Machine Learning in Education
What Is Machine Learning?
The term “machine learning” is used to describe one kind of “artificial intelligence” (or AI) where a machine is able to learn and adapt through its own experience. For example, picture an electronic chess or checkers game that allows you to play against the computer. Using machine learning, the computer can improve its game over time, based on new information it gets every time it plays, without any new input from a programmer (Samuel, 1959). Machine learning uses algorithms to detect patterns in very large data sets and generate new insights and/or behaviors.
While often used synonymously, it is important to know the difference between artificial intelligence (AI) and machine learning. When we refer to AI, we are really speaking to a broader concept. In AI, we are referring to “smart” technology where the machines autonomously make decisions on their own without input from a human. Machine learning, on the other hand, is a series of techniques (such as neural networks, decision trees, etc.) that allow a machine to understand and make use of relationships between inputs and outputs.
Machine learning works especially well for prediction and estimation when the following are true:
- The inputs are well understood (you have a pretty good idea of what is important but not how to combine them).
- The output is well understood (you know what you are trying to model).
- Experience is available (you have plenty of data to train the algorithm).
One example of how this technology is used for assistive purposes is Microsoft’s app Seeing AI, which interprets visual data collected via a smartphone camera and turns the information into speech. The app can read a document, describe the user’s physical environment, identify objects (including money), or even interpret the emotions of people in view of the camera. The app uses both machine learning and artificial intelligence to take in significant visual data on the fly and turn it into auditory output for the user.
Why is This Important?
Machine learning is transforming the way most of the world’s businesses operate, from health care, to financial services, to transportation, and beyond. However, when it comes to assistive technologies, machine learning is a game changer because the tools adapt to optimize the experience for the end user.
“The world is quietly being reshaped by machine learning. We no longer need to teach computers how to perform complex tasks like image recognition or text translation: instead, we build systems that let them learn how to do it themselves.” (Hern, 2016).
The International Data Corporation (IDC), a global leader in research and intelligence for information technology, predicts that by 2020 industry will spend $47 billion on machine learning and artificial intelligence (Which 50, 2016).
Who is Working on This?
All the major software and technology companies are investing in this emerging technology: Amazon, Apple, Facebook, Google, IBM, Microsoft, Twitter, and YouTube, to mention a few. Some companies are focusing on creating machine-learning libraries, studios, and application program interfaces (APIs) that provide collaborative drag-and-drop tools that can be used to build, test, and deploy predictive analytics solutions based on a set of data.
In this chapter, we will highlight some of the implementations that have the potential to help students in the classroom.
Here are some of the applications of machine learning that companies are working on.
Speech Recognition – The ability for a computer to use complex algorithms to convert the sound waves of a human voice into machine-readable data and output. For example:
- Amazon’s Echo (aka “Alexa”) – A voice recognition interface that allows users to enter commands by speaking in a natural voice instead of using a keyboard. Alexa Voice Service lets you “add intelligent voice control to any connected product that has a microphone and speaker.”
- Apple’s “Siri” – Voice recognition software that uses machine learning to understand user interests and present content tailored for their preferences.
Image Recognition – Allows a machine to use advanced algorithms to identify specific persons, places, or things in a camera, photo, or video. For example:
- Apple’s integrated camera – Provides face recognition, smile recognition, and blink detection. Apple gives machine-learning image recognition APIs to developers that lead to innovative uses.
- Facebook – Provides image recognition with basic auto-generated image descriptions.
- IBM’s “Watson” – Provides a visual recognition service that allows users to tag, classify, and search visual content.
- Microsoft’s “Custom Vision” – Allows users to upload and tag images. The software then “learns” which concepts are important to the user and develops the ability to automatically tag future uploaded images
- An “Emotion API” built on Microsoft’s Azure lets developers create tools that analyze faces in images to detect feelings and customize the app’s responses accordingly. This is an example of how machine learning can provide a different kind of categorization of something visual.
