1. “Video games are not the enemy, but the best opportunity we have to engage our kids in real learning. ” (Prensky, 2003) “A sine qua non of successful learning is motivation: a motivated learner can’t be stopped.” (Prensky, 2003) How would you like your learning? The influence of reward uncertainty on motivation for learningin computer games. SkeviDemetriou CPLiC
2. Broader project BROAD AIM: To investigate the potential link between prediction error (PE), engagement and learning in laboratory experiments involving adults, and then use this understanding to explore how children engage with learning games involving chance-based uncertainty in more “real world” classroom contexts, seeking to interrelate this understanding with the discourse and social constructionsassociated with such games. Interdisciplinary approach: Education & Neuroscience. Computer game studies on Learning. Game built for the purposes of this project. Classroom & Laboratory based studies. Child & Adult participants. Mixed methods approach (interviews, video recordings, drawings, Electro Dermal Activity measurements, statistical data, etc). Qualitative & Quantitative data. Individually & Collaboratively working participants.
3. Theoretical framework Neuroscientific evidence to show that: Dopaminergic reward activity in the brain (mid brain areas) has been shown to vary with prediction error (Daw et al., 2006). The dopaminergic activation (ventral striatum, NAcc) depends on the magnitude of the prediction error (e.g. Fiorillo et al., 2003; Schultz, 2006). Neuroscientific evidence to show that dopaminergic activity in the brain aids memory formation and thus factual learning – Two notions: Direct impact: Uncertain rewards may promote memory formation through the dopamine release in the brain area called the hippocampus (Adcock, 2006; Callan & Schweighofer, 2008) Indirect impact: The uncertain reward - memory link is mediated by attention (i.e. Loftus, 1972, Muzzio et al., 2009). Dopamine is a neurotransmitter produced in the midbrain and transferred to cortical and subcortical regions (Treber et al, 2005). It is released at the synapse between 2 neurons and allows the transfer of impulse (information) . It is the transmitter used in these specific parts of the brain (4 main dopaminergic pathways).
4. 4B. How are discourse and constructions influenced by whether the artificial opponent is matched in terms of gaming skill or academic ability? Quasi-Experimental Study 4A. What types of discourse and constructions are associated with competitive learning games involving chance-based uncertainty and an artificial competitor in the classroom, and how might these be interrelated with our biological understanding? 1. How does prediction error, in learning games employing chance-based uncertainty, influence memory in adults? RESEARCH QUESTIONS (RQs) 3B. How is this learning influenced by whether the artificial opponent is matched in terms of gaming skill or academic ability? 2. How might prediction error, in learning games employing chance-based uncertainty, influence emotional engagement as measured by EDA? 3A. How does prediction error, in a computer-based learning game employing chance-based uncertainty and an artificial competitor, influence children’s memory?
5. Quasi-Experimental Study Aim: To explore how reward uncertainty and in particular, positive prediction error (PPE), is related to fundamental learning processes (i.e. orientation of attention, memory encoding and recall). In Specific: To identify instances when material was learnt and look for a relationship between them and the size of the PE in whatever game event had just previously occurred.
17. Definition of learning in the computer game. PEENC Question presented F Correct answer revealed PEREC Question presented F T Unsuccessful learning Successful learning (PE =Box’s_Score_Now –Box’s_Score_Last sampled)
18. Methodology 1. How does prediction error, in learning games employing chance-based uncertainty, influence memory in adults? Sample:16 adults (7 males and 9 females). All participants were working individually. Methods - Measures: Pre and Post tests Recordings of participants’ choices in the game (PPE & learning) Videorecording Variables: Two continuous pseudo dependent variables (they were an alleged cause rather than effect): Prediction error at encoding (PEENC) and recall (PEREC). One pseudo independent variable: Learning (2 levels: successful/learning-SL and unsuccessful/non-learning-UL). Hypotheses: 1 (for RQ1): In a learning computer game employing chance-based uncertainty, for encoding, prediction errors would be higher prior to successful than for unsuccessful learning. (PEENC) Successful learning > (PEENC) Unsuccessful learning 2 (for RQ1): In a learning computer game employing chance-based uncertainty, for recall, prediction errors would be higher prior to successful than unsuccessful learning. (PEREC) Successful learning > (PEREC) Unsuccessful learning
19. Results & Discussion On average participants scored significantly higher in the post test (M = 19.25, SE = .78) than in the pre test [M = 8.75, SE = .36, t (15) = -16.11, p < .0005, r = .97]. In response to RQ1 and Hypotheses 1&2: On average, for encoding, PE for successful learning (M = 17.41, SE = 1.35) was not significantly higher than PE for unsuccessful learning (M = 15.27, SE = 1.31) even though the trend was in the direction hypothesised [t(15) = 1.71, 1-tailed: p = .054, 2-tailed: p = 1.1]. Discussion:- This was expecting the possible attention effect of the PE to survive about 18-24 seconds, (i.e. from when participants had entered their question answer and indicated their confidence rating, to when they received feedback). However, literature suggests that dopaminergic reward effects are quite short-lived – only a few seconds (Bogacz et al., 2007). - The encoding was also occurring in the negative context of being told their answer was wrong.
20. Results & Discussion On average, for recall, PE for successful learning (M = 20.10, SE = 1.36) was significantly higher than PE for unsuccessful learning [M = 17.55, SE = 1.35, t(15) = 3.51, 1-tailed & 2-tailed: p < .005, r = .67]. Positive prediction error (PPE) is linked to successful learning. Happy surprise is linked to successful learning.
21. Conclusions Learning to be potentially influenced by positive prediction error (PPE). In line with literature suggesting that positive prediction error triggers memory and factual learning formation. PPE as a “memory enhancer”. Reward uncertainty increases dopamine release in the brain and therefore attention and motivation to engage (e.g. Fiorillo et al., 2003). This was examined in the context of an educational computer game. Potential for Education. This study was followed up with studies using this computer game in classroom settings. Statistically significant results. This could help device educational computer games that could promote motivation and learning. Design “learning provoking” situations. Neurocomputational modelling. Gaming as a computer based activity, is known for its potential to promote learning even in formal educational settings when embedded properly and with awareness of its benefits and constraints. Technology is not a “panacea”. It is a tool. This study was followed up with studies on uncertainty-involving not computer based gaming. Very encouraging results. Whether involving computers or not, the gaming element itself, obtains powerful dynamics in enhancing individuals’ motivation and is potentially powerful in making learning derived from such instruction efficient (e.g. Randel et al., 1992; Whitehall, & McDonald, 1993; Ricci et al., 1996). Playing is, above all, “a privileged learning experience” (Rosas et al., 2003).