Online learning — ripe for picking. Part 2 — AI
This is a part of 2 of a 3 part series, part 1 can be found here.
As I tried to argue in Part 1, I think the technological advancements and the unexpected fallout of the COVID lockdowns, has opened up an opportunity to democratize good education via innovative online learning.
I was and am a great fan of the work of The Great Courses company (previously known as The Teaching Company). I spent a decade of my life, devouring their content. They produced university-level audio and video courses, taught by the best professors in the world. I left like a discovered a gold mine, and bought (and pirated , sorry) all I could find. Several gigabytes of precious audio content. Biking an hour to and from work allowed for a lot of listening time. I studied ALL of history, art, religion, philosophy that they had, from many different perspectives, from many different professors. It had an absolutely awesome, formative time.
Now you can access recorded lectures from many universities, which is absolutely great.
But what about K-12 learning? Online accredited schools, that offer recorded teacher lead instructions are growing and provide an excellent alternative to traditional schooling. Students can access on-demand videos that explain the topic, and can follow a curriculum of carefully planned out exercises. However, as mentioned in part 1, the problem of marking the exercises and tests has not yet been solved. Schools provide answers keys, but parents have to have the time (and a bit of know-how) to take advantage of those. You can pay for marking but that is expensive — some person needs to perform this action. This is not a viable option for most parents.
This is where Artificial Intelligence can help. Tools like ixl.com, which ask and evaluate practice questions are good, but only for simple problems. It is easy to say that 2+2=5 is wrong and offer an explanation, but simplifying polynomials like 28 + x2 + 7y2 − 2xy + 3x2 + 4 + 10xy− 6y2 takes many steps and is a harder problem to evaluate and give feedback on. Plus, evaluating such a problem takes pen and paper and entering the solution back to some program, is an unnatural steps that will not only extend the work unnecessarily, but will likely be skipped altogether — it is like double checking your work, we have to force ourselves to do it. But what if a program could read the solution, via a camera or photo, right off the paper on which the student worked out the solution? Photomath, a company that recently raised $23 million dollars and boasts a download count of 220 million, attempts to do just that. I am not impressed with the app’s “explain the solution” part but it is a step in the right direction, and investors certainly thinks so. This app is a clear proof that parents are desperate for some help with their kid’s homework. I think AI system could interpret a written problem, point out mistakes, and point to a WELL PRODUCED video that explains how to solve the problem. Now that would be something!
What if that AI could figure out (after a repeated use for example) the root of the difficulty the student is having? What if, when asking to solve long division, the system could figure out that the root of the problem is a lack of knowledge of the multiplication table? It is a common deficiency. What if that program could then help the child break down the long division problem into steps and aid the student to solve the problem? The system could show the multiplication table, perhaps even highlighting the right cell at first, pointing the exact multiplication that is needed at this step. As student’s familiarly of the multiplication table increased, the system could gradually back off and offer fewer and fewer hints. System could offer suggestion, like the area of study that the student should focus on, like practicing the multiplication table. Could this AI system become the child’s private tutor? I think so… How much would that cost? Pennies, at most… I engineer software systems for a living, I know how much computational power for such system costs.
What impact would that have on the children from households that cannot afford tutors, let alone good ones? If an inscription on my tombstone read “helped millions of disadvantaged kids get good primary education”, I’d die a satisfied man…
What if AI could interpret speech and thus be able to ask and interpret answers to math question, even something as simple as addition and multiplication? Mimicking a loving, patient parent, willing to spend countless hours practicing multiplication with their child? Wouldn’t that be awesome…Think of the impact on education options for the disabled…
Failure to learn basic arithmetic has disasters consequences on later math learning. I focus on math because I am a mathematician by training and a teacher by heart, but English is a great candidate for AI too. Teaching sentence structure is a perfect example but I leave it to linguist to explore this area. I almost failed senior year English because for my project, I compared and contrasted Lord of the Rings to some other fantasy book — Miss Meredith despised the genre.
I am a software engineer. A programmer. A geek. I cannot judge if my educational ideas are on the money, but if they are, I do know that such software can be built; I know that what I described here is possible with today’s technology. I believe that AI’s ability to comprehend speech, ability to recognize handwriting and the ability to adjust learning based on results, could eliminate or lessen the need for a teacher/tutor/parent to review student’s work in a online, self paced learning environment. I think this is the only hope to bring good education to all. Hope to be a part of such an adventure one day.
In part 3, I will explore gamification of learning.