Sound Recognition – Allows a machine to use advanced algorithms to identify speech for transcription into text. For example:
- Google’s YouTube platform provides automated closed captioning with the ability to visualize sound effects such as [LAUGHTER], [APPLAUSE] and [MUSIC].
Prediction and Personalization
Using machine learning, developers can personalize outputs to an individual user by learning to detect or predict what the user may need. Examples include:
- Cerego uses machine learning to create personalized learning pathways for individuals and optimizes their learning experience by measuring their progress and learning preferences.
- Watson’s “Natural Language Understanding” tools allow developers to analyze text to extract concepts, understand sentiment and emotion, and understand text in multiple languages.
- Content Clarifier helps people with cognitive or intellectual disabilities such as autism or dementia. It can replace figures of speech such as “raining cats and dogs” with simpler terms and trim or break up lengthy sentences with multiple clauses and indirect language.
- Twitter Cortex developed “timelines ranking algorithms” that use machine learning to help Twitter predict whether a particular Tweet will be engaging to a particular user.
- “SmartReply,” built on Google’s TensorFlow, predicts the likely response to an email and automatically generates the text so that the user can just click to send the reply without having to type it.
How is Machine Learning Applied in Education?
Machine learning has the potential to be a game changer for education through the use of automatic image descriptions, speech recognition, personalized learning, smart search, and early detection of learning disabilities. These developments could profoundly help teachers reach all of their students in effective and meaningful ways. Targeted feedback allows students to gain deeper understanding of the material with the opportunity to have one-on-one instruction where they need it the most. This technology is very early in its development and will take a few years to mature, but the potential is very promising. Many of the biggest educational publishing companies, including McGraw Hill, are shifting to an “Education as a Service” (SaaS) model, providing web-based learning systems driven by machine-learning algorithms to customize learning (Gaskell 2016).
Challenges and Opportunities for Students with Disabilities
Machine learning in special education is still in its infancy and will take some time to mature before it can address the needs of all learners. While the potential for this technology to revolutionize the way in which students learn is worth monitoring closely, there are still many barriers for students with disabilities.
Several challenges must be overcome before the true potential of machine learning in education can be a reality. Having a reliable data set for training the machines is by far the most difficult problem. Students with disabilities tend to be highly variable compared to their non-disabled peers (Odom et al., 2005), so gathering enough data to create reference points can be difficult (Alnahdi, 2015).
Virtual assistants such as Amazon Echo that use speech recognition as a primary user interface often fail to recognize students with accents, stuttering impairments, and younger children with high-pitched voices. In addition, students who pause while asking a question will encounter problems because the system thinks they are finished and acts on the partial command.
While facial recognition and emotion detection technologies can be valuable for supporting students with autism, the time required to train the machine may be significant and therefore a deterrent to use. (See the Speech Recognition chapter.)
Additionally, in the last few years there has been a lot of concern over student privacy and data collection. Parental permission is required both for the initial data collection needed to train the system and again in order to monitor all aspects of the students during class. In 2011, inBloom created a computer system to store data in a secure, common format that gave the schools complete control over what data they collected, how it was used, and with whom that data was shared. While schools embraced the data collection for informed decision-making, pressure from parent and privacy advocates contributed to the decline of inBloom (Herold, 2014), and they closed their doors in 2014.
While privacy concerns are a big barrier to collecting data, the opportunities may outweigh the risks. Some schools are opting to sign documents to protect student privacy like the Student Privacy Pledge that is intended to comply with federal and state laws and regulations.
With increased data points, the technologies will learn more about student behavior and be able to accurately support and identify the need for more interventions. The same technology also has the potential to detect learning disabilities at an early stage (David & Balakrishnan, 2010). This technology has significant implications for teachers as well. The use of “smart” technology can help teachers identify the ways students learn best and modify the instruction to meet their needs. Additionally, it can adapt the material instantaneously. If a student answers a question incorrectly, the machine knows to offer a simpler question or language. This feature could help students with intellectual disabilities who may need simplified language to understand a concept (Fajardo et al., 2014). Computerized adaptive testing is one such example.
