Abstract:

We offer an analytical narrative of the conceptual development, specification, building, and testing of an open source conversational chatbot designed to guide graduate students in critical thinking and problem solving in the context of a business case. The bot will be used in an online learning environment offering support for asynchronous interactions between co-working dyads of geographically-distributed student learners. The case in question, authored by technical experts but written for a non-technical audience of readers, offers a provocative analysis singling out the practice of “detailing” – pharma sales reps. visiting doctor’s offices to encourage prescription of certain drugs – as causing the opioid crisis in North America in recent years. Graduate students enrolled in this on-line preparatory course are asked to read the case and respond to a clutch of questions on critical thinking fundamentals: causation vs. correlation, the law of large numbers, inductive and deductive reasoning, and discerning “junk” science and spurious claims. Students are also asked to respond to the professional ethical dilemmas raised by the case. The on-line course and the chatbot constitute the iBridge Pre-Orientation Prep, a required non-credit course that all admitted graduate students must complete before commencing their program of study (Fall and Spring). Through cases like the opioid case and others, the iBridge course is designed to equip students in critical thinking skills in order to succeed in demanding courses of graduate study like data science. Developing students’ ability to grapple with ethical dilemmas in data science and technology, and their sensitivity to gender and other kinds of biases using “bias literacy” training are also part of the iBridge Pre-orientation Prep program. The fact that admitted graduate students are predominantly residents of nations outside US before they arrive on the Syracuse University campus (in the US) requires us to offer the iBridge Prep course, with the conversational AI agent serving as an e-learning guide, over the Internet.

The conceptual framework is adapted from the model in Michaels et al., (2007) called APT, Academically Productive Talk. The APT Model condenses reasoning modalities in social settings to three, and all three modes are concerned with the accountability of the learner as a condition for reasoning and reasonable justification in argumentation. We adapted the APT Model with elements from an implementation of an AI bot using the Bazaar framework reported in Dyke et al., (2013).

The conversational agent architecture was detailed to drive the iBridge implementation of the :
(a) the training set and the NLU module (natural language understanding) using the (open source) RASA conversational agent & AI markup language plat-form, and
(b) the schema for the contextualized interactive dialog consisting of two response categories: automatic re-voicing (where the bot responds to the learner with something like, for example, “So what I hear you saying is XXX. Is that right?” (from Dyke et al., ), and academically productive feedback (where the bot responds with positive feedback to signal recognition of the learner’s success approximating the APT’s reasoning and accountability norms).

Keywords:

Critical thinking, Conversational AI Bot, Academically Productive Talk (APT), Re-voicing, Feedback.