I have always believed that my habit of sleeping long was a bad one. Instead, I was intrigued by stories of famous and successful people, who optimized their sleeping time to be more productive during the day. Now, I am certain that most of these stories are only legends, neither based on reality nor helpful inspirations for my own sleeping habits. In case you are still interested in sleeping myths of famous personalities such as Donald Trump or Albert Einstein, check out the best-of infographic at DisturbMeNot.

The starting point of my mind change has been the book Why We Sleep by Matthew Walker. He is a professor of neuroscience and psychology at UC Berkeley, contributing to highest-profile research on the impact of sleep on human health. Walker likes to compare sleep to a revolutionary drug that makes us feel happier, smarter, and healthier. If such a drug existed, would we not all want to take it? What particularly struck me as a student was the impact sleep has on our cognitive performance. Have the painful late-night learning sessions before an exam been useless, in fact, counterproductive for me?

Sleep is more than the absence of wakefulness

Humans do not just sleep but cycle through two completely different types of sleep. These types are called rapid eye movement (REM) and non-rapid eye movement (NREM) sleep. The brain activity of REM sleep is almost identical to the state when we are awake and marks the time during sleep, in which we are dreaming. It is also associated with paralysis of muscle systems in the body and rapid movements of the eyes (that is where the name comes from).

NREM sleep is subdivided into four separate stages depending on their depth. The deepest sleep we experience is stage 4 NREM sleep, meaning it is the time in which it is most difficult to wake us up. Below is a graph that illustrates the interplay between REM and NREM sleep throughout a typical night between 11 pm to 7 am:

The graph shows a continuous alternation between REM and NREM sleep in approximately ninety-minute cycles. However, the ratio of NREM to REM sleep within each cycle changes dramatically across the night. In the first half, the vast majority of our sleep is consumed by deep NREM and very little REM sleep. As we transition through into the second half of the night, the ratio shifts, with most of the time dominated by REM sleep, and little, if any, deep NREM sleep. This pattern suggests that going to bed either very late or waking up very early deprives us of one stage of sleep disproportionally, elevating the harm of unnatural sleeping habits.

Sleep increases our ability to memorize content

In his book, Walker lays out plenty of benefits related to sufficient sleep, including the positive effects for memorization of fact-based, textbook-like content:

• Sleep before learning refreshes our ability to make new memories
• Sleep after learning saves newly acquired memories, an operation called consolidation

The memory refreshment before studying is related to lighter, stage 2 NREM sleep with short, powerful bursts of electrical activity called sleep spindles (see graph below).

The more sleep spindles we obtain during sleep, the greater is the refreshing effect for our short term storage in the brain. The reason behind it is that during stage 2 NREM sleep, electrical current is pulsating back and forth between a temporary storage (the hippocampus) and a long-term one (the cortex). In doing so, sleep delightfully clears out the hippocampus, leaving this short-term information repository with plentiful free space. We awake with a refreshed capacity to absorb new information within the hippocampus, having relocated yesterday’s experiences to a more permanent depot. Note that sleep is equally relevant after a learning session. Particularly early-night sleep with abundant deep NREM facilitates memory retention. The more deep NREM sleep we get, the more information we remember the next day.

Have a tough problem? Sleep on it!

In the online course Learning How To Learn on Coursera, instructor Dr Oakley explains that toxic products aggregate in our brain and sleep washes these toxins out. But sleep does even much more than that. It’s actually a crucial part of the learning process of complex problems. During sleep, we continue to work on the concepts we were just learning about. Our brain strengthens the neural patterns to contextualize the content and get back to it the upcoming day. Plenty of anecdotal evidence suggests that sleep is remarkably helpful to figure out challenging problems that we could not solve during the day in our focused mode:

“It’s as if the complete deactivation of the conscious you in the prefrontal cortex at the forefront of your brain helps other areas of your brain start talking more easily to one another allowing them to put together the neural solution to your learning task while you’re sleeping.”

Dr Oakley, Instructor of Learning How to Learn

For me, sleep always had a bad reputation. I associated sleeping long with laziness and waste of time. Whenever I woke up late or slept too long, I felt guilty that I was not doing something else instead. “Sleep is for the weak” was the motto I believed in. Today, I track my time in bed using the app Sleep Cycle, and I feel full of drive when I wake up, seeing that I’ve had a solid 9-hour rest. Sleep can enhance our memory and ability to solve complex problems. Contrarily, lack of sleep can hinder us unfold our full potential in stressful situations such as during exams. Therefore, I take better care of my sleep, which I also enjoy very much. Now, time for a little nap?

Digitalization has impacted almost every sphere of our lives. We take care of our finances through online banking, communicate with our friends via social media, and search for potential partners on dating platforms. However, digitalization has not only affected our private lives but also the vocational world. In her book “The New Education”, university professor Davidson predicts that 65% of jobs that will be available for students in 15 years do not exist yet. Most of these new jobs are expected to evolve in the quaternary sector, which is composed of knowledge-based and highly demanding service jobs such as in business, legal, and IT functions.

To cope with these rapid developments, we require an excellent education system which prepares students adequately for a career in an unforeseeable environment. So far, the measures taken by universities and schools to satisfy these high requirements are disappointing. For a long time, employers report an increasing gap between graduates’ skill set and employers’ expectations, while students complain that the cost of higher education is increasing rapidly. As a result of these issues, the education technology (EdTech) industry has evolved to a considerable instance with a market size estimated to be around US$165 billion in 2016. One of the forerunners of the EdTech industry is the former head of Google’s self-driving car team and Stanford professor Sebastian Thrun. In 2011, he launched the online course “Introduction into AI” in which 160,000 students enrolled in the first batch. Based on the success of the online course, Thrun went on to found the massive open online course (MOOC) platform Udacity in 2012. His main motivation was to make high-quality and practice-oriented education accessible to people all around the world independent of their social background:

I felt that if we could just build a new kind of university that could democratize education and really reach everybody, we could have a bigger impact on the world than just building a self-driving car.

Will MOOCs be the future of education and teachers become dispensable?

Unlikely. The hype around MOOCs has flattened out and despite enormous investments in recent years, MOOCs do not seem to become the first-choice learning format for students, which is also indicated by the current high dropout rates (often >95%). Blended learning approaches, in which classroom learning is combined with digital sessions outside the classroom, are much more likely to disrupt traditional education. This concept offers the best of both worlds; the benefits of education technology such as the permanent accessibility of content and the advantages of face-to-face classes, for example, the formation of a community and the development of social skills. Because educational institutions have continued to rely on traditional lecture/classroom settings in recent years, they oversaw unconceivable potential from AI solutions to complement traditional learning by improving current common practices and unburdening teachers.

Here are three of the hottest topics that the EdTech scene is currently working on using artificial intelligence:

Personalized learning

Because of simple practicality, our current education system is based on grouping students merely by age, disregarding the differences in students’ learning paces, interests, and talents. Consequently, in every classroom there are a few students who become bored because they have understood a subject very fast and others who are discouraged quickly because they cannot follow the teacher’s explanations. In other words, education has moved towards a one-size-fits-all solution.

Rochelle, head of product management at Google, expects that AI applications will soon personalize the learning experience of students by suggesting individual learning objectives, selecting instructional approaches and displaying exercises that are based on the interests and skill level of every student. Just like Netflix shows us the film that we could also like, or Spotify creates a personal playlist based on our historical music choice, AI could suggest to students their most suitable educational setting. The provision of an individual learning journey would allow students not only to learn at their own pace but also regain the enjoyment and excitement that excellent education offers.

What is needed to implement personalized learning successfully is a close interplay between teachers and AI machines. While AI can help students to study content and complete exercises at home, teachers will have to adjust their lecture formats and focus on integrating interactive elements, revising content, and practicing soft competencies such as communication and presentation skills. Realizing that AI systems can only do half of the job towards high-quality education, the importance of well-trained teachers is once again emphasized.

Automated grading

I can still recall my former English teacher claiming that the worst part of his profession is grading. It is a repetitive task that consumes a lot of resources which could be utilized as valuable classroom time interacting with students. In fact, Kraus, president of the German teacher union, estimates that teachers in Germany spend up to 1,000 hours every year on grading depending on the classes and subjects they are teaching. Apart from teachers’ dissatisfaction with the status quo, also students criticize that the grades they receive are in part subjective, inconsistent, and opaque.

AI technologists are working on solutions to automate the tedious grading process and one successful example is the AI-based grading solution Gradescope, a system which is already used in many universities including Berkeley and Stanford University. Gradescope asks teachers to scan the students’ handwritten test solutions and automatically applies pre-defined grading criteria to all tests, thereby reducing grading time significantly and providing a transparent grading key to students.

While Gradescope already works robustly for standardized test formats such as multiple choice tests, a further challenge lies in the grading of longer texts, a discipline also referred to as automated essay scoring (AES). One approach to AES is finding objective measures such as the word length, the number of spelling mistakes, and the ratio of upper case to lower case letters. However, these obvious and quantifiable measures are not insightful for evaluating crucial aspects of an essay such as the argument strength or conclusiveness. To see how well machines would grade thousands of essays and replicate teachers’ gradings, the William and Flora Hewlett Foundation started a competition in 2012. Almost surprisingly, the output of the winning team was in 81% agreement with the teachers’ gradings, an impressive result that marked a turning point in teachers’ perceptions towards education technology. Today, AES engines are used to support human raters in scoring academic essays such as the GRE and TOEFL. For anyone who is interested in finding out how an AES engine would score their writing skills, I can recommend to sign up for the TOEFL test preparation for free and jump to the writing practice test.

24/7 personal student and teacher support

Have you ever seen your teaching assistant, who forwards you relevant course information and answers all your questions regarding the course syllabus on campus? If not, he or she probably does not even exist but is just an AI bot. At least this is what happened to 400 students at the Georgia Institute of Technology when professor Goel introduced the new teaching assistant Jill Watson to the class. Ms. Watson replied to questions, sent out reminders for due dates, and asked questions in the middle of the week to trigger discussions. Students described her responses as reliable and her tone as colloquial. The only difference to an ordinary teaching assistant was that Ms. Watson was not a real person but an AI bot developed to reduce the workload of professor Goel’s Ph.D. students. Ms. Watson took over tasks in which she was at least 97% confident that she knew the right answer. Such an AI-based bot is helpful for both teachers because it reduces their overall workload and students because they can get help on common problems instantly and round the clock.

Conclusion

Educational institutions have to find new approaches to prepare students sufficiently for their future. To achieve that, teachers and digital systems such AI engines will have to learn to collaborate effectively to overcome the remaining challenges on the path towards better education. Therefore, it is crucial that AI systems are tailored to act side-by-side with teachers, taking on the repetitive tasks and leaving the challenging ones to the human. This will allow educational institutions to increase productivity, save cost, and create new opportunities. The recent developments in education technology and particularly in AI make me confident that we will be successful in providing future students a high-quality education and enabling teachers to do more of what they love most — teaching.

The Future of Education and How AI Can Help Shape It was originally published in Luminovo on Medium, where people are continuing the conversation by highlighting and responding to this story.