[ CogSci Summaries home | UP | email ]
http://www.jimdavies.org/summaries/
@Article{HerbergSaylorRatanaswasdLevinWilkes2008,
author = {Herberg, Jonathan S. and Saylor, Megan M. and Ratanaswasd, Palis and Levin, Daniel T. and Wilkes, D. Mitchell},
title = {Audience-contingent variation in action demonstration for humans and computers},
journal = {Cognitive Science},
year = {2008},
volume = {32},
pages = {1003-1020}
}
The actual paper can be found at http://csjarchive.cogsci.rpi.edu/2008v32/6/HCOG_A_302425_O.pdf
When we interact with others, we must not only understand their beliefs, desires, and goals, but must also tailor our behaviours to accommodate their knowledge and perceptual skills. When demonstrating simple actions for infants, adults produce larger actions with more salient and more frequent pauses.
These findings are important when considering full range of potential audiences for action demonstrations.
Asked participants to demonstrate actions for human or computer audiences.
"Identifying differences in the demonstrations will not only help to develop a practical understanding of human-machine interaction, but will also illuminate basic cognitions about fundamentally different kinds of intelligent systems." [p2]
How do individuals succeed in action segmentation?
One proposal: both bottom-up and top down knowledge of the world play essential roles
Bottom-up: Rely on salient perceptual features in motion stream to mark boundaries between action units [p2]
Top-down: Recruit knowledge about the world, such as actors' goals and intentions, in identifying boundaries [p2]
What guides observers' attention to top-down versus bottom-up features?
Zacks (2004): When observers are shown moving shapes, those who were told the movements were produced by two people playing a video game segmented the motions into larger units than those who were told that the movements were randomly generated. Being told that the motions were generated by intentional agents may lead to top-down segmentation which produced larger units
Killingsworth, Saylor, and Levin (2006): Adults segment action into larger units if they believe the segmentation is for the benefit of a person or anthropomorphic robot than for a non-anthropomorphic computer. Participants' judgments of the intentionality of agents predicted the tendency.
If segmentation differs according to an audience's capacity for intentional reasoning, demonstrations may differ as well.
Will people modify their actions for computers? Mechanical agents (e.g., computers and robots) are able to engage in action, but may not understand others' intentions Although people sometimes treat computers as social actors, they may apply psychological modes of reasoning less deeply to computers than to people, or may apply qualitatively different modes of reasoning about these systems
(Nass, Fogg, & Moon, 1996): In human-human interactions, when an individual works with another person to solve a problem, their interdependence leads the individual to conform his or her opinions to the teammates'. This effect is seen in human-computer team situations
People prefer interacting with a computer who exhibits a "personality" similar to their own (Nass, Moon, Fogg, Reeves, & Dryer, 1995)
People sometimes differentiate between computers and humans during interactions
(Mishra, 2006): People are likely to take praise or blame offered by a computer at "Face-value", regardless of the difficulty of the task they attempted.
People do not always align their interpretation of computer and human behaviour
(Levin & Beck, 2004): Describing a computer system in non-anthropomorphic terms influences how people reason about it
People will give less weight to an actor's intentions when segmenting for a non-anthropomorphically described computer than a person (Killingsworth et al., 2006). Non-anthropomorphically described computer systems may not be viewed as capable of intentional reasoning which may lead people to treat computers and people differently during action demonstrations.
How? One way may be to selectively highlight portions of action sequence the individual thinks the computer is capable of producing and may be relatively unlikely to provide discrete social behaviours that may draw the attention of an audience which is capable of intentional reasoning to their goals
Study had different trial "audiences" (Photograph of trial group): Amazonian tribe member lacking knowledge of Western culture, a toddler and a computer. Task is to instruct the "audiences" on how to perform one of three tasks: tying a shoe, Tower of Hanoi and sorting cards. Videos of the trials (demonstrations) were coded for social modifications (e.g., looks to the audience picture, points, smiles, repeats and does purposely incorrect actions). Using a social modification correctly requires some recognition that one's social partner is an intentional agent.
Motion measurements (e.g., velocity, pace, path efficiency, path length and pause abruptness) were analyzed by a handtracker.
Velocity: how fast the hand moved
Pace: How continuous and uninterrupted the motion was
Path efficiency: How direct and small the motions were
Authors predicted that if people fail to represent a computer as capable of intentional reasoning, then they should use fewer social modifications and more motion modifications for the computer relative to the humans. [p6]
If people view the toddler as still developing their action analysis skills, then they should include more social and motion modifications for the toddler than for the adult. [p7]
Human demonstrations contained a higher rate of social modifications than the computer demonstrations. [p10]
This was present for both the toddler and the adult audience
Analysis of results revealed a marginal effect on motion modification production for computer audiences. [p11]
"Post-hoc comparisons revealed that the human-computer difference was present for both human audiences but that the path length effect was only present for the adult audience". [p11]
There was a marginally higher rate of social modifications for the toddler audience than for the adult audience. [p12]
People adjust their demonstrations to a mechanical audience to include fewer social modifications than included for human audiences
Demonstrators would vary how much they highlighted their motions for mechanical versus human audiences.
Participants increased their speed relative to the human audiences for a computer. Possible reasons:
Range of motion increase for the computer
Participants partially mimicked a mechanical style of movement when demonstrating for a computer
Action sequences may be analogous to a basic-level concept, with different movement styles for achieving the action as subordinate-level concepts. [p14]
human-human scripts are only activated when features of human-computer interaction suggest the appropriateness of such scripts
There were no features in the interaction that suggested that the computer was a social agent. The finding of more social modifications for human than computer audiences may have resulted from automatic responses to human faces, as opposed to a more cognitive representation of the human audiences as having a capacity for intentional reasoning. [p16]
Demonstrators may have overcome a default tendency to take their social partner's cognitive and reasoning capacity for granted when they were faced with a non-human agent. [p16]
"It would be necessary to add a social module that would be sensitive to within-species differences in capacities and would produce variations in levels of social behaviour according to these capacities" [p16]
Future research:
Investigate how beliefs about canonical motion patterns of an audience influence people's demonstration style
Whether intrinsic representations of audiences may lead to more automatic demonstration differences
Within-subjects design can be interpreted as a limitation as it may create artificial demonstration differences based on demand characteristics Adults consider differences in their audiences when demonstrating actions, which was present with two human audiences as well as human and mechanical audiences.
As mechanical artifacts become more ingrained into the everyday lives of individuals, there will be a greater need to understand the factors that shape beliefs about the systems and interactions with the technology.
Page numbers result from pdf document of article
Back to the Cognitive Science Summaries homepage
Cognitive Science Summaries Webmaster: