Previously, Reynolds and Picard have described a dimensional metaethical position and procedure to anticipate and improve consideration of affective computing systems [reynolds2005AUGCOG]. This section briefly describes this procedure and applies it to situations described by the Poker and Interview experiments.
Dimensional metaethics is a procedure that sees "what is good" about a system as related to the value of a number of different social dimensions. As such, it first advocates that designers of systems describe different social dimensions that they think are relevant to their system's design and use. The next step is then for the designers to list their assumptions about values along these dimensions. Lastly designers are asked to vary these assumptions and to speculate or (even better) to observe the resulting changes in ethical acceptability.
Table 7.1. Social dimensions relevant to evaluation of systems that mediate the communication of affect (a non-exhaustive list)
| Dimension | Examples | Description |
|---|---|---|
| Whom | Supervisor, Friends, Nicholas | The individual or individuals who receive the communicated affective message. |
| What | Telephone, Emotemail, Learning Companion | Artifact that acts as a transmitter or receiver for the communicated affective message. |
| Goal Relationship | Adversarial, Cooperative | The degree of conflict between the goals of the sender and receiver, which can be (but does not have to be) modeled from a game-theoretic perspective. |
| Power Relationship | Dominant, Submissive, Peer | Role that reflects the ability of either source or destination to alter the political, economic, or social situation of the other. |
| At Stake | Nothing, Ego, Money | What things are risked in using a system. |
| Genre of Emotion | Valence-Arousal Space, Categories, Emotional Orientation | Model used by the system to describe and encode emotion. |
| Valence | Positive, Neutral, Negative | Classification of transmitted emotion using an axis with positive or negative poles to describe feeling state. |
| Demeanor of Recipient | Angry, Sad, Excited | Emotional state of the message destination. |
| Gender | Female, Male, Intersex | Classification of either message source or destination based on reproductive role. |
| Ethnicity | Latino, Multi-Ethnic, Asian, Caucasian | Classification of either message source or destination based on racial or cultural identity. |
| Age | 18, Middle-Aged, Mature, Minor | Classification of either message source or destination based on duration of life. |
| Culture | Rural, Icelandic, Traditional | Cultural context of communication and of either message source or destination. |
| Risk | Dangerous, Safe, Hazardous, LD50 (lethal dose for 50% of population), LC50 (lethal concentration for 50% of the population) | Potential impact of communication on goals of message source or destination. |
| Symmetry | Balanced, Skewed | Information or power balance between users of communication system. |
| Trust | Trustworthy, Deceitful | The degree to which the message source trusts either the destination or the channel. |
| Designer | Affective Computing Group, Microsoft, GNU, Jussi Angesleva, Employer | Person or organization who created the system that mediates the communication of affect. |
| Experimenter | Stanley Milgram, Carson Reynolds | The person who conducts an experiment that evaluates the ethical acceptability of communication system. |
| Time | Now, Ten Years Ago, Tomorrow | When the system that mediates the communication of affect is used. |
| Informed Consent | None, Compliant with CFR Title 45 Section 46.116 | Does message source voluntarily consent to transmission of affective signals? |
| Security | None, C2, RC5-64, Hardened, Encrypted | Classification of security level of communication system or encoded signal. |
| Control | None, Partial, Complete | Degree to which message source can control the transmission of affective signals. |
| Feedback | None, Partial, Complete | Can the message source access the transmitted affective signal? |
| Transparency | Opaque, Open | Are the workings of the system that mediates the communication of affect visible for inspection, and by whom? |
| Proximity | Near, Far | Distance between message source and message destination. |
The above table presents a non-exhaustive list of many factors that could influence an ethical evaluations of systems that mediate the communication of affect. Let us consider a subset of these that change between the Poker and Interview Experiments.
