A Path Out of the "Uncanny Valley"?

A Path Out of the "Uncanny Valley"?

Postby JohanEklund on 20 Nov 2009, 13:03

Hello good people of Robots Podcast forum!

My name is Johan Eklund. I'm a psychologist and have written a thesis about our psychological reactions to robots. Early in the research for my thesis I naturally came across Masahiro Moris hypothesis about "the uncanny valley". In this way my thesis became a critical examination of Moris hypothesis. The short version is that I argue for an abandonment of "the uncanny valley" and the need for a more nuanced view of human reactions on humanlike robots.

Before I give you the long version, let me say that I'm posting this with the hope that you might comment with your qualified thoughts.

Now, the long version:

By reviewing the empirical investigations in the field I examined our reactions on three levels: immediate, mediated and reflective reactions.

To access the “immediate” reactions on robots, I have reviewed literature on both facial attractiveness and on experiments with “the uncanny valley” that used static images. From these reviews, I argue that the "near-realistic" robot face can elicit negative reactions, such as disgust, but that it is by no means unavoidable. By careful selection of facial features with high symmetry, averageness, and slightly enlarged features it is possible to create a near-realistic robot that does not elicit negative reactions on the immediate level.

In my section on the “mediated” reactions on robots I explain and discuss higher cognitive reactions towards robots. I explain how our above mentioned immediate reactions towards the robot "as if it were alive" can create a dissonance with our mediated knowledge of the robot's artificial nature.

On the reflective level the illusion of the "living" robot can be so rich and compelling that it can challenge our notion of “humanity” and our existential defense: "if this object (the robot) is alive, aren't I just an object as well?". This can challenge our existential defenses and elicit existential anxiety on the reflective level. The mediated and reflective reactions are not unavoidable ones, but they are more difficult to overcome and cannot be mediated by better design of the robot. To overcome some of the cognitive and reflective reactions would require prolonged exposure to near- realistic robots so that one’s cognitive schemes can be adjusted or new ones constructed. An alternative would be to circumvent the challenge of our existential defense by treating the robot as a human.

Our more positive reaction to the abstract (or less familiar, in Mori's words) robot is also examined. This aspect of Mori's hypothesis is validated and I add a missing theoretical explanation with foundations in Scott McClouds examination of the comic as a media.

I conclude that a more nuanced view of human reactions on robots is needed. I see the hypothesis of “the uncanny valley” as on par with ancient teleological explanations of “fire as seeking to reach back to the sun”. An explanation which serves one fine in simple usages, say fry an egg. But if one is to harness the power of fire to reach the stars, the simple hypothesis leaves you helpless. Analogous to the old hypothesis of fire, the hypothesis of the “uncanny valley” seems to work well in the design of the abstract robot, but leaves you helpless when walking into the minefield of heterogeneous reactions to the humanlike robot. Mori's hypothesis must therefore be rejected as a first step to a more differentiated view of human reactions to the humanlike robot.

I'm looking forward to your comments!

Thanks for your interest!

Kind regards

Johan Eklund

(You can also contact me on twitter. Username: LowImagination)
JohanEklund
 
Posts: 4
Joined: 20 Nov 2009, 12:33

Re: A Path Out of the "Uncanny Valley"?

Postby nano on 24 Nov 2009, 00:33

Hi Johan,

Very interesting analysis. It is great to see psychology in robotics!
I personally agree that the uncanny valley theory does not fully capture the complexity behind human-robot interactions. Although I generally do enjoy humanoids that have more cartoon like features than those which resemble humans too closely.

My main question is whether or not we will actually want to introduce near-realistic robots and for what applications. Currently, roboticists are concentrating on making robots that people will want to interact with. It would be interesting to know when it is useful to push humans to interact with or accept robots that are currently not in their comfort zone.

That being said, a couple questions (off the top of my head) or limitations regarding the uncanny valley come to mind:
1) Can robot behaviors be uncanny. Likewise, can the behavior of a robot be decoupled from its visual features. As an example, the Big Dog behavior is quite uncanny:
http://www.youtube.com/watch?v=cHJJQ0zNNOM
2) What is the impact of culture in how people perceive robots.
3) You mention facial "attractiveness". It is often assumed that people want beautiful robots, is this actually the case? Might they in fact prefer robots that look friendly but are not necessarily attractive so as not to feel inferior to the robot? Especially in the case of near realistic robots.
small is beautiful
User avatar
nano
 
Posts: 61
Joined: 25 Jun 2008, 15:34

Re: A Path Out of the "Uncanny Valley"?

Postby JohanEklund on 24 Nov 2009, 18:04

Hi nano!

Thanks for your questions! All of them very relevant and interesting. In the following I'll I will try to answer as best as I can.

To your main question about why we should implement humanlike robots, I have a broad answer and three specific applications: Mainly I think that it's not going to be a question whether we should, but rather about how we will react to and cope with the humanlike robots that inevitably WILL be constructed. The same fascination that propelled my thesis has always existed. From paintings on the cave wall to the statue of David, humans have always sought to produce replicas of themselves while aiming for the creation of "the perfect human". Our new technology has now made it possible for us to pursue this even further. And because this pursuit is fueled by our existential reactions to the circumstances of our life, it will continue and almost inevitably lead to more and more advanced humanlike robots. When we add to this the potential the sex-industry (sadly) will see in this (http://www.guardian.co.uk/technology/2009/sep/16/sex-robots-david-levy-loebner) I think that it's more than indicated that it's not something that anybody will be able to control. Instead I think we should prepare for the obvious problems (e.g extreme attachment and the problems of mistreating humanlike robots) and focus on the possible positive applications of humanlike robots. To exemplify some of the positive applications I will list the above mentioned three specific applications:

David Hanson: Robots with empathy. Natural interface and therapeutic robots.

David Hanson has made humanlike robots with the ability to mirror human emotions in facial expression (http://www.ted.com/talks/david_hanson_robots_that_relate_to_you.html). Hanson explains that this work might result in "robots with a capacity for empathy". Even though Hansons formulation is somewhat problematic, it's still a perspective on an application for humanlike robots: the familiar and natural interface. We have an intuitive knowledge about how to interact with other humans. Therefore a robot with a humanlike appearance and/or behavior will be very easy for us to communicate with. Hanson also gives a perspective on robots with therapeutic abilities (think advanced Paro) which might work even when we acknowledge the "therapeutic robot" as artificial (think ELIZA: http://en.wikipedia.org/wiki/ELIZA_effect).

Ishiguro and Macdorman: Robots as a stable variable in psychological experiments

Hiroshi Ishiguro and Karl Macdorman suggests that an application for humanlike robots is psychological experiments where the humanlike android could act as "a stand-in for a human being" and thereby providing much better controlled variables and eliminating of confounding variables (e.g mood and unintentional eye-movmements) (Macdorman, 2006).

Prosthetics: the cyborg as ambassador for the humanlike robot

A "spin-off" of the development of very humanlike robots will be better prosthetics (e.g Dean Kamen's "Luke arm"). Personally I also think that the more advanced cyborgs we will see in the future will act as a sort of "ambassador" for humanlike robots. We will get used to very humanlike technology and thereby be more at ease when this technology acts autonomously (actually I think that the cyborg is the only thing able to compete with the development of humanlike robots. If humans should stop making humanlike robots it'll only be because we've channeled all our focus onto to the development of the cyborg).

Now to your other questions:

1):Can robot behaviors be uncanny?

Yes, robot behavior is also a factor and just represents one of the different levels which humans can react to (other levels being "appearance" and "symbolic communication" (i.e talk)).

Pollick et al. (2001) managed to isolate the effect of movement and indicated that fast movements where associated with emotions such as "angry" and "happy". On the other hand slow movements where associated with emotions such as "sad" and "content". That is to say that fast and powerful movements more often are interpreted by humans as "hot emotions" versus the slower movements more "cooler emotions". In this way one can reduce the chances of ones robot being interpreted as "angry" simply by reducing the speed of the robot (the robot could still be interpreted as "sad" but this would probably be preferable to "angry").

Besides a "direct" reaction to movement there is also a chance of a negative reaction to movement, due to a inconsistent interaction with appearance. Separate visual pathways in the brain initially process form and motion and then at a later stage integrate this information into a representation of a human (Giese & Poggio, 2003). In this way visual information about a robot with humanlike appearance but slow jerky movements may be hard to integrate into a mental representation of a human. These problems with the integration of the different information might lead to a state of unease or at least uncertainty about what is being observed. Here it's the inconsistency between behavior and appearance that causes the negative reaction (Pollick, in press).

Other issues regarding behavior is human reactions to proximity and so on, but I'll move on to your next question now.

2) What is the impact of culture in how people perceive robots?

Culture is in some extend a variable in our reactions to robots. The western focus on the individual is what underlie the existential reactions I mention. If we saw objects and people as having the same qualities and value, ( as the Shinto religion does) we might not have the reaction on humanlike robots as we apparently do. Macdorman et al. (2009) states in their conclusion that: "philosophical doctrines of Japan—Shinto, Buddhism, and neo- Confucianism—have never impeded Japan’s progress in science and technology. This contrasts with the enduring conflict between science and religion in the West." Other factors that explain the prominence of robots in Japan is that robots have played an important role in maintaining Japan’s manufacturing base.

3) Do people want beautiful robots?

No, I don't think that people necessarily want beautiful robots. I agree that an acceptable appearance would be preferable and this was also the reason i looked into attractiveness. With knowledge about attractiveness it's fairly easy to move from a "disturbing" or "repulsive" to an "acceptable" humanlike appearance. This move from "repulsive" to "acceptable" makes a big difference compared to the minor tweak you can achieve by changing the robots appearance from "acceptable" to "very attractive"

So yes, not a focus on "attractiveness" per se, but rather a focus on making the robot "non-repulsive" and "non-discusting"; i.e "acceptable".

Long post. Got a bit carried away there! Hope this answers your questions! Dont hesitate to write back if you have further questions or would like elaboration.

Kind regards

Johan
JohanEklund
 
Posts: 4
Joined: 20 Nov 2009, 12:33


Return to Podcast

Who is online

Users browsing this forum: No registered users and 0 guests

cron