Praise in Computer-Mediated Learning

Grace Fuisz, Psychology, '19

As computer-mediated teaching becomes more and more prevalent, it is important to account for the difference in social experience for learners when their teacher is an avatar or AI system. This essay describes some of the differences in learning in a mediated versus unmediated context, providing suggestions for creating successful
digital instructors and future directions for research. 

         In Punished by Rewards, Alfie Kohn criticizes conditional academic praise, claiming that social rewards can be as enduring and impactful as tangible rewards when it comes to decreasing intrinsic motivation. We want our teachers, our parents, our authority figures to like us, and social rewards exacerbate our need to please others by verbalizing when we have performed adequately. There is essentially no difference in receiving an A or a ‘check’ on an essay, and receiving a confident “good job!” from the teacher. 

Source: https://www.inc.com/gordon-tredgold/the-thing-that-many-people-get-wrong-about-giving-praise.html

            Either way, we learn that our performance contributes to our value and being
 socially accepted.This, of course, does not mean that teachers should not interact with their students or avoid feedback altogether. Social interactions, both with peers and teachers, are incredibly helpful for efficient learning. Collaboration with peers and comfortable, social classroom environments foster confident and deep learning. Respect, perceptions of credibility and competency, and feelings of trust towards a teacher all the quality of a student’s learning. As grade school students are handed iPads in the classroom and computer-mediated learning environments are integrated into nation-wide curriculum, one might wonder how the social aspects of education will change.

Social computers

            There is overwhelming evidence that computers are treated as social beings. People demonstrate the same in-group/out-group effects with computers as they do with humans, rating computer agents of their same ethnic identity as more attractive, trustworthy, persuasive, and intelligent than a computer of a different race (Nass & Moon, 2000). Just like humans, we are able to recognize the personality of a computer and the similarity of a computer’s personality to our own. Submissive people rate a submissive computer as friendlier and more competent than dominant people do, representing another in-group bias (Nass, Moon, Fogg, Reeves, & Dryer, 1995). Even a minimal manipulation (assigning a participant and computer to the same arbitrary team) is enough to signal human-computer in-group bias behaviors (Nass & Moon, 2000). 

            We apply gender stereotypes to computer agents, rating computers with male-voices more competent with regards to stereotypically masculine topics (computers, technology) than female-voices; female-voiced computers are rated more competent on feminine topics such as love and relationships. Furthermore, when a computer with a male-voice praised another computer during an evaluation, the evaluated computer was rated more competent than when the same praise was given by a female-voiced computer (Nass & Moon, 2000). This paradigm, wherein people mindlessly react to computers as social beings, has been termed Computers as Social Actors (CASA) (Nass & Moon, 2000). It seems that it does not matter if participants are aware or not that the computer is not human; computers are still perceived as social entities and receive the same patterns of cooperation, trust, and socially desirable behavior that is expected from human-human interactions. When participants were asked what they wanted to watch in a waiting room by a computer agent, they were more likely to pick the socially desirable choice (watch a documentary) compared to an action movie or television, even though the social influencer was a computer (Kim & Baylor, 2006).

Computers as teachers

Source: https://www.newsroom.co.nz/2018/08/22/203646/digital-teacher-in-kiwi-schools

,             The image of a young child playing on an iPad in a stroller, watching videos on the subway, or playing on his mother’s iPhone is all too familiar in the age of the internet and smart phones. Many of the apps marketed for children’s entertainment promise learning in the form of strategy, mathematics practice, and language exposure. People interact with computers as social beings, which implies that the importance of social learning environments, trust and respect for computers, and the harm of social rewards should apply to computer-human interactions and learning, as well. So how do we create effective computer teachers?

Multimedia representations used for teaching are commonly referred to as pedagogical agents (Atkinson, Mayer, & Merrill, 2005), and range from animated text, to video recordings of lectures, to avatars and artificial intelligence. Current applications of computers-as-teachers use artificial intelligence for simple content delivery, and feedback, mainly in higher education (Popenci & Kerr, 2017). Essentially, it seems that the most effective computer teachers are similar to human teachers, and are used to augment (rather than replace) traditional teaching methods. College students are more motivated to learn and have more favorable ratings of an artificial intelligence (AI) tutor with an older voice than of an agent with a young voice. This finding suggests that students recognize the AI in the stereotypical professor role, associating age as a sign of authority and experience (Edwards, Edwards, Stoll, Lin, & Massey, 2018).

Pedagogical agents can personalize interactions with content, which motivate and support the learner more than stagnant material like a textbook. The addition of a pedagogical agent is more important to the learner’s subjective experience and enjoyment than to the actual performance of the student, although there is evidence that pedagogical agents can increase transfer success (Krämer &Bente, 2010).

