Introducing your future teacher: Artificial Intelligence (AI)
It seems inevitable that the ongoing pandemic will impact the education of kids around the world. Especially in countries lacking in advanced tech solutions, the consequences of closed schools could be devastating. UNESCO points out a number of concerns: first and foremost, parents are unprepared for homeschooling and let’s face it, it’s impossible to do it well while also working. It’s insanely expensive, demanding, isolating for kids, harmful for health, and last but not least – they learn less. Sitting in front of the screen isn’t as much fun, and you can’t even get proper, personalized feedback.
If you’d ask investors about industries worth looking into right now, chances are they will mention educational tech. The opportunities go way beyond your good old Coursera and Udemy – tech spiced up with AI can adjust to your learning style and deliver what you need.
It sounds counterintuitive to connect technology and personal(ized) education – what about the relationship between teachers and students, the spontaneousness, the “humanness”? Yes, that will never lose its importance, but artificial intelligence can help teachers understand their students, set personal biases aside, and get accurate, objective insights into their performance. How different could the past year of schooling look like if we widely implemented AI-fueled teaching systems? We’d guess that while it can’t replace the school – a living organism, we’d also see less bored kids with a huge gap in their education.
So what are some of the education areas that we can transform with a helping hand from a friendly, objective, informed AI?
A comprehensive report by JRC suggests that AI can play a significant role in special needs education. The opportunities represent a groundbreaking development for these kids, mainly because child-robot interaction can unlock new forms of diagnostics and special needs educational applications. To illustrate, Swedish company Lexplore can detect dyslexia by tracking the reader’s eye movements. Otsimo helps nonverbal children in various languages while Gnous’s video games assess and create a map of brain cognitive abilities.
On top of that, AI can pinpoint what these kids need, which can be a highly advanced task. It’s also tremendously helpful for parents struggling to grasp a highly sophisticated approach to education.
Testing & Assessment
JRC also analyzed the use of AI for automatic test generation and assessment. The problem with grading in testing is that standardized tests don’t capture students’ skills and knowledge perfectly. Unfortunately, a teacher’s personalized assessment can be far from objective and marked by bias, even if it’s unconscious and unintended. And of course: unless automated, it’s time-consuming and prone to human error.
If an AI system has sufficient data on students and their peers and can recognize their patterns, grading homework is perfectly realistic. The study mentions that AI can take it one step further: diagnosing student attention, emotion, and even conversation dynamics in computer-supported learning. What does that mean? We will be able to pick the students who fit well within the same study group or a task and monitor if someone is likely to drop out, and they can get help to prevent that from happening.
Joel Hellermark, the Founder of Sana Labs, a Swedish company focused on developing and applying AI for learning, points out no two learners are the same and estimates that over 60 percent of them find misalignment between learnings and their skill gaps. AI can identify this and offer an individual learning plan for each student. “With adaptive assessment, learning, and review, the learning assistant personalizes the path to the individual needs of the learner — surfacing just what they need, right when they need it,” Joel explains, adding that achieving mastery of the learning material takes half as long. The newly acquired knowledge sticks over 3x more.
However, as he mentioned in the interview for 52 Insights, human teachers aren’t going anywhere anytime soon. “I think teachers are going to be one of the last jobs to get automated,” he says. Instead, they will become empowered by AI and spend less time on time-consuming tasks that can be automated. According to Joel, that could look like a classroom full of students who have already studied the basics. Their teacher would know who is at which level and which areas need help. That will look different for each student, but thanks to AI, it will be manageable. “The class will be spent mostly on project-based learning,” Joel concludes.
This can be a massive game-changer for learning new languages, as it depends on the availability of preferably native speakers. That’s not a given in all corners of the world, and kids end up learning from inadequate sources and acquire wrong accents. Emotech, which has introduced the first multimodal AI Teaching Assistant solution, has leveraged the current situation and has been helping children worldwide access quality language education. “At this moment, technology is quite important for the efficiency of education – because you don’t have the limitation of whether you can go out to another country or a time-zone issue,” said Chelsea Chen, co-founder of Emotech, in her talk at Reflect Festival. “Through AI, you can get education and information more efficiently. Compared with one teacher facing 20-30 students, the computer doesn’t have any limitations – there can be thousands of students, while all their feedback will be tracked and provided to the teachers,” she adds.
All these opportunities bring so much hope for a brighter future for kids, as well as for the efficient reskilling of adults adjusting to tumultuous changes in workplaces. But the fear is there as well, and it’s perfectly reasonable.
A vast amount of data is a prerequisite for effective AI, and that’s where ethical questions arise. As JRC points out, student behavior needs to be monitored and tracked, but it should in no way be intrusive and cross ethical lines. You also need this data to represent everyone fairly to produce adequate results: as Hellermark points out, if you train on datasets of white, middle-aged men, that’s only acceptable if the system is intended for this demographic. If a diverse group will use it, the data-set needs to be varied as well, and that’s certainly a challenge to be tackled.