
treatments like chemotherapy. Chemotherapy is instrumentally good
because it can prolong cancer patients’ lives. Yet, as it requires the
administration of highly toxic drugs into a patient’s body, it could be
seen as harmful, or intrinsically bad. For many people, exercise is
painful, and pain is intrinsically bad, but exercise can be
instrumentally good.</p>
<p><strong>There is no consensus about what is intrinsically
good.</strong> Some philosophers believe that there are many intrinsic
goods. Others believe there is only one value. One common view is that
the only intrinsic good is wellbeing, and everything else is valuable
only insofar as it promotes wellbeing.<p>
Value pluralists believe that there are many intrinsic goods. These
values may include justice, rights, autonomy, and virtues such as
courage. Other philosophers believe there is only one fundamental value.
Among these, one common view is that the only intrinsic good is
wellbeing, and everything else is valuable only insofar as it promotes
wellbeing.</p>
<h3 id="sec:wellbeing">Wellbeing</h3>
<p><strong>Wellbeing is how well a person’s life is going for
them.</strong> It is commonly considered to be intrinsically good,
though there are different accounts of precisely what wellbeing is and
how we can evaluate it. Generally, a person’s wellbeing seems to depend
on the extent to which that person is happy, healthy, and fulfilled.
Three common accounts of wellbeing characterize it as 1) net pleasure
over pain, 2) preference satisfaction, or 3) a collection of objective
goods. Each account is elaborated below.</p>
<p>Some philosophers, known as <em>hedonists</em>, argue that wellbeing
is the achievement of the greatest balance of pleasure and happiness
over pain and suffering. (For simplicity we do not distinguish, in this
chapter, between “pleasure” and “happiness” or between “pain” and
“suffering,” though neither pair is interchangeable.) All else equal,
individuals who experience more pleasure have higher wellbeing. All else
equal, individuals who experience more pain have lower wellbeing.</p>
<p><strong>According to hedonism, pleasure is the only intrinsic
good.</strong> Goods like health, knowledge, and love are instrumentally
valuable. That is, they are only good insofar as they lead to pleasure.
It may feel as though other activities are intrinsically valuable. For
instance, someone who loves literature may feel that studying classic
works is valuable for its own sake. Yet, if the literature lover were
confronted with proof that reading the classics makes them less happy
than they otherwise would be, they might no longer value studying
literature. Hedonists believe that when we think we value certain
activities, we actually value the pleasure they bring us, not the
activities themselves.<p>
Hedonism is a relatively clear and intuitive account of wellbeing. It
seems to apply equally to everyone. That is, while we all may have
different preferences and desires, pleasure seems to be universally
valued. However, some philosophers argue that hedonism is an incomplete
account of wellbeing. They argue there may be other factors that
influence wellbeing, such as the pursuit of knowledge.</p>
<p>Some philosophers claim that what really matters for wellbeing is
that our preferences are satisfied, even if satisfying preferences does
not always lead to pleasurable experiences.<p>
One difficulty for preference-based theories is that there are different
kinds of preferences, and it’s unclear which ones matter. Preferences
can be split into three categories: stated preferences, revealed
preferences, and idealized preferences. Each of these categories can be
informative in different contexts.<p>
To illustrate different kinds of preferences, consider voter preferences
in a democratic election.<p>
In a democratic election, citizens choose which candidate they want to
elect by casting their vote on a ballot. Their choice to vote for a
given candidate can be impacted by a number of different factors.
Perhaps they have an existing political affiliation, are influenced by
social pressures, believe in the candidate’s policies, or maybe they
just like one candidate’s demeanor and personality. Importantly,
citizens may not always vote for the candidate they outwardly support,
and the choice to vote for a specific candidate can change when voters
discover new information.<p>
A voter’s <em>stated preference</em> is the candidate that they state
they support. Voters may express their stated preferences in
conversations, polls, and while campaigning.<p>
When a voter casts their vote on a ballot, they express their
<em>revealed preference</em>. Generally, a voter’s revealed preferences
align with their interests. For example, a voter who supports increased
funding for education might vote for a candidate who wants to increase
budgets for local public schools. A revealed preference is expressed by
your actions, not your words.<p>
People change their preferences upon learning new information.
