Narrowing down the research topic ...

Prompt: Below you see several (6-7) artifacts that can be pedagogical agents or intelligent computer assisted instructional systems, each at the different stages on the uncanny valley curve. How do you think EACH of these agents/systems will differ on 

1. how the student will interact with the agent/system (and why)?
2. what the agent/system may be capable of when it comes to evaluating/assessing student's learning (and why)? 
3. for what kind of learning would such agent/system be good for (and why)? 

Please reflect on two or more of the required readings from this week or (past readings) to justify/support your answer. You can answer each separately or all together in one paragraph. 

Hal (2001: A Space Odyssey)

It's been a long time since I have seen this movie (!), but Hal is a classic example of feedback delivered by a Voice of God (VoG) agent as described by Goldberg & Cannon-Bowers (2015), a disembodied voice that could interact with a learner in a number of ways (at least until he took control of your ship and all its decisions). Instructional uses could include serving as a guide in simulated learning or gaming environments that could provide hints or feedback on learner's performance. This feedback could be delivered either after the performance or learning task has been completed, or at a more granular level, providing feedback at each step of the performance or learning activity (VanLehn, 2011). 

In this particular case, Hal is frightening because 1) he is hardwired to the ship and its controls and 2) seems to be superior in intelligence to the protagonist of this film, as if playing a game of chess or an equivalent game of strategy, and 3) is aware of these facts. In a different learning setting, however, and because Hal is a learning machine, it could be useful agent in complex or nuanced situations, such as scanning the environment for physical cues on how to behave in a different cultural setting. And because he is a disembodied voice, he would be particularly useful for heavily visual learning environments, so as to not cognitively overload the student by overtaxing the dual channels of information intake (Mayer & DaPra, 2012). That said, "student self-efficacy was significantly negatively impacted when the agent was evasive and without motivational messages" (Baylor, 2007, p 12), as sadly was the case with Hal. Finally, according to Goldberg and Cannon-Bowers (2015), learners showed significant increase in transfer success when the feedback actually came from an embodied agent, not just a voice (sorry, Hal). Still, VoG guidance could be particularly useful for heavily visual learning environments, so as to not cognitively overload the student by overtaxing the dual channels of information intake (Mayer & DaPra, 2012).

Keepon Robot

This interactive, yellow robot exudes rubber duckie-like cuteness as well as playful expressiveness. This robot was created to help children with autism develop social skills, who might otherwise be overwhelmed by interacting with real human children. Accordingly, Keepon delivers joyfully emotive feedback and reactions (such as dancing if music is played) in a way that makes the autistic child feel safe through "simplifying the social context" when interacting. I found it interesting that Keepon can take a bit of a beating because of its spongy softness, but that it tends to grow on these children. In fact, Baylor states that non-human but anthromorphic shape agents (including 3D shapes and animals) indicates that even non-human agents are perceived as emotional, believable, and human-like" (2007, p 12).

Nexi (MIT Media Lab)

Nexi is a pedagogical agent that is described as "mobile, dexterous and social" which could be useful in a wide variety of learning contexts. For example, Nexi can actually move (mobilize) among students to interact and deliver feedback to the instructor, as well as for one-on-one learning interactions. According to Baylor, "an anthropomorphic pedagogical agent can impact learning and particularly motivation, regardless of the back-end computer system" (2007, p 11). Even though it does not try to look human, per se, it is anthropomorphic and possesses a wider range of emotional feedback response abilities, both gestural and spoken. Because Nexi is dexterous, it could be useful at showing how to perform physical tasks or processes. Even though it does not appear "human", per se, it is anthropomorphic and possesses a wider range of emotional feedback response abilities, both gestural and spoken. 

And another use could be as a conversational agent, since her communication level (it seems) is quite polished and seemingly "natural." Nexi's voice is pleasant and human-like not machine-like, which makes the learning experience more agreeable, but does not seem to interfere with the learning process itself (Baylor, 2007); others find the voice quality and characteristic--that it is human--does matter effect learning for the better, a principle known as the voice effect (Mayer, 2012). Less disputed perhaps is that because Nexi's can non-verbally express states of joy, confusion, sadness or other emotional states, it is likely to increase learner motivation because it provides a feeling that the learning experience is a shared one, that the learner and Nexi are "in it" together (Mayer, 2012) and so may be particularly suitable for younger students. 

