How AI Can Reach Its Potential in Education
Last year, I had the great opportunity to speak about the use of AI in education at LearnLaunch’s Across Boundaries Conference in Boston. During the Q&A session, some educators conveyed their mistrust and trepidations around AI. They had an array of concerns, ranging from the accuracy of certain solutions to minimizing the role of teachers in the classroom. Now, a year later, we are seeing various industries accelerating their efforts in AI. The retail sector is creating operational efficiencies through robotics and physical automation. The healthcare industry is leveraging computer vision to detect visiting hospital patients who may be exhibiting symptoms of COVID-19. In education, we are seeing more online and distant learning, but what will AI’s role in this space be? How will it redefine the role of educators? To be frank, I don’t see a future where robots will be replacing teachers. If there are solutions that strive to automate teaching, then I think these companies are missing the point. The benefit of AI lies in its ability to support educators and relieve them of arduous tasks so that they can focus on the most important aspect of their job—teaching. If AI is truly going to live up to its promise and have large-scale impact in education, then we must have solutions that are traceable and transparent, data-light, and capable of solving high-value problems.
Ensure There is Transparency
In order for educators to trust an AI solution, it must be evident how the machine came to the decision that it did. Unfortunately, some AI technologies are black boxes. There has been a lot of excitement around personalized learning, which, broadly speaking, is technology that can adapt to each student’s needs. If educators are to have trust in these systems, then they must be traceable and transparent. If not, there can be unfortunate consequences.
For example, if a platform reveals that a student is a low performer, then an instructor may give this student an easier workload to help them catch up to their peers. Not understanding how the machine came to this specific conclusion is problematic; the worst-case scenario is that the initial diagnosis was wrong, and the student is actually a high performer, so giving them a lighter work load would hinder their academic development. This is a stark example where a machine can inhibit a student reaching their true potential. In the end, educators will not and should not blindly trust machines. They will demand the need for traceability and transparency, especially when the stakes are so high.
Be Data Light
For the last few years, “big data” has been a buzzword in the AI space. The basis for this is that the more data a machine has, the better it can make decisions, and ultimately predict future outcomes—advances in technology have made processing vast data sets faster and cheaper, thus enabling big data solutions. While these solutions may be useful in industries that have large troves of data that require in-depth analysis (e.g. finance), this is not necessarily the case in the education space. When training data sets, representative samples are required for accurate and generalizable results. Unfortunately, we have seen in education that companies have used “flawed” training data sets that can hurt specific groups. Additionally, school systems and institutions may have a limited amount of data in their systems, and won’t have the bandwidth or resources to generate the large amounts of data that machines require to be effective.
For example, grading is a time-consuming and sometimes grueling process for educators. For traditional machine learning solutions to automate the grading process, they require hundreds (and sometimes thousands) of graded assignments to sufficiently train themselves. Most educators would not have the quantity of data that these solutions would need, but would also lack the quality of a representative training set. In brick-and-mortar classrooms, one particular assignment might be administered to 30 students once a semester—if this same assignment is used unchanged over 30 years (an unrealistic assumption), that’s only 1,800 total responses. In addition to this relatively small data set, how do you determine what a representative student sample looks like from such a longitudinal population? This is why it’s critically important that AI systems do not require their engines to be trained on large amounts of data. If systems are data-light, it will create less of a burden on educators.
Solve High-Value Problems
To truly create transformative impact in the classroom, AI solutions must solve high-value problems. While there are various areas to focus on, I would like to address two in particular. The first is grading. With the rise of MOOCs and online courses, we are seeing thousands, and in some cases, millions of students in a singe course. Educators do not have the bandwidth to grade and give feedback to these students individually; simultaneously, institutions don’t have the resources to hire an army of graders. The inclination is just to rely on multiple choice questions, but open-response assessments and essays are critical in determining if a student truly grasps a subject. AI can help alleviate the headaches of the grading process by being able to rapidly understand a rubric and assess millions of assignments against it in a short amount of time. As mentioned earlier, traditional machine learning solutions would have limitations when it comes to grading, but there are new and emerging technologies that can leverage machines that are entirely based on computational linguistics.
The second area that AI can help educators is by amassing knowledge on demand. While many educators have spent years, or even decades, becoming thought-leaders in their respective fields, information is increasing at a rapid pace. It is nearly impossible for educators to stay up to date on everything that’s happening in their industries. AI, specifically natural language processing (NLP), has the power to analyze large amounts of information and synthesize it into digestible portions for educators to consume. Giving educators the ability to have this knowledge on demand will create a more vibrant learning environment where students are constantly up to date on the world around them.
The COVID-19 crisis is accelerating change in every industry. AI will play a key role in the education sector’s transformation, but all of us have the power to decide what that process will look like and how many lives will be impacted. Will we adopt black box technologies that will benefit a few while hurting many? Or will we embrace traceable and transparent solutions? When it comes to data, will AI only work for a limited amount of schools that have large troves of data, or will there be data-light solutions that can empower even the most under-resourced schools? When it comes to empowering educators, will AI be used to make small strides in the classrooms, or will it tackle some of the most daunting challenges? This global pandemic has created even more inequities in education. When the world emerges from this crisis, we have a lot of work to do address these systemic issues. We must ensure that we fully leverage AI to create large-scale and inclusive impact for every student in the world.
– Emil Kuruvilla, Vice President at Intelligent Machines Lab