           The visual presentation of a pedagogical agent is important for successful learning. Multimedia instruction is more effective for increasing test scores when it includes social cues (a video of a speaking professor instead of a stagnant relevant image) (Töpper, Glaser, & Schwan, 2014). Compared to on-screen text, an animated, speaking pedagogical agent displaying the same content increased performance on transfer tests and interest ratings in material (Moreno, Mayer, Spires, & Lester, 2001). Similarly, an animated image was rated more engaging, person-like, credible, and instructor-like compared to a static image, although animacy did not affect performance directly). An animated image was, surprisingly, not significantly more person-like than on-screen text without an accompanying image, although it was more person-like than a static image (Baylor & Ryu, 2003). Perhaps this effect is due to our current familiarity with texting and e-mail, which normalizes on-screen text as a representation of live human communication. How the avatar is represented visually seems to only matter when the instructor is demonstrating a physical skill. If the pedagogical agent is using observational learning to demonstrate a physical movement, it is most effective if it is represented with a human-like body (Krämer & Bente, 2010).

            As previously mentioned in the gender stereotyping of computer agents, a computer’s voice shapes our perceptions of competency, and credibility. Consistent with the importance of animacy to learning, computer voices are most effective in teaching when they are human-like (as opposed to off-the-shelf machine/robotic voices). In a study by Atkinson, Mayer, and Merrill (2005), students performed better on practice problems, and had higher rates of both near and far transfer when learning from a pedagogical agent with a human-like voice, compared to a machine voice. This effect was consistent, even while accounting for the effect of novelty and learning contexts – the study was replicated in a high school classroom, in addition to a lab setting.

            The implication of the CASA paradigm is that social rewards coming from a computer would be as detrimental and complicated as those coming from a human. Praise from computers, then, should come only in the same form as Kohn suggests. Computer-human praise should focus on what people do, as opposed to quality, to draw focus away from performance goals praise should be specific; praise should be genuine (Kohn). An animated pedagogical agent with a thumbs-up and a “good job!” speech bubble is counter-productive to learning and, in theory, decreases intrinsic motivation as much as a grade, or getting a sticker for good behavior. Instead, animated agents could focus on learning achievements: “you used a new word,” “you tried a new strategy.” With artificial intelligence, specific and unique praise may be more achievable in the near future, as it may be difficult to program a simple avatar to provide unconditional and genuine support.   

Future directions

            Future research should explore the three conditions of harmless praise. As defined by Kohn, praise is less harmful than more tangible rewards when 1) praise is less salient 2) praise is less controlling, and 3) praise is not promised in advance. This effect could easily be applied to pedagogical agents by testing intrinsic motivation after a pedagogical agent violates any or all of these rules. For example, to violate the controlling praise rule, a pedagogical agent could socially reward a learner with: “Congratulations! You met the standard score for students of your age.” This statement would be controlling as it sets an expectation for the learner, so that if they do not receive similar praise going forward they know they have failed or disappointed the agent. It also violates a socially supportive learning environment by comparing the learner to other students, so that the student feels the need to compete with their peers for success. This aspect of cooperation could also be studied in digital learning environments by testing the effectiveness of computer-mediated learning in cooperative versus competitive peer environments.

References

Atkinson, R.K., Mayer, R.E., & Merrill, M. M. (2005). Fostering social agency in 

            multi-media learning: Examining the impact of an animated agent’s    

            voice. Contemporary Educational Psychology, 30, 117-139. 

            https://doi.org/10.1016/j.cedpsych.2004.07.001.

Baylor, A. L., Ryu, J. (2003). The effects of image and animation in enhancing 

            pedagogical agent persona. Journal of Educational Computing                         Research, 28(4), 373-394.

Edwards, C., Edwards, A., Stoll, B., Lin, X., & Massey, N. (2018). Evaluations 

            of an artificial intelligence instructor’s voice: Social Identity Theory 

            in human-robot interactions. Computers in 

            Human Behavior, in press. https://doi.org/10.101g/j.chb.2018.08.027

Krämer, N. C. & Bente, G. (2010). Personalizing e-learning: the social effects             of pedagogical agents. Educational Psychology Review, 22 (1), 71-87.

Kohn, A. (1993). Punished by rewards: The trouble with gold stars, incentive 

            plans, A’s, praise, and other bribes. Boston: Houghton Mifflin Co.

Moreno, R., Mayer, R. E., Spires, H. A., & Lester, J. C. (2001). The case for 
            social agency in computer-based teaching: do students learn more 

            deeply when they interact with animated pedagogical agents? 

            Cognition and Instruction, 19(2), 177-213.

Nass, C., & Moon, Y. (2000). Machines and mindlessness: Social responses 
            to computers. Journal of Social Issues, 56, 81-103. 

            https://doi.org/10.1111/0022-4537.00153

Nass, C., Moon, Y., Fogg, B. J., Reeves, B., & Dryer, D. C. (1995). Can 
            computer personalities be human personalities? International 

            Journal of Human-Computer Studies, 43, 223-239.

Popenici, S. A. D. & Kerr, S. (2017). Exploring the impact of artificial 
            intelligence on teaching and learning in higher education. 
            Research and Practice in Technology Enhanced Learning,12 (22). 
            https:/doi.org/10.1186/s41039-017-0062-8.

Sproull, L., Subramani, M., Kiesler, S., Walker, J. H., & Waters, K. (1996). 
            When the interface is a face. Human Computer Interaction, 11(2), 
            97-124.

Töpper, J., Glaser, M.,& Schwan, S. (2014). Extending social cue based 
            principles of multimedia learning beyond their immediate effects. 
            Learning and Instruction, 29, 10-20.