CB2 Child Robot

CB2 is a perfect demonstration of uncanny valley, because it is a step closer to being humanlike than Nexi, but there is something jarring about its form ... it just doesn't sit right with me. In the same way, CB2's skin is made to be soft and human like, which I also think is a bit creepy. That aside, it promises rich interaction with humans because of its ability to actually listen, see and feel (with sensors all over its body), but it does not appear to be as facially expressive as Nexi but it does respond to touch and seems to be reacting to its environment. Its maker admits that people are usually uncomfortable with CB2 until "they get used to it." As a pedagogical agent, I could see it being very useful in higher education and soft skills development, such as interviewing practice or counseling. As well, this could be useful in providing deictic gestures to help facilitate learning example (Baylor, 2007).

Tom Hanks (CGI, Polar Express)

Who wouldn't want a friendly, knowledgeable CGI pedagogical agent in the form of Tom Hanks? Per Baylor, "learners tend to rate attractive agents as deserving more respect, as being more similar to themselves, as who they would like to be like, and as their preferred agent from which to learn" (2007, p 12). As an embodied pedagogical agent, this avatar could be useful for instruction, and granular-level feedback (as with Hal, above) in a TUI dashboard as described in Goldberg & Cannon-Bowers' article (2015). Also, Mayer et al points out that "students learn better when onscreen agents engage in human-like gesturing, movement, eye gaze, and facial expression during instruction" (2012). Of course, this would apply to Nexi and Emily as well.

Emily (Image Metrics)

OK, wow, I had to watch this a few times, because I wasn't actually certain whether "Emily" was the real person or not. We are now squarely in uncanny valley, which I suppose in our times of rampant "fake news" this kind of technology could easily be far more dangerous than constructive. But in a learning environment, Emily could be very useful in simulations or video games (its raison d'etre, after all), provide realistic conversational in simulations for a wide spectrum of different industries. And such a realistic human-like agent could augment historical events and personalities (Veletsianos et al, p 980). Finally, I believe interactions with Emily, Nexi or CB2 illustrate perfectly the kind of heartbeat-elevating, curiosity-raising interest in Veletsianos and Miller's (2008) phenomenological exploration of digital entities in which one gets pulled in and loses oneself in the flow of conversing with these digital agents.

* * * * *

In closing, it seems each of these designs can teeter on a thin line between being attractive, or not; helping with the learning transfer, or adding to cognitive overload; or appealing to the affective side of cognition and being "relatable", versus becoming a distraction. All of these agents prove useful based on the persona effect, which "posits that the mere presence of a lifelike character in an interactive learning environment can have a significant positive effect on the perception of the learning experience" (Goldberg 2015, p 2). 

But the readings this week have also made clear the need for additional research on such boundary limitations, as well as going deeper into which specific kinds of learning tasks and activities might benefit from which specific kind of agent.

References

Baylor, A. L. (2007). Pedagogical agents as a social interface. Educational Technology, 47(1), 11–14.

Goldberg, B., & Cannon-Bowers, J. (2015). Feedback source modality effects on training outcomes in a serious game: Pedagogical agents make a difference. Computers in Human Behavior, 52, 1-11.

Mayer, R. E., & DaPra, C. S. (2012). An embodiment effect in computer-based learning with animated pedagogical agents. Journal of Experimental Psychology: Applied, 18(3), 239-252.

VanLehn, K. (2011). The relative effectiveness of human tutoring, intelligent tutoring systems, and other tutoring systems. Educational Psychologist, 46(4), 197-221. 

Veletsianos, G., & Miller, C. (2008). Conversing with pedagogical agents: A phenomenological exploration of interacting with digital entities. British Journal of Educational Technology, 39 (6), 969-986.