Personalized learning and personalized learning platforms provide the ability to customize lesson plans to meet the needs of each student. Teachers have been creating individualized instruction for students with disabilities for decades, in the form of an IEP (Individualized Education Plan), but the use of machine learning has the potential to automate some of the adaptations. Machine learning can help teachers create personalized lesson plans based on each student’s strengths and weaknesses and recommend steps to help students with a variety of disabilities. (See the Personalized Learning chapter.)
Story from the Field
Administrators for Tacoma Public Schools have long sought to understand why students drop out of high school. In 2010, only 55% of the students in the Tacoma Public School District graduated high school compared to the national average of 81%. Administrators wondered if they could use data to predict which students would drop out of school. Shaun Taylor, CIO of Tacoma Public Schools said,“By using predictive analytics, we thought we would be able to intervene earlier and work closely with those at-risk students. Then we would be able to reach our ultimate goal: getting that graduation number close to 100%.” Shaun Taylor, CIO of Tacoma Public Schools. Shaun and others reached out to Microsoft and Azure Machine Learning to see if they could help them understand their data.
Through the use of predictive analytics, Microsoft and Tacoma Public Schools created a data warehouse and analyzed data such as student grades, health, and attendance to see if there were any correlations in the data. They found that the data gave them a better understanding of the students’ needs within a 90% confidence level. Teachers were able to intervene much earlier and the district was able to graduate 78% of their students in 2014. They are now developing the next level of data-driven improvement with the help of machine learning.
While the authors of the study did not specifically address students with disabilities, it is estimated that in 2014, 63% of students with disabilities graduated from high school. In Washington, where Tacoma Public Schools are located, that number falls somewhere between 50-60% (Grindal and Schifter, 2014). While analysts predict that the U.S. will have a 90% graduation rate by 2020, the 2015 Building a Grad Nation report estimates that only 61.9% of students with disabilities will graduate (Kineavy, 2016). More work needs to be done to make sure that the use of machine learning will benefit all students.
Conclusions / Actions
Look for machine learning applications in the classroom as the technology matures and more educators become familiar with how to leverage machine learning to benefit their students.
- Consider ways machine learning would be beneficial for your students. Join pilot studies or use Microsoft’s Azure to help collect the training data needed to improve the machine learning algorithms.
- Start experimenting with virtual assistants such as Alexa from Amazon to see how they can be used in a classroom setting.
- Machine learning relies on massive amounts of potentially sensitive student data. Take steps to understand the privacy and security implications as well as your students’ rights.
- Since machine learning must collect data on your child, work with your child’s educators to understand where this data is stored, how it is used, who has access to it, and how it is being protected and anonymized.
- Look for pilot programs or studies that your school could participate in, or other schools that are using machine learning, and connect your school with them.
- Promote machine learning to other parents and educators as a tool that will help other students.
- Experiment with applications such as Microsoft’s Seeing AI on the iPhone with Voice Over, or any number of apps in the app store with ties to machine learning.
- Keep abreast of the latest uses of machine learning and suggest areas where the technology could be implemented.
- Experiment at home with machine learning-related projects and applications such as Alexa, Google Glass, Xbox, etc.
- Alnahdi, G. H. (2015). Single‐subject designs in special education: advantages and limitations. Journal of Research in Special Educational Needs, 15(4), 257-265.
- David, J. M., & Balakrishnan, K. (2010). Machine learning approach for prediction of learning disabilities in school age children. J. of Computer Applications, ISSN-0975-8887, 9(10).
- Fajardo, I., Ávila, V., Ferrer, A., Tavares, G., Gómez, M., & Hernández, A. (2014). Easy‐to‐read Texts for Students with Intellectual Disability: Linguistic Factors Affecting Comprehension. Journal of Applied Research in Intellectual Disabilities, 27(3), 212-225.
