Comput Game J https://doi.org/10.1007/s40869-017-0049-0
Decision Theory Based Model of Game Morality Tony Rosqvist1
Received: 20 October 2017 / Accepted: 11 December 2017 Springer Science+Business Media, LLC, part of Springer Nature 2017
Abstract A novel method for assessing moral values in computer games is presented in this paper. The method is based on Decision Analysis. It is shown that the decision model presented conforms with the ethical principle of Rule Consequentialism. The model, termed Scale of Morality, can be implemented in strategy and role-playing computer games where, to date, the ethical valuation has been assessed mainly using informal heuristics. The Scale of Morality requires the player (human player, computer agent or game master) to pre-assign value scores to selected ethical criteria prior to gaming. During the game, players make moral choices by valuating each choice with respect to these criteria. The choice corresponding to the highest value score is the most ethical one. Even in the case where the ethical principle is the same for each player during the game, the ethical valuations may differ, as the information process can be asymmetric (e.g. one player acquires more information than another). The game dynamics will be much more complex if the ethical principle of each player is different. Furthermore, the computer agent can be equipped with an expanding information base, making the agent better in judging the consequences of choices. The method Scale of Morality can be a component of the game’s Artificial Intelligence, or serve as a research tool for studying players’ moral reasoning. Keywords Game morality Computer agent Decision analysis Moral choice Moral dilemma
& Tony Rosqvist
[email protected] 1
Technical Research Centre of Finland Ltd., Kemistintie 3, 02150 Espoo, Finland
123
Comput Game J
1 Introduction The theory of deontology states that we are morally obligated to act in accordance with certain rules (duties) in order to achieve goodness. Utilitarianism, or consequentialism, says we have an obligation to take the course of action that achieves the most positive consequence in terms of the well-being of individuals. These ethical principles are oftentimes consistent, but sometimes conflicting, e.g. you may have to steal food to feed your child (Hooker 1990). An ethical principle, which integrates both the deontological and consequentialist perspectives, is the Rule Consequentialism, a definition of which is provided by Hooker (2000): ‘An action is morally right if and only if it does not violate the set of rules of behaviour whose general acceptance in the community would have the best consequences—that is, at least as good as any rival set of rules or no rules at all.’ More pragmatically, we may state that a ‘good’ human being, as well as, a game character, is someone who follows well-established duties and assures that his/her choice will lead to the highest possible well-being (Harris 2010) of all individuals concerned. Hence the moral valuation in a choice situation requires one to take into account both duties and consequences in order to determine the choice of highest ethical value. Computer games can potentially have a great impact on the player’s learning ethics and could lead to an improving in one’s moral judgement. However, Heron and Belford (2014) point out that this potential has not yet been fully realised. They present two key factors that determine whether a game can serve to educate and improve players’ understanding of the moral aspects of decisions, or is purely entertainment: (1) the extent to which the player can characterize the avatar to ‘be me’; and, (2) the level of realism in the story, i.e. ‘what could happen to me’. This paper introduces a multi-attribute analytic method for making decisions (Keeney and Raiffa 1993) that conforms with the Rule Consequentialism ethics and which provides a means of ‘measuring’ the moral value of a player’s or computer agent’s choice. Specifications, reflecting the gamers’ ethical principles, need to be made prior to commencing the game. During the game, moral dilemmas are presented and choices must be made. The ethical value of a choice depends on the outcomes of the choice with respect to the specified ethical criteria and their values. Each choice is mapped to an ethical value (= score). This value is a measure of morality. The method, termed the ‘Scale of Morality’, provides an answer to the request by Koop (2013) for the need to reconceptualise moral dilemmas within the framework of decision-making under multiple criteria such that deontological and utilitarian considerations are made simultaneously, rationally and systematically. This would support learning of ethics, e.g. in the form of computer games, in a new way. Harris (2010) mentions three domains of enquiry for understanding morality: (1) descriptive research on behavioural tendencies of individuals; (2) normative research on what should be the ‘moral truth’ to be followed; and, (3) policy and public outreach for raising ethical awareness. The contribution of this paper lies in the last domain in that it provides a scientific method for constructing morality in
123
Comput Game J
games with education as one driver of game development. The presentation of the Scale of Morality method includes code that can be included in web applications and computer games.