Uninformed preferences can be reached quickly. For example, a voter
might have an uninformed preference based on a “gut reaction” to a
candidate. Voters can arrive at more <em>idealized preferences</em> once
they have gathered and evaluated all relevant information. They might
not actually do this—few people have the time or ability to perfectly
gather and evaluate all the relevant information that they would require
to find their idealized preferences. However, their preferences can
become more idealized over time. A voter might have an uninformed
preference for Candidate A and, after learning new information about
each candidate’s platform, they may arrive at a more informed preference
for a different candidate. In other words, preferences can change, and
they often do change as people become more informed.</p>
<p><strong>It is easy to learn about people’s stated preferences–—simply
ask them.</strong> Political polls and surveys, for example, are an easy
way to gather information about people’s stated preferences. However,
stated preferences may not always predict what people will actually
choose. A voter may outwardly express support for Candidate X, but when
it comes to casting their ballot, they may vote for Candidate Y.
Similarly, someone might express a stated preference to eat healthier,
but that doesn’t necessarily mean that they will. Their behavior (such
as eating only chocolate for a week) may indicate that their revealed
preference is for unhealthy food.<p>
Revealed preferences can be harder to observe, but they are generally
more useful for predicting people’s behavior. Someone with a stated
preference for vegetables but a revealed preference for chocolate is
more likely to purchase and consume chocolate than vegetables. When
researching consumer behavior, economists often prefer to study
consumers’ revealed preferences (i.e. what they buy) rather than stated
preferences (i.e. what they say they’d like to buy).</p>
<p>Others believe that wellbeing is the achievement of an objective set
of “goods” or “values.” These goods are considered necessary for living
a good life regardless of a person’s individual experiences or
preferences. There is disagreement about which particular goods are
necessary for wellbeing. Commonly proposed goods include happiness,
health, relationships, knowledge, and more. Objective goods theorists
consider these values to be important for wellbeing independently of
individual beliefs and preferences.<p>
There is no uncontroversial way to determine which goods are important
for living a good life. However, this uncertainty is not a unique
problem for objective goods theory. It can be difficult for hedonists to
explain why happiness is the only value that’s important for wellbeing
and for preference satisfaction theorists to determine which preferences
matter most.<p>
While people disagree about which account of wellbeing is correct, most
people agree that wellbeing is an important moral consideration. All
else equal, taking actions that promote wellbeing is generally
considered morally superior to taking actions that reduce
wellbeing.<p>
In the future, it is conceivable that AIs might be conscious and have
preferences but not experience pleasure, which would mean they could
have wellbeing according to the preference satisfaction theorists but
not hedonists. It is also possible that in the future AIs may have
wellbeing according to all three accounts of wellbeing. This would
require that we dramatically reassess our relationship with AIs.</p>
<h2 id="constraints-and-special-obligations">A.3.2 Constraints and Special
Obligations</h2>
<p>We’ve covered the moral consideration of intrinsic goods, and focused
on the intrinsic good wellbeing. Special obligations and constraints are
key considerations when we make ethical decisions.</p>
<p><strong>Special obligations are duties arising from
relationships.</strong> We can incur special obligations when we promise
someone to do something, take a professional position with
responsibilities, have a child, make a romantic commitment to a partner,
and so on. Sometimes we can have special obligations that we did not
volunteer for—a child to its parents, or our duties to fellow
citizens.</p>
<p><strong>Constraints are actions that we are morally prohibited from
taking.</strong> A constraint is something that places limits on our
actions. For example, many people think we’re morally prohibited from
lying, stealing, cheating, harming others, and more.</p>
<p><strong>Constraints often come in the form of rights.</strong> Rights
