Critical Analysis: Too real for comfort? Uncanny Responses to Computer Generated Faces

MacDorman, K., Green, R., Ho, C. and Koch, C. (2009). Too real for comfort? Uncanny responses to computer generated faces. Computers in Human Behavior, 2, pp.708-709.

 

The Uncanny Valley effect is characterized by increasing photorealism in CGI characters which in turn make audiences feel increasingly uncomfortable, (Karl F. MacDorman et al., 2009). This is often accompanied by the inability to feel empathy or emotional connection as the on screen persona is perceived as fake or artificial. The reason why I chose this article is because, as a character artist, the Uncanny Valley is a phenomenon that has, for many years remained a limiting factor in the creation of realistic and believable human faces across the film and gaming industries. This article is significant for my work as it explores psychological theories like empathy, mate selection, threat avoidance, cognitive dissonance and psychological defenses when it comes to character perception and studies how individual factors like facial proportions, skin texture and level of detail can increase the believably of a character and can in turn directly aid me in the creation of more believable or if needed unsettling characters.

The article first appeared in the Computers in Human Behavior Journal in 2009 by Robert D. Green, Chin-Chang Ho, Clinton T. Koch and Karl F. MacDorman who is currently employed at the Indiana University, School of Informatics and Computing. The journal itself focuses on the use of computers from a psychological perspective (Behavior, 2016) – I believe this to be an important factor as despite major innovations in GCI character rendering, the Uncanny Valley remains as much of an obstacle now as it did 20 years ago. For this evaluation I will discuss the final two pages of the paper where the authors focus on summarizing and defining their findings from four experiments they conducted.

One of the first observations discussed shows that faces which were regarded as looking more believable had higher levels of detail in terms of polygon-counts as well as use of photorealistic textures. The explanation provided for this is that photorealistic textures made it easier for subjects to identify the character as natural looking, however this use of photorealistic textures also produced very “eerie” looking results when combined with varied facial proportions. A later study however, showed that the most photo-realistic CG character was not perceived as the least eerie, instead participants agreed that the most believable looking GC character was one that relied on 75% photorealism, while also adding stylization like bronze and line texture materials.
This observation can be traced to contemporary GCI characters where films like Star Wars Rouge One, had CG characters that made many audiences feel uncomfortable (while striving for %100 accurate characters), while games like Uncharted 4 that used inferior rendering, shading and lighting models, received praise for their life-like protagonists (which were a mixed product of scanned real world data and artistic input).

While normally the results documented here can be difficult to accurately record (as multiple visual factors often work together to improve visual fidelity), the researchers have taken care to isolate and link outcomes across variables by focusing on specific character aspects like materials, geometry and textures in separate and combined experiments. The findings, therefore not only show how different visual factors can positively or negatively affect character believability, but also highlight which contribute more towards the Uncanny Valley effect. For example the findings make texturing a strong contributor to the effect in all instances, while shifting proportions have a strong effect only under certain conditions like their combination with high quality, realistic textures.

When it comes to critical observations, there are several prominent ones. However it must be noted that the majority of these are commonly found across most studies in this field and as such are limitations experienced by the scientific community as a whole and not limited to this paper alone.

One of these is that the data collected should not be taken at face value. This is due to the fact that all studies on the uncanny valley effect lack a common unit of measure across different articles and journal publications. While the “eeriness scale” used in the article aims to unify the results of the different experiments, there is no coherent method or quantifiable unit used to compare the outcomes of these experiments with others in the field. For example many scientific papers can compare size and temperature in cm or °C, but the scientific community has yet to agree as a whole on a unit that can define how a visual stimulus can affect a test subject. As such the numeric results documented in such studies can be difficult to reproduce.

Another factor that is commonly left underexplored is the rendering aspect of the character. The methods of stylization here do explore the use of non-physically based shaders, however shading models that focus on photorealism that use sub-surface scatter through subdermal maps or specular maps for gloss are left unexplored. While there are extensive studies on the rendering of human skin that do employ such shaders, they are almost never explored in combination with more fundamental character aspects like the experiments into facial proportions or texture detail, conducted here. This potentially leaves out critical factors like skin smoothness or softness which can lead a character to appear plastic and very unnerving if combined with accurate textures and proportions.

One significant limitation exclusive to the studies here however, is the artistic control over the base model. The only variations used in the experiments are that of a white middle aged male and that can be seen a strong limitation to some of the results and by extent – recommendations. This is because factors like proportions and stylized materials work very differently across characters of diverse ethnicities and genders. The rule of mixing stylization with photorealism to exaggerate proportions or mute details can be potentially controversial should it be used incorrectly and highlight racial stereotyped – an example of this can be seen in games like Heavy Rain (the character of Mad Jack) and Indigo Prophecy (the character of Takeo). This I believe is the strongest limitation of the article as it is the only one that can be exclusively linked to the experiments conducted here and can also have a direct impact on the final recommendations.

In the end the article follows a strong, clear structure that separates the overall study into clear and logical sections that are then tested in an isolated and mixed environments. The results are clearly documented, displayed and used to generate recommendations and conclusions for character creation that are based in psychological theory. The findings themselves are important as they show several correlations between texture, model and material quality and how different combinations of the three can produce results of varied believably. On the other hand while the paper suffers from two limitations common for studies in the field it also suffers from a limiting artistic control over the tested model that can undermine the universal application of its recommendations. A key lesson from this paper none the less is the fact that while photorealism can take a design up to the uncanny valley it is through stylization and artistic liberty that it can cross it.

 

Bibliography

Behavior, C. (2016). Computers in Human Behavior – Journal – Elsevier. [online] Journals.elsevier.com. Available at: http://www.journals.elsevier.com/computers-in-human-behavior [Accessed 18 Dec. 2016].

MacDorman, K. (2016). Karl F. MacDorman. [online] Macdorman.com. Available at: http://www.macdorman.com/ [Accessed 10 Dec. 2016].

 MacDorman, K., Green, R., Ho, C. and Koch, C. (2009). Too real for comfort? Uncanny responses to computer generated faces. Computers in Human Behavior,2, pp.708-709.

Advertisements

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s