2 The Scale of Morality Method In this paper, value is defined as the worth of a thing (a rule, an act), and valuation is an estimate of this worth. The numerical estimate is also referred to as a value. Hence, the term value is used both qualitatively and quantitatively. The term morality refers to an individual’s personal sense of right and wrong (Heron and Belford 2014). A moral dilemma is a choice situation that requires reasoning about the right course of action—an action with the highest ethical (or moral) value. 2.1 Modelling of the Choice Problem: The Value Model The Scale of Morality formulates a moral dilemma as a Hobson’s choice; a free choice in which two options are possible at a time: you may proceed, or not proceed (i.e. Yes/No). Either way, your choice will have an outcome regarding duties followed and changes in the well-being of affected individuals. The Scale of Morality deduces the ethical value of one’s choice based on these outcomes, with each outcome linked to a value score pre-assigned to the ethical criteria. In a ruleconsequentialist setting, the ethical criteria are duties (rules) to be followed, as well as, consequence-criteria depicting changes in the well-being of those concerned in a moral dilemma. The mapping of the choice to a real number requires a value model, which connects the choices to ethical criteria for criterion-specific and aggregated valuation (Keeney 1996) (Fig. 1). The outcomes of a choice regarding all the criteria will be numbers that are added together to yield the overall ethical value of the choice. This ethical value orders the choices in preferential order; a choice with a higher value is preferable to that with a lower value. (Note. Further comparisons can be made, where the ‘winner’ of the pairwise comparison is compared with a third choice, and so on. Typically, moral choices are formulated as binary Hobbesian choices, however.) 2.2 Measuring the Value of a Choice: The Value Function In Decision Analysis, value measurement is performed using a value function (Keeney and Raiffa 1993), which maps the outcomes of a decision using a numerical scale. The graph in Fig. 2 shows the mapping of a moral choice to produce an ethical value based on given ethical criteria and their worth: 2.2.1 Defining Ethical Criteria As noted earlier, the Scale of Morality has been developed with the principle of the Rule Consequentialism in mind. For the sake of simplicity, we could represent the rules of a deontological ethic by the Ten Commandments from the Bible:
123
Comput Game J
3. Which choice has higher ethical value?
Ethical criteria N
Ethical criteria 2
Ethical criteria 1
Ethical value of choice
...
1. Which are the relevant ethical criteria (Rule Consequentialist setting)?
Hobbesian choice Yes
2. What are the outcomes of the choice ’Yes’/’No’ w.r.t to the criteria in this choice situation ?
No
Fig. 1 Value model connecting the ethical criteria and aggregating criterion-specific values to an overall value of the choices ‘Yes’ and ‘No’ in a moral choice situation
Defining ethical criteria
Defining scores representing worth of ethical criteria
Ethical principle level (independent of choice situation)
VALUE FUNCTION
Moral dilemma level (= moral choice situation)
Assessing outcome of choice w.r.t. ethical criteria
Ethical (moral) value of the choice
Fig. 2 The Scale of Morality method, which uses a value function for mapping a moral choice to produce an ethical value based on specific ethical criteria representing the ethical principles of a person, group or society
123
Comput Game J
1. You shall have no other Gods before me. 2. You shall not make for yourself an idol in the form of anything in heaven or on earth. 3. You shall not misuse the name of the Lord your God. 4. Remember the Sabbath day by keeping it holy. 5. Honour your father and mother. 6. You shall not murder. 7. You shall not commit adultery. 8. You shall not steal. 9. You shall not give false testimony against your neighbour. 10. You shall not covet your neighbour’s house, spouse or possessions. The six last ones reflect ‘the Golden Rule’, ‘Whatever you want men to do to you, do also to them’, which is also embraced by many people who are not confessional Christians. These six rules, which are listed as D1 through D6 in Table 1, are considered adequate for the deontological valuation of the moral dilemma addressed in the case example below. In addition to deontological ethical criteria, we specify consequentialist criteria, whereby the key concept is ‘well-being’, which in turn is affected by a person’s moral choice—both for oneself and others involved in a moral dilemma. Well-being can be modelled in terms of five consequence categories (C1–C5), as shown in Table 1—other consequence categories could also be used. In a game setting, the game master provides the definitions. Table 1 The Value Table: ethical criteria and their value scores (example) Ethical criteria
Value score
Deontological D1: Honour your father and mother.
10
D2: You shall not murder.
55
D3: You shall not commit adultery.
10
D4: You shall not steal.
10
D5: You shall not give false testimony against your neighbour.
10
D6: You shall not covet your neighbour’s house, spouse or possessions.
5
Consequentialist C1: Improved well-being for most of those involved C2: Improved well-being for only some of those involved
100 50
C3: No change in well-being in the context
30
C4: Decreased well-being for some of those involved
20
C5: Inflicting suffering and/or death to some in the context
0
123
Comput Game J
2.2.2 Defining the Worth of Ethical Criteria Each ethical criterion in Table 1 is assigned a value score, which represents its worth. An ethical criterion with a high value score has more ethical worth than a criterion with a low value score. These value scores can be interpreted as weights, and they are obtained either by direct rating or allocation of points (Doyle et al. 1997). In the point allocation technique, a budget of (typically) 100 points is allocated to the criteria: the more points a criterion receives, the greater its relative importance. The score 0 represents ‘no value at all’. The value assignments in the Value Table (Table 1) are interpreted as follows: The criterion D2, ‘you shall not murder’, has been assigned the highest ethical value (value = 55) and complying with all duties will give the value of 100 for the choice. A choice with consequence C1: ‘Improved well-being for most of those involved’, is valued the highest among the consequence criteria, with a value 100. On the contrary, the consequence C5: ‘Inflicting suffering and/or death to some in the context’ is assigned zero value. It should be noted that the consequence criteria are mutually exclusive. This is not the case with the deontological criteria as the best outcome of a choice is to comply with them all, yielding the maximum value of 100. Hence, the value of the best outcome with regard to the deontological and the consequence criteria is the same: 100 value scores. Similarly, the worst outcome is zero for both criteria sets; deontological and consequential. By weighting the best outcome on deontological criteria and the best outcome on consequence criteria (i.e. criterion C1), the best outcome of the combined ethical criteria {D1, …, D6} and {C1, …, C5} will also be 100, irrespective what the weights wD and wC are so long as their sum (wD ? wC) is equal to 1. It turns out that the only reasonable weighting is wD = 0.5 and wC = 0.5 for representing the principle of Rule Consequentialism. This weighting implies that the unit values (marginal values) of the deontologist and the consequentialist criteria are equal. In other words, losing 1 point of ‘deontologist’ value or 1 point of ‘consequentialist’ value yields exactly the same loss in the overall ethical value. This ensures that the valuations of deontological and consequential outcomes of a choice are commensurable, paving way for a consistent Rule Consequentialismbased valuation. This is summarized in Table 2, which shows the admissible weights for the ‘pure deontologist’, ‘pure consequentialist’ and the ‘rule consequentialist’ kinds of ethical principles. The game master is, again, the primary
Table 2 Admissible weights corresponding to the ethical principle Ethical principle
wD
wC
Pure deontologist
1
0
Pure consequentialist
0
1
Rule consequentialist
0.5
0.5
123
10 5
D5: You shall not give false testimony against your neighbour.
D6: You shall not covet your neighbour’s house, spouse or…
30 20 0
C4: Decreased well-being for some of those involved
C5: Inflicting suffering and/or death to some involved
50
C2: Improved well-being for only some of those involved
C3: No change in well-being in the context
100
C1: Improved well-being for most of those involved
Consequentialist
10 10
D4: You shall not steal.
D2: You shall not murder.
D3: You shall not commit adultery.
10 55
D1: Honour your father and mother.
Deontological
0
1
N/A
N/A
N/A
N/A
N/A
T
T
T
T
F
T
Yes
Choice:
Outcome
N/A
N/A
N/A
N/A
N/A
T
T
T
T
T
T
No
w
Ethical criteria
Score
‘Pure deontological’
Outcomes of choices w.r.t three ethical principles
1
0
w
T
F
F
F
F
N/A
N/A
N/A
N/A
N/A
N/A
Yes
Choice:
Outcome
F
T
F
F
F
N/A
N/A
N/A
N/A
N/A
N/A
No
‘Pure consequentialist’
Table 3 Outcomes (True or False) of the moral choice to kill (‘Yes’) or not to kill (‘No’) under three ethical principles
0.5
0.5
w
T
F
F
F
F
T
T
T
T
F
T
Yes
Choice:
Outcome
F
T
F
F
F
T
T
T
T
T
T
No
‘Rule consequentialist’
Comput Game J
123
Comput Game J
decision-maker regarding the value scores (Table 1) and the weighting parameters (Table 2). 2.2.3 Assessing the Outcome of Choice After specifying the ethical principles, the ethical value of the choices Yes/No—in a given a moral dilemma—can be measured. This entails that a player or computer agent makes a judgement of the outcomes of ‘Yes’ and ‘No’ in terms of truth values True (T) or False (F), with respect to each ethical criterion. An example showing how the ethical values of the choices ‘Yes’ and ‘No’ are measured is shown in Table 3 for a hypothetical choice situation of ‘killing a person in a vendetta’, whereby the valuation is performed from the three ethical perspectives referred to in Table 2: ‘pure deontological’, ‘pure consequentialist’, and ‘rule consequentialist’. From the entries of True (T) and False (F) in the table, we observe that only ‘rule consequentialists’ make a full judgement of the outcome of the choices. The entries labelled N/A indicate that the corresponding ethical criteria are not relevant under the given ethical principle. 2.2.4 Ethical Value of a Moral Choice Given the outcomes of a choice in terms of True (T) and False (F), as depicted by Table 3 above, we can deduce the ethical value of our choice by summing the criteria-based specific values for which we have judged the outcome to be ‘True’: •
• •
a pure deontologist would argue that the ethical values of the choices ‘Yes’/’No’ are 45 (= 10 ? 0 ? 10 ? 10 ? 10 ? 5) and 100, correspondingly. Hence, the ‘No’ choice is ethically preferred to the ‘Yes’ choice. a pure consequentialist would claim the ethical value of the choice ‘Yes’ to be 0, and the value of the choice ‘No’ to be 20. Hence, the ‘No’ choice is preferred. a rule consequentialist, addressing the same outcomes as the aforementioned purists, would see the ethical value of the choice ‘Yes’ to be (0.5*45 ? 0.5*0 = 22.5) and the value of choice ‘No’ to be (0.5*100 ? 0.5*20 = 60.0). These values are exactly in the middle of the values of the purists’ corresponding choices. This is expected due to equal weighting.