are claims that individuals may have over their community. For instance,
many people believe that humans have the rights to life, freedom,
privacy, and so on. Some people argue that any individual with the
capacity for experiencing pleasure and pain has rights. Non-human
individuals (including animals and AI systems) might also have certain
rights.<p>
An individual’s rights may require that society intervene in certain
ways to ensure that those rights are fulfilled. For instance, an
individual’s right to food, shelter, or education may require the rest
of society to pay taxes so that the government can ensure that
everyone’s rights are fulfilled. Rights that require certain actions
from others are called positive rights.<p>
Other rights may require that society abstain from certain actions. For
instance, an individual’s right to free speech, privacy, or freedom from
discrimination may require the rest of society to refrain from
censorship, spying, and discriminating. Rights that require others to
abstain from certain behaviors are called negative rights.<p>
Many AI researchers think that, for now, we should avoid accidentally
creating AIs that deserve rights <span class="citation"
data-cites="sebo2022chatbot">[1]</span>; for instance, perhaps all
entities that can experience suffering have natural rights to protect
them from it. Some think we should especially avoid giving them positive
rights; it might be fine to give them rights against being tortured but
not the right to vote. If they come to deserve rights, this would create
many complications and undermine our claim to control.</p>
<h2 id="what-does-it-mean-for-an-action-to-be-right-or-wrong">A.3.3 What does
it mean for an action to be right or wrong?</h2>
<p>Some of the first questions we might ask about ethics are: Are all
actions either right or wrong? Are some simply neutral? Are there other
distinctions we might want to draw between the morality of different
actions?<p>
The answers to these questions, like most moral questions, are the
subject of much debate. Here, we will simply examine what it might mean
for an action to be right or wrong. We will also draw some other useful
distinctions, like the distinction between obligatory and non-obligatory
actions, and between permissible and impermissible actions. These
distinctions will be useful in the following section, when we discuss
the considerations that inform our moral judgments.</p>
<h3 id="options">Options</h3>
<p>Special obligations and constraints tell us what we should not do,
and sometimes, what we must do. Intrinsic goods tell us about things
that would be good, should they happen. But philosophers debate how much
good we are required to do.</p>
<p><strong>Options are moral actions which we are neither required to do
nor forbidden from doing.</strong> Even though it would be good to
donate money, many people do not think people are morally required to
donate. This is an ethical option. If we believe in options, not all
actions are either required or forbidden.<p>
We now break down actions onto a spectrum on which we will simply
examine what it might mean for an action to be right or wrong. We will
also draw some other useful distinctions, like the distinction between
obligatory and non-obligatory actions and between permissible and
impermissible actions.</p>
<p><strong>Obligatory actions are those that we are morally obligated or
required to perform.</strong> We have a moral duty or obligation to
carry out obligatory actions, based on ethical principles. For example,
it is generally considered obligatory to help someone in distress, or
refrain from hurting others.</p>
<p><strong>Non-obligatory actions are actions that are not morally
required or necessary.</strong> Non-obligatory actions may still be
morally good, but they are not considered to be obligatory. For example,
volunteering at a charity organization or donating to a good cause may
be good, but most people don’t consider them to be obligatory.</p>
<p><strong>Permissible actions may be morally good or simply neutral
(i.e. not good or bad).</strong> In general, any action that is not
impermissible is permissible. Moral obligations, of course, are
permissible. We can consider four other actions: volunteering, donating
to charity, eating a sandwich, and taking a walk. These seem
permissible, and can be classified into two categories.<p>
One class of permissible actions is called <em>supererogatory
actions</em>. These may include volunteering or giving to charity. They