- Gaskell, A. (2016, November 07). Machine Learning and the Future of Education. Retrieved August 16, 2017, from https://www.forbes.com/sites/adigaskell/2016/11/04/machine-learning-and-the-future-of-education/
- Grindal, T., Schifter, L. (2016, January 14). The Special Education Graduation Gap. Retrieved August 7, 2017, from http://www.huffingtonpost.com/todd-grindal/post_10880_b_8976972.html
- Hern, A. (2016, June 28). Google says machine learning is the future. So I tried it myself. Retrieved August 23, 2017, from https://www.theguardian.com/technology/2016/jun/28/google-says-machine-learning-is-the-future-so-i-tried-it-myself
- Herold, B. (2014, April 21). InBloom to Shut Down Amid Growing Data-Privacy Concerns. Retrieved August 17, 2017, from http://blogs.edweek.org/edweek/DigitalEducation/2014/04/inbloom_to_shut_down_amid_growing_data_privacy_concerns.html
- Kineavy, F. (2016, December 12). Students with Disabilities More Likely to Drop out of High School. Retrieved August 7, 2017, from http://www.diversityinc.com/news/students-with-disabilities-more-likely-to-drop-out-of-high-school/
- Odom, S. L., Brantlinger, E., Gersten, R., Horner, R. H., Thompson, B. & Harris, K. R. (2005) ‘Research in special education: scientiﬁc methods and evidence-based practices.’ Exceptional Children, 71 (2), pp. 137–48. From https://www.researchgate.net/publication/259545422_Single-subject_designs_in_special_education_Advantages_and_limitations
- Samuel, A. L. 1959. Some studies in machine learning using the game of checkers. IBM Journal of Research and Development 3:211–229. Reprinted in E. A. Feigenbaum and J. Feldman, editors, Computers and Thought, McGraw-Hill, New York 1963.
- Which 50. (2016, November 15). Machine Learning And AI Spending To Surge Toward $47 Billion By 2020: IDC. Retrieved August 7, 2017, from https://which-50.com/machine-learning-ai-spending-surge-toward-47-billion-2020-idc/
- Machine Learning: The “Next Big Thing” in Education http://www.gettingsmart.com/2017/04/next-big-thing-education/
- AI and the Classroom: Machine Learning in Education; http://blog.trueinteraction.com/ai-and-the-classroom-machine-learning-in-education
- Xbox Live Machine Learning with Recommendation Suggestions; https://www.youtube.com/watch?v=zG6nt7Y-DWI
- Blog posting from IBM “Simplifying Content for People with Cognitive Disabilities” September 21, 2016 by: Ram (P G) Ramachandran
- What is Machine Learning – and Why is it Important? https://www.interactions.com/machine-learning-important/
- 8 Ways Machine Learning Will Improve Education
- 20+ Emotion Recognition APIs That Will Leave You Impressed, and Concerned
- Machine Learning Is Redefining The Enterprise In 2016 https://www.forbes.com/sites/northwesternmutual/2017/06/29/which-asset-class-will-justify-investors-first-half-love/#7819b23b2f69
- Machine learning libraries in particular serve as the foundation for tools and solutions being applied today. Some of the core available libraries available today include:
- Google’s TensorFlow is an open-source software library for machine learning that allows researchers and developers to create custom enhancements to their platforms and services.
- Microsoft’s Azure is an integrated set of cloud services that Microsoft offers.
- “Machine Learning Studio” lets developers prepare large datasets for machine learning and predictive analytics to generate insights “buried” in the data.
Scholarly Publications: Machine Learning
- B. Kotsiantis – Use of machine learning techniques for educational proposes: a decision support system for forecasting students’ grades: Artificial Intelligence Review April 2012, Volume 37, Issue 4, pp 331–344; https://link.springer.com/article/10.1007/s10462-011-9234-x
- Thomas Way, Adam Bemiller, Raghavender Mysari and Corinne Reimers – Using Google Glass and Machine Learning to Assist People with Memory Deficiencies; http://worldcomp-proceedings.com/proc/p2015/ICA6330.pdf