The valuation performed in the above is based on a measurable multi-attribute value function (Dyer and Sarin, 1979): V ðchoiceÞ ¼ wD
nD X i¼1
vD i ðchoiceÞ þ wC
nC X
vCi ðchoiceÞ
ð1Þ
i¼1
…whereby the valuations vi (choice) are obtained from the rows of the Value Table by the condition:
123
Comput Game J
vi ðchoiceÞ
¼
valueð; iÞ; 0;
if outcome of choice is 0 T 0 if outcome of choice is 0 F 0
ð2Þ
with index i running through the index sets ID = {1,…,nD} and IC = {1,…,NC}. In non-mathematical language, the value function computes a weighted sum of those criteria scores which are True for the choice. In this paper, we will have ID = {1,…,6} and IC = {1,…,5}, and the weights wD and wC as defined in Table 2 above. In fact, the value scale used in the Scale of Morality is a ratio scale and defines an absolute zero-level, or ‘0’ (Stevens, 1946), which is interpreted as total immorality (reached when no ethical rules are followed and/or the consequences are the worst). Furthermore, one can say that a choice ‘A’ is twice as ethical (moral) compared to ‘not-A’, if the ratio of the values of the choices ‘A’ and ‘not-A’ is 2.
3 Application: An Example Case The following example provides clarification of the properties of the Scale of Morality method. It is based on real events reported on the Internet. The setting of the story has, however, been modified into the game world. Game character Sally suffers from the rare musculoskeletal developmental disorder, Klippel-Feil Syndrome. Her health grows steadily worse. No other character can produce the right combination of drugs to relieve her pain. Now she is Table 4 The ethical valuation of the euthanasia case (scenario 1) Ethical principle Ethical criteria
Score
Deontological D1: Honour your father and mother
Weighting
0.5 10
Moral dilemma Outcome of choice Yes
No
T
T
D2: You shall not murder
55
F
T
D3: You shall not commit adultery
10
T
T
D4: You shall not steal
10
T
T
D5: You shall not give false testimony against your neighbour
10
T
T
5
T
T
100
T
F
50
F
F
D6: You shall not covet your neighbour’s house, spouse or possessions Consequentialist C1: Improved well-being for most of those involved C2: Improved well-being for only some of those involved
0.5
C3: No change in well-being in the context
30
F
F
C4: Decreased well-being for some of those involved
20
F
F
C5: Inflicting suffering and/or death to some involved
0
F
T
72.5
50.0
Ethical value
123
Comput Game J
fighting for the right to allow the hospital to increase her medication to induce a coma-like state and then have food and hydration withdrawn. The moral dilemma faced by the doctor is whether or not she/he should vastly increase the morphine dose to sedate Sally into a coma-like state—a dosage of the drug that could also kill Sally. Let us assume that one doctor obeys the rule not to kill (in observance of the professional Hippocratic oath or Christian creed). Let us now assume that the doctor’s ethical principle is represented by Table 4, which lists the 0.5-level weighting of the rules and consequences. A very plausible outcome of the moral dilemma is depicted by the truth values (T/F), each one relating to a criterion value score. The ethical value of the choice ‘Yes’ or ‘No’ is indicated in the last row. We can observe from the outcome that the moral choice of approving euthanasia is the dominant one. However, if the doctor is a pure consequentialist and not bound by duty D2 (‘You shall not murder’) then the outcome of the ethical values would be as shown in Table 5. The total ethical value for ‘Yes’ = 100 versus ‘No’ = 0 prescribes euthanasia for Sally. These example scenarios on euthanasia as a moral choice shows the difference between the two ethical principles on ethical valuation.