are generally considered good; in fact, we tend to believe that the
people who do them deserve praise. On the other hand, we typically don’t
consider the failure to do these actions to be bad. We might think of
supererogatory actions as those that are morally good, but optional;
they go “above and beyond” what is morally required.<p>
Another class of permissible actions is called <em>morally neutral
actions</em>. These may include seemingly inconsequential activities
like eating a sandwich or taking a walk. Most people probably believe
that actions like these are neither right nor wrong.</p>
<p><strong>Impermissible actions are those that are morally prohibited
or unacceptable.</strong> These actions violate moral laws or principles
and are considered wrong. Stealing or attacking someone are generally
considered to be impermissible actions.<p>
</p>
<figure id="fig:action-types">
<img src="https://raw.githubusercontent.com/WilliamHodgkins/AISES/main/images/action_types.png" class="tb-img-full"/>
<p class="tb-caption">Classes of permissible and non-obligatory actions</p>
<!--<figcaption>Classes of permissible and non-obligatory actions-->
<!--</figcaption>-->
</figure>
<p>Some philosophers believe that all actions fit on a scale like the
one above. At one end of the scale are impermissible actions, like
murder, theft, or exploitation. At the other end are obligatory actions,
like honesty, respect, and not harming others. In between are neutral
and supererogatory actions. These are neither impermissible nor
obligatory. Many people believe that the vast majority of our actions
fall into these two categories. Crucially, in designing ethical AI
systems that operate in the real world, it is important to determine
which actions are obligatory and which actions are impermissible.<p>
However, some philosophers do not believe in options; rather that
actions are all on a spectrum from the least moral to the most moral. We
will learn more about these positions, and others, when we discuss moral
theories later in this chapter.</p>
<h3 id="from-considerations-to-theories">From Considerations to
Theories</h3>
<p><strong>Moral considerations can guide our day-to-day decision
making.</strong> Understanding which factors are morally relevant can
help us think more clearly about what we should do. Of course, we don’t
always stop to consider every factor before making a decision. Rather,
we tend to draw broader conclusions or moral principles based on our
evaluations of specific cases. For instance, once we consider a few
examples of the ways in which stealing can harm others, we might draw
the conclusion that we shouldn’t steal.<p>
The considerations discussed in this section provide a basis on which we
can develop more practical, action-guiding theories about how we should
behave. The types of fundamental considerations in this section comprise
a subfield of ethics called <em>metaethics</em>. Metaethics is the
consideration of questions like “What makes an action right or wrong?”
and “What does it mean to say that an action is right or wrong?” <span
class="citation" data-cites="fisher2014metaethics">[2]</span><p>
These considerations are important in the context of designing AI
systems. In order to respond to situations in an appropriate way, AI
systems need to be able to identify morally relevant features and detect
situations where certain moral principles apply. They would also need to
be able to evaluate and compare the moral worth of potential actions,
taking into account various purported intrinsic goods as well as
normative factors such as special obligations and constraints. The
challenges of designing objectives for AI systems that respect moral
principles are further discussed in the Machine Ethics chapter.<p>
In the following section, we will discuss some popular moral
theories.</p>

<br>
<br>
<h3>References</h3>
<div id="refs" class="references csl-bib-body" data-entry-spacing="0"
role="list">
<div id="ref-sebo2022chatbot" class="csl-entry" role="listitem">
<div class="csl-left-margin">[1] J.
Sebo, <span>“Op-ed: What should we do if a chatbot has thoughts and
feelings?”</span> [Online]. Available: <a
href="https://www.latimes.com/opinion/story/2022-06-16/artificial-intelligence-morals-ethics-sentience-thinking">https://www.latimes.com/opinion/story/2022-06-16/artificial-intelligence-morals-ethics-sentience-thinking</a></div>
</div>
<div id="ref-fisher2014metaethics" class="csl-entry" role="listitem">
<div class="csl-left-margin">[2] A.
Fisher, <em>Metaethics: An introduction</em>. Routledge, 2011.</div>
</div>
</div>