Many adult workers need to keep up with advances in technology to remain relevant in the job market. Adults now need 21st century skills including Computational Thinking. It is challenging for adults to find training opportunities that take into account their limited time, educational, and resource constraints. Our approach provides support to adult learners and their tutors to help them reach their goals. This support can take the form of facilitation messages that suggest possible learning activities, hints on study and time management for learners, and instructional suggestions and alerts for tutors based on current information gleaned from the learner and the interaction. We have collected data with adult learners and their tutors, and are designing an automated facilitator system that can provide tutors and learners with feedback according to their needs.
Existing online programming tools and mentoring programs
do not provide enough support for adult learners looking to improve
the skills necessary to practice the basic concepts of computer
science necessary for these adults to have access to many jobs. [
Systems that teach computer programming have often focused
on syntax-level, basic computer programming concepts, and have
mostly been used with first year undergraduate students [
Working in a general framework for lifelong learning
supported by technology [
Computational thinking is considered a subset of computer
science. CT refers to solving problems by making use of concepts,
methods and processes central to computer science [
According to Brennan and Resnick [
Various types of assessments of CT have been developed and
deployed with middle school students [
We collected data from ten tutors and ten adult learners at a Bay Area Outreach program (N = 20). The adult learners ranged in age from 18-42 years old, 7 males and 3 females, 3 identified themselves as Asian or Asian American, 4 as Black or African American, 2 as Mexican or Mexican American, and 1 with 2 or more than one ethnicity, and 90% had at least some college education. Demographic information also showed that 20% of the learners felt completely confident with writing computer programs, 50% moderately confident, 30% somewhat confident, and 10% not at all confident. Three were employed full time, 2 partial time, 1 was a stay at home parent, and 5 were unemployed.
Before the tutoring session, the adult learners (i.e., novices) completed surveys that allowed us to gather demographic information. They also completed individual differences assessments to aid us in better understanding the populations’ needs. Then, 10 dyads (i.e., learners and tutors) were audio recorded while completing a tutoring session. Next, we interviewed key tutors and experts in areas such as adult learning (i.e., communications with adults at a Bay Area Outreach center) and tutoring adults on computer science topics for non-profit organizations.
After completing the surveys and interactive tutoring sessions, we quantitatively analyzed the surveys, and then proceeded to transcribe the audio recordings of the tutoring session taking great care to code discourse moves in a pedagogically meaningful way. Finally, we interviewed experts to ensure that our findings make sense in a practical way.
Survey results based on this admittedly small sample revealed
that the population is indeed unique in that learners scored higher
than average on individual difference measures such as Grit (M =
4.51, SD = .66) with a maximum score of 5, Growth Mindset (M =
4.52, SD = .07) and Cognitive Flexibility (M = 4.71, SD =.53), both
with maximum scores of 6. The Grit scale measures one's
persistence in the face of failure and passion over a long period of
time [
Next, we analyzed transcripts and discovered 25 dialogic moves corresponding to pedagogical tactics displayed by tutors and learners. For example, moves include “knowledge check,” where the teacher asks the students questions about understanding a topic of discussion; “modeling,” where the teacher types code and the student watches; and “procedural tell,” where the student actively types code and is the primary participant while the tutor scaffolds the students understanding. After discovering these techniques, we were able to condense the 25 discourse moves into an over- arching framework of tutoring for computer science which is beyond the scope of this paper to discuss.
Interviews with tutors and experts provided additional insight. For example, we discovered that tutors try to make the topic relevant to the learner to help motivate him or her (e.g., by working on problems relevant to the interests of the learners).
We also discovered that the tutors tend to have learners work on their own and help them only as needed. However, quite often the tutors spend too much time explaining basic concepts that learners should learn prior to working with the tutor. Also, tutors described situations that learners usually find difficult to overcome and contribute to increasing their risk of dropping out. For example, balancing work and family commitments, missing deadlines, and accumulating overdue assignments. These types of activities could be handled by a facilitator that provides supporting features like the ones shown in Table 1. This facilitator can provide learners with alerts on assignments due soon and support on basic programing concepts and examples. This is particularly important because experts that serve as tutors often are working software engineers and only have a set amount of time to help others.
Rather than implementing a complete intelligent tutoring system, we propose to develop a facilitator that can provide relevant hints and alerts to learners and tutors. Table 1 shows sample supporting features. Tutors often spend a large amount of time explaining basic concepts such as loops, arrays, and so on. Rather than taking up the tutor’s time on these concepts, the facilitator can instead point them to a relevant source.
By providing these hint and alerts, we expect that both adult learners and tutors will engage in more productive sessions. Reducing the burden on the tutor by decreasing the amount of time, cognitive energy and attention necessary to tutor a novice adult, it becomes possible for tutors to focus on other aspects that may require close attention (e.g., providing additional help on challenging topics or planning appropriate activities for learners). Also, tutors may be able to help more adults.
The tool is implemented on top of the ETS Platform for
Collaborative Assessment and Learning (EPCAL) [
Description e.g., sample code, discussion
forums, sample dialogue exchanges, and similar projects
available on the web.
Time and project management
activities (e.g., list of assignments, due dates, and
meetings with the tutor).
Hints about how to deal with common errors and relevant best
practices.
Information about results on assignments and process log data to keep track of learner progress (e.g., alerts on learner missing assignments, problems with particular pieces of content or other risk factors associated with dropping out).
In the above interaction, both the learner and the tutor are sent private messages by the facilitator. The learner has already answered the question and now the facilitator (referred to as “system”) is helping the learner better understand the underpinnings of the problem. This provides supplementary information to the learner and eases the burden on the tutor. Furthermore, the tutor is provided information to aid them in teaching the learner, as we recognize that experts in computer science may not necessary have expertise in pedagogy. Thus, the facilitator should ease the burden on the tutor by aiding the learner and the tutor.
The creation of the facilitator involves collecting large quantities of data from learners and tutors and collecting and analyzing existing data from online forums and through crowdsourcing mechanisms. Machine learning algorithms will be used to find potential domain topics that learners find challenging, resources that have been used provide help (e.g., sample code, sample dialogue exchanges), and areas/situations that require the supporting alerts. This iterative data-driven approach will result in gradually refining the different supporting mechanisms of the facilitator.
By implementing a facilitator rather than a whole intelligent tutoring system, we expect to produce a system that can provide effective support in situations that adult learners and tutors find challenging. The resulting supporting features and the iterative, data-driven approach have the potential to be repurposed in other adult learning domain areas. The proposed approach can be used to continually refine the system (e.g., improving current adding supporting mechanisms) as more data are collected.
The proposed system builds on advances in artificial intelligence to provide the needed support to adult learners and tutors. We expect the types of alerts and hints provided by the system will be effective in improving adult learning of CT concepts and practices while reducing the dropout rate characteristic of these types of programs. Also, we expect the facilitator will help tutors become more effective at providing learning support.
Future work will involve collecting data to evaluate different types of alerts and hints and implementing the components of the facilitator that will keep track of learner and tutor interactions to decide what alerts and hints to provide in which situations.
We expect to use this type of approach to support adult learners with other learning and training needs that they may encounter throughout their lives. The results of this project will help us refine our approach to focus on the alerts and hints that are more effective at supporting adults’ and tutors’ educational needs.
Our thanks to Irvin R. Katz, Jung Aa Moon, Burcu Arslan, and Donald E. Powers for their feedback on a previous version of this paper, as well as Jennifer Lentini and Stephanie Peters for their work on this project.