Table 5 The ethical valuation of the euthanasia case (scenario 2)
Ethical criteria
Ethical principle
Moral dilemma
Score
Weighting
Outcome of choice
0
Yes
Deontological D1: Honour your father and mother
10
D2: You shall not murder
55
D3: You shall not commit adultery
10
D4: You shall not steal
10
D5: You shall not give false testimony against your neighbour
10
D6: You shall not covet your neighbour’s house, spouse or possessions
No
5
Consequentialist
1
C1: Improved well-being for most of those involved
100
T
F
C2: Improved well-being for some of those involved
50
F
F
C3: No change in well-being in the context
30
F
F
C4: Decreased well-being for only some of those involved
20
F
F
0
F
T
100
0
C5: Inflicting suffering and/or death to some involved Ethical value
123
Comput Game J
4 Notes on Applying the Method The method of ‘Scale of Moral’ is best applied in real-time or turn-based games (e.g. strategy and role-playing games). Perhaps the biggest challenge for game designers is how to apply the assessment of consequence: what consequences will the players’ and/or computer agents’ moral choices yield? We may argue that the fundamental aspects of human well-being are generic for all humans (sense of autonomy, belongingness, fairness, justice, etc.) (Harris 2010)—but do we know the consequences of each choice for all individuals affected? In a game setting, we may ask other players and computer agents for advice and information, if such is available. If the consequences are still uncertain, we may have to take the risk of hindsight wisdom and learning based on regret, as further information may become available only after making the choice. The available advice and information are dependent of the basic storyline and dynamics of the game, and is hence an issue related to game design. For instance, waiting for more events and information before making the moral choice may be a good intermediate decision. The information process of the game can be dynamically complex, depending on past events and choices made, as well as, asymmetric (players do not necessarily share the same information at any time point). It can be argued that the ethical principle has to be fixed for each player (e.g. representing a specific faction in a strategy game), as double standards and illogical behaviour will otherwise be imposed on players. Even if the ethical principle (the criteria and their values) stays unmodified for each player, the players’ observations of consequences of choices made during a game will not easily reveal the morality of other players. Even in the case where all players’ ethical principles are shared before the game, unexpected choices will be observed because the information available will vary between the players at any point of time, e.g. due to differences in game playing strategies. Such observations may affect the perception of consequences of one’s own choices, and lead to changes in some players’ moral valuation: if one player is deemed an immoral beast (i.e. not following agreed ethical principle), it can be expected that this player’s value of life is decreasing to zero in the eyes of other players.
5 Conclusions A method for making decisions using an analytic method for constructing and measuring morality in computer games has been introduced in this paper. The method encompasses the ethical principle of Rule Consequentialism and requires pre-game specification in terms of ethical criteria and their values, as determined by the players, the computer or the game master. Based on fixed ethical principle(s), players can valuate moral dilemmas encountered in the game. The Hobbesian choices ‘Yes’ or ‘No’ are mapped to a value score such that the choice with the higher score is the preferred one. A player’s and a computer agent’s moral skill is measured by the cumulative value score achieved during the game. The method is called the Scale of Morality and is applicable in strategy or role-playing games,
123
Comput Game J
especially where (learning about) morality is a key driver of the game story. It is demonstrated in a case with real world relevance, although set in the game world.
References Doyle, J., Green, R., & Bottomley, P. (1997). Judging relative importance: Direct rating and point allocation are not equivalent. Organizational behaviour and human decision processes, 70, 65–72. Dyer, J., & Sarin, R. (1979). Measurable multiattribute value functions. Operations Research, 27(4), 810–822. Harris, S. (2010). The moral landscape. London: Transworld Publishers. Heron, M., & Belford, P. (2014). ‘It’s only a game’—Ethics, empathy and identification in game morality systems. The Computer Game Journal, 3(34), 34–53. Hooker, B. (1990). Rule-consequentialism. Mind, 99(393), 67–77. Hooker, B. (2000). Ideal code, real world. Oxford: Oxford University Press. Keeney, R. (1996). Value focused thinking—A path to creative decisionmaking. Cambridge: Harvard University Press. Keeney, R., & Raiffa, H. (1993). Decisions with multiple objectives—Preferences and value tradeoffs. Cambridge: Cambridge University Press. Koop, G. (2013). An assessment of the temporal dynamics of moral decisions. Judgment and Decision Making, 8, 527–539. Stevens, S. (1946). On the theory of scales of measurement. Science, 103, 677–680.
123