Blended Learning: The Rise of Machine Learners
In 2010, only 54% students from grades 9-11 in the Dallas Independent School District (DISD) passed the math portion of the assessment test. In contrast, ~80% students from grades 3-6 met the minimum standard on Texas’s assessment test. That means nearly half these students (~12000 in DISD) fell behind on basic math skills by high school.[1] Ideally, teachers could identify and prioritize the diverse needs of students, especially those who need intervention. However, resource constraints in school districts can seriously limit the personalized attention that students receive.
In such cases, introduction of “intelligent” tutoring systems that use machine learning (ML) to gauge pre-existing knowledge, provide personalized tutoring and analyze performance, can dramatically improve learning outcomes. Reasoning Minds (RM), an adaptive video game that teaches math, does exactly that – A virtual “genie” guides middle-schoolers through personalized math lessons where students progress through levels only after they’ve mastered certain concepts.[2] Meanwhile, it alerts teachers if a student is stuck, allowing them to monitor individual progress and effectively intervene when needed instead of having to simultaneously field wide-ranging questions. Furthermore, students can access this learning platform (provided internet availability) even in the absence of teachers.
Figure 1: Sample guided study screen
Given the promise of ML solutions, DISD decided to pilot RM in the short-term. ~2500 third and fourth graders were enrolled onto the platform in 2009 and another ~1000 in 2010. As the program demonstrated success, DISD expanded it to all grade 2 students in 2011, and to grade 3 students in 2012. [3] By 2016, ~33,000 students were enrolled for RM, adoption and usage continuing to flourish. Researchers measured 89% time on-task in RM when implemented with fidelity, potentially leading to up to 40 extra hours of instruction in a year.[4] Hence, it is no surprise that DISD teachers revealed strong support for RM, with 86% of supported and 71% of non-supported teachers stating RM benefits students, and 80% and 62% of supported and non-supported teachers willing to recommend it to others. [5]
Such ML-based tools that can “personalize” learning in classrooms (and outside) are expected to grow to a $6B+ world-wide market by 2024, with U.S. K-12 classrooms and consumers driving ~20% of that growth.[6] Tools like RM that customize content to individual learning patterns will continue to enhance student engagement and learning outcomes. In addition to the contribution to students’ learning, ML offers significant advantages for teachers too. It can reduce their workload, allowing them time to focus on students one-on-one or on more complex problems. ML tools also provide teachers (and parents) with real-time feedback on students’ progress, allowing them to intervene timely instead of waiting until periodic exams, when it might be too late.
Figure 2: Expected growth of machine learning application in education
For DISD specifically, as it extends usage of ML in the near-term, it should first assess the culture and technical readiness of schools before integrating these tools into traditional class formats and curricula. It must invest in technology infrastructure and standardize data sets (e.g. grade records) throughout the network to facilitate quicker implementation and adoption. While the technology adoption is at its nascent stage, DISD can also codify learnings from its RM programs thus far, defining best practices on how to efficiently leverage these tools for maximum benefit. Finally, DISD can start establishing strict security protocols in advance, as data privacy concerns are bound to arise.
Longer term, as ML undertakes more responsibilities away from teachers, like time-consuming routine tasks of grading and record keeping, teachers’ roles will have to evolve. Teachers will have to be trained in new skills, primarily developing a sophisticated understanding of technological tools so that they can utilize them effectively. To that effect, DISD should also start involving teachers in co-designing ML tools so that their proven pedagogical techniques can be embedded into these intelligent tutoring systems. This will also help DISD address teachers’ fears of being replaced by technology and facilitate better adoption. On a more strategic level, as classrooms move towards a blended learning model – combining human and artificial intelligence, DISD will need to define a) the right balance between the two components, and b) the pace and stage(s) at which it introduces virtual teaching agents into students’ lives.
For schools to answer the above-mentioned questions, it will be critical to evaluate the design and impact of blended learning more holistically. Hence, I’m curious to hear opinions on (1) the potential negative impacts, if any, of using ML tools on student learning, and (2) what if we flipped ML application in education on its head? Instead of machines learning how children think/learn, what if these tools could proactively shape how children think – urging their curiosity and expanding their imagination?
(785 words)
References
[1] Dallas News (Dec 2011). “Dallas ISD to expand computer-based math program”– Melissa Repko https://www.dallasnews.com/news/education/2011/12/19/dallas-isd-to-expand-computer-based-math-program. Accessed on November 12, 2018
[2] www.reasoningmind.org. Accessed on November 11, 2018
[3] Victor Kostyuk, Leigh Mingle, Steven Gaudino, Diedre Douglas (2015). “Reasoning Mind DISD 2015 Report”. https://www.reasoningmind.org/rmwp/wp-content/uploads/sites/4/2015/12/2014-2015-Dallas-ISD-Report.pdf. Accessed on November 11, 2018
[4] Jaclyn Ocumpaugh, Ryan S.J.d. Baker, Steven Gaudino, Matthew J. Labrum, Travis Dezendorf (2011). “Field Observations of Engagement in Reasoning Mind”, Springer-Verlag Berlin Heidelberg 2011. Accessed on November 12, 2018
[5] Bush, J. & Kim, M. (2015). “Evaluation of Reasoning Mind: Final Report 2014–15”. Accessed on November 13, 2018
[6] Global Market Insights report (2018). “Artificial Intelligence (AI) in Education Market worth over $6bn by 2024”. https://www.gminsights.com/pressrelease/artificial-intelligence-ai-in-education-market. Accessed on November 10, 2018
Note: Supported teachers refers to those instructors who received additional training and implementation support from Reasoning Minds specialists
This was a very engaging article and a completely unconventional application to deploy ML. My only reservation with this is the fact that traditional ML systems need a ‘training data-set’ that is robust and exhaustive (to whatever extent possible) to learn benchmarks against which to then base difficulty levels and other stratification. In the context of this essay, this seems possible by using RM repeatedly across several student populations. But that could be because the core subject matter itself is fairly quantifiable (math!). Will this application be as effective for students studying history or art? And if it IS effective, then will more families opt for this mode of education, at the cost of the social experience that schools offer? For me these are the two possible negatives for an ML application in education.
On the flip side, I can see ML being deployed to study a variety of languages, where a student’s understanding of Spanish or Hindi can be measured against a benchmark, allowing her to then adapt the pace of her learning in sync with her imagination and creativity.
Education is such a space that is very behind given the technological advances we have made in the last decade so this piece is very interesting. I specially like your last question about helping shape the way students think. I think this can be an advantage as well as a risk. As you mentioned, curiosity is extremely important especially as we try and develop students into critical and analytical thinkers so this could be a huge benefit if all machines knew how to do this versus having to spend time training teachers doing when there is a resource constraint. The danger with standardized teaching (and therefore programable teaching essentially) is that the way a program such as RM can be dangerous if it is programmed with incorrect data/underlying assumptions and all students are using it. I think with RM we are heading in the right direction so we have to take into account the possible dangers as well.