Table 7.2. Assumptions in the Poker Experiment
| Dimension | Value |
|---|---|
| Whom | Friends |
| What | MixedEmotions |
| Goal Relationship | Adversarial |
| Power Relationship | Peer |
| At Stake | Small Sum of Money |
| Demeanor of Recipient | Neutral |
Participants in the Poker Experiment, it could be conjectured, were involved in a relatively friendly game among peers. In keeping with the tradition of poker they might have tried to assume neutral poker faces. In terms of what was at stake in using the system, there was a small amount of money to be won or lost. This may or may not affect one's ego depending on how seriously the player views poker play. Other mitigating factors include that participants played only one real hand, and it was with a stranger, so what might be blamed on skill in a higher-stakes situation was easily blamed on chance and other less personal factors in this situation. Also in a game like poker, bluffing (deception) is expected and not unethical.
Table 7.3. Assumptions in the Interview Experiment
| Dimension | Value |
|---|---|
| Whom | Potential Co-Workers |
| What | MixedEmotions |
| Goal Relationship | Adversarial |
| Power Relationship | Dominant-Submissive |
| At Stake | Ego, Small Sum of Money |
| Demeanor of Recipient | Nervous |
In contrast, participants in the Interview Experiments were placed into a relatively tense situation in which one had the power to hire or not hire the other. Over and above this the stakes of the situation were quite different: not being hired reflected not upon one's luck but potentially also on one's self-esteem or ego. Many subjects commented on how real this was as many were looking for jobs; one remarked "I am interviewing for jobs this week, so this seemed pretty real to me." Additionally, for many subjects being put into a condition where they had to lie (in a non-game situation) was stressful.
In seeing how these values differ between the two situations we may be able make some sense of why the questionnaire reports were so different. However, it doesn't appear that any single variable is directly related to these changes.
Dimensional metaethics is neither the only nor the first approach at providing a metaethical position for the evaluations of systems. In the paragraphs below we will briefly describe other metaethical positions that have been applied to computer ethics and compare them to dimensional metaethics.
Discussed previously in chapter 2, Value-Sensitive Design [friedman2002] considers Human Welfare, Ownership and Property, Privacy, Freedom From Bias, Universal Usability, Trust, Autonomy, Informed Consent, Accountability, Identity, Calmness, and Environmental Sustainability as values that may be of ethical consequence.
In many ways, a dimensional metaethics is an extension of value-sensitive design. Both provide a list of criteria which can be used to help structure evaluations and critiques of computing system. The chief difference between Value-Sensitive Design and a dimensional metaethics is what Kagan refers to as "evaluative focal points" [kagan2000]. Value-Sensitive Design is essentially a virtue ethics that focuses on different values that are of import to the design of computer systems. A dimensional metaethics instead focuses on dimensions along which the context of use of affective computing systems may vary.
Disclosive Computer Ethics [brey2000] "is concerned with the moral deciphering of embedded values and norms in computer systems, applications and practices." In contrast to value sensitive design, disclosive computer ethics focuses on justice, autonomy, democracy and privacy. Brey contrasts "mainstream" approaches to computer ethics (which he views as limited) with disclosive computer ethics. Brey sees the disclosive metaethical position as more of a process which is concerned with "disclosing and evaluating the embedded normativity in computer systems."
Dimensional metaethics position differs from this approach by not focusing on the embedded norms and instead considering the context in which the technology is used and factors, perhaps even very subtle, that might influence ethical judgments. Put another way, dimensional metaethics is not just artifact-centric, but also is fixated on the environment in which ethical judgments are formed.
Let us make these comparisons more concrete by providing an example ethical analysis of the interview experiment. Value-Sensitive Design would ask to consider the virtues of Human Welfare, Ownership and Property, Privacy, Freedom From Bias, Universal Usability, Trust, Autonomy, Informed Consent, Accountability, Identity, Calmness, and Environmental Sustainability as they pertain to the MixedEmotions affect sensing and communication system as it is used in the Interview Experiment. Disclosive computer ethics, on the other hand asks us to examine how a technology embeds various normative judgments. In the case of the interview experiment, we would examine how justice, autonomy, democracy, and privacy are embedded and supported by the MixedEmotions systems design. Dimensional metaethics position, in contrast would ask us to consider the use of MixedEmotions while the value of the different dimension vary. Thus we might consider MixedEmotion's usage when there is and is not a power relationship present between users.