Talk:Affective computing
This is a very informative article and I have enjoyed reading it.
I have two suggestions: (1) To enhance the discussion of emotion modeling, and (2) to explicitly mention ethics
(1) Discussion of emotion modeling
The initial motivating example could also explicitly mention that the robot gradually builds an affective model of the human (a user model) and that affective modeling methods are used for this... which are then discussed in the last section of the article.
The section 'Sensing, recognizing, modeling, and predicting affective state'... doesn't really discuss modeling, unless the brief mention of predicting the (n+1)th affective state from the nth state is considered modeling. Modeling generally refers to the topics the author describes in the last section - emotion simulation. It would be good to tie the discussion of emotion simulation to emotion (affective) modeling, and perhaps differentiate the more complex architecture-based models used in simulation, from the essentially stochastic models that may be used to predict a future affective state from an existing state, based on historical subject data.
(2) Ethics
The end of the last paragraph of the section'Sensing, recognizing, modeling, and predicting affective state' would be a good place for a sentence or two about special ethical considerations in affective computing, and any developments in this area.
also, the last sentence of the paragraph beginning with: "Affective Computing faces mane decades of challenges.." is unclear.
===============
I generally agree with the peer reviewer on the article being basically informative. I would however suggest that there are a few important aspects that should be addressed to improve its value (some of which overlap partially with recommendations put forward by the peer and not yet addressed in a fully satisfactory manner).
(1) I am generally missing a critical stance towards the field as a whole, and regarding the state of the art in the different areas addressed. Issues connected to the introduction of such technologies as a whole should not be washed over lightly (<-- please DO forgive me for exaggerating with my formulations, this of course is to be understood for the sake of making my point only!); and the known technical and theoretical difficulties and limitations ought to be explained more clearly. In this context, a short explanation of the evolution of the field so far (including e.g. the understanding/correction of the use of 'emotion' vs. 'affect') would be helpful, as well as a succinct the explanation and underscoring of the genuine _novelty_ of the research field (in comparison to established emotion research in psychology, sociology, anthropology, philosophy, etc.): relevant results are scattered all over these different fields, and the mapping of the full richness of "emotional life" to what is covered in the different fields is very incomplete. Of course there are also methodological reasons:
It could therefore be worthwhile to report some aspects regarding "Psychology 101" and human research: E.g., as also experienced e.g. within Humaine, basics of good practice of research concerning/involving humans are often not taught in the context of engineering curricula, even at globally leading institutions; the inherent difficulties of singling out, measuring and modelling psychological quantities, and of releasing technologies interacting with humans - as humans will inevitably adapt to whatever technology they are exposed to and there thus e.g. is no "full undoing" of released solutions, and there are fundamental methodological issues that have to be considered.
The _well-known_ "incompleteness" of existing emotion theories also relates to the second-but-last paragraph of the section "Research scope and challenges": it is not so surprising after all that "subdued" should not be covered in "the dominant psychology theories" (if this is in fact the case? the relation of the notion of "subdued" to terminology in sociology and emotion research should be scrutinised with care) -- cf. e.g. Clark Elliott's extension of the OCC model for pedagogical purposes reported in
Elliott C.: Affective Reasoner personality models for automated tutoring systems, Proc. Workshop on Pedagogicalk Agents, 8th World Conference on Artificial Intelligence in Education (AI&ED-97), Kobe, Japan, 1997. (available online [2008-06-22] at http://condor.depaul.edu/~elliott/papers/aied97/aied97.html)
(2) Some coverage of ethics has been introduced at the location suggested by my peer. But given both, the horizontal (=pervasive) nature of ethical concerns, as well as the need to establish it as core, routine, and *monitored/supervised* principle, I would suggest that it should be allocated a section of its own.
As aspects to be addressed (list not meant to be exhaustive!), we would suggest:
- Procedural _and_ outcome issues:
. procedural: how to carry out research; basics of human research: informed consent; ethics committees; the importance of the researcher _not_ being the sole/main judge of a proposed research activity, etc.
. outcome: what are the consequences if research is successful and products are released? What kind of educational, legal, and _social_ scaffolding (which stakeholders, ...) is required? How can it be achieved?
- Symmetrical aspects: abuse of affective technology (i.e., e.g.: abuse of affectively sensitive ECAs by humans)
- The impact of scope and context of use (i.e. the "semi-intelligent information filter" issue: for this kind of technology in particular, how can it be ensured that all premises for inferences and functions in fact do hold? How is the technology to operate when these do not hold?)
- The scope and limitations of ethical clearance: assessments of ethicality of research...products need to be revisited (i.e. there's no "once-and-for-all" ethical clearance)
(3) The references need to be examined with care, making sure the individual records are complete: many entries lack the title of the paper (!), as well as issue numbers (in particular for the recommended special journal issues); page numbers are missing for Breazeal & Picard 2006 and Burleson et al. 2004.
Furthermore, the references need to be linked with the article text at the appropriate/related places.
We are also not sure about classifying the ACII proceedings as "recommended reading" (at least, we would suggest to group the two entries under a heading such as "Affective computing and intelligent interaction conference proceedings").
Some further more detailed/local comments have been inserted in the article text.
=
added comments PP 2009-02-07:
Overall, I do agree with Eva's positive evaluation.
General remark ad Figures: Not sure whether it currently is the author's or the editors' responsibility, but: the "alt" text of a figure should summarize/describe what is displayed in the figure itself, and not repeat what is anyway already stated (and thus accessible also to impaired or text-only users) in the legend of that figure.
ad "Research scope and challenges:"
Figure 1 shows boxes in different colors that are not explained; Fig 1 in fact is never referenced in the text. The (number of) areas identified in the text (5) also do not match the content of Fig 1. Perhaps the content of Fig 1 could be rephrased as application scenarios of Affective Computing, and be referred to in the text as providing a different additional perspective to better grasp the broad scope of the domain.
area (2): "(2) .. and predicting emotional and affective states, e.g., the customer looks and sounds angry now and <<if I say this>> <-- this latter bit should be omitted, as this belongs to area (3). It could e.g. be replaced by: "...is only likely to grow angrier still;"
((side-remark: perhaps some reference to the recently growing area of sentiment analysis could be placed here))
vs. "(3) methods for computers to respond intelligently and respectfully to handle perceived affective information, e.g., the strategy of acting subdued around a person who is upset
suggested re-phrasing (minor):
"For years, Apple Macintosh... when failing to boot (the current expression, "not booting successfully" could be misunderstood as actually booting, if not 'completely' successfully in one way or another)
ad "Displaying, communicating, or mediating expression of affect:"
"Some people prefer non-affective interaction with computers and do not wish for ..."
We would suggest to place this paragraph within the broader context of good principles of deployment of *any* technology. This could e.g. be addressed with a short introductory sentence, such as: "The key issue of end-users retaining control of course also applies to Affective Computing: Some people will always prefer non-affective interaction..."
"(see ethical/social/philosophical section at end of this article for more examples of issues like this.) "
suggested rewording:
(see the section on ethical/social/philosophical issues at end of this article for more examples.)
ad "Sensing, recognizing, modeling, and predicting affective state:"
"Artists may also use affective technologies to provoke reflection, debate and discussion about its potential uses and misuses. " => why would/should use of affective technologies for such purposes necessarily be limited to artists; without wanting to diminish the important contributions of artists: why should artists be the main channel (as opposed to e.g. a public discourse instigated by the very researchers, designers, stakeholders)?
The section "Sensing, recognizing, modeling, and predicting affective state" could acknowledge more clearly the issues of/amplified by moving out of the laboratory, in particular the challenge of relating correctly measured sensor data to the actual (cognitive) causes, or also the issues currently held implicit in expressions such as "likely to occur", ...
ad "Synthesizing and simulating emotions or implementing their regulatory and biasing functions:"
end of first paragraph:
For example, cognitive appraisal models can be used both for recognizing antecedents that may give rise to an emotion (analysis) **and** for actually giving rise to a state in a computational system (synthesis).
second paragraph:
"this idea comes out of studies showing emotion influences **on** human creativity"
ad "Ethical issues, etc.,:"
4th paragraph:
"it not only violates the fundamental principles of affective computing research to respect affective preference, but it also violates standard ethical practice. "
The status of the notion of "standard" ethical practice is unclear. Perhaps it could be replaced by "ethical good practice"?
"No affect recognition system is always perfect" Given that this whole paragraph is about pointing out how a notion of "perfection" is misplaced altogether in the context of natural settings (in particular, but not only, for technology), the wording of this opening sentence should be reconsidered, e.g. along the lines of:
No technology deployed "in the wild" can ever be perfect, and affect recognition is certainly no exception.
(if considered useful, it could be pointed out how pragmatics of technology deployment can very well make up for various technological deficiencies)
And along the same line, we would like to suggest to revised the final part of the paragraph: what does matter at the end of the day is that users *are* careful in *practice* (cf. the often-cited case of having to remind people that fresh coffee in fact may be HOT(!)). We certainly do agree that full responsibility to achieve this final aim cannot be offloaded completely onto the shoulders of technology designers. In this sense, we do agree that their aim is to inform users sufficiently so that they _can_ be very careful. The emphasis on such empowerment (stress on "can") could perhaps be made clearer.
Re: misuse of affective information:
A possible addition would be the danger of abuse of affective technologies by exposing users to conditions outside their volitional control (e.g. this reviewer did face a research proposal suggesting to punish a student's tamagotchi all the way to its virtual death in case of the student's lack of success with their studies!)
Final paragraph:
suggested changes:
"need to not fall prey to" --> "*must* not fall prey to" "common scientific tendency" --> "scientific bad practice"
re: "and seeking public input to steadily discern what should be done in developing technology to improve human experience." According to our understanding, a key point of ethical good practice is the understanding that the developer/research should _not_ be the judge of their own work (This is why proposals are submitted to ethical boards, and why such boards in turn must include laypersons). Thus, what is (to be) sought for is not so much to seek for "public *input*", rather, what ought to be done is to submit and expose the work (and before actual work: the proposals of work) to scrutiny by empowered public entities throughout the process (research: ethics boards, deployment: jurisdictional/political/societal/stakeholder bodies).
<EOT>
