AI in Education
AI in Education: Where is it now and what is the future?
AI science in education is more than imagination. One study found that 34 hours of language education in the application is equivalent to a full university semester. But the broader category of educational AI and educational technology (ad tech) goes better than learning a language. Companies such as Carnegie Learning and Fuel Education apply artificial intelligence to K-12 teaching. One of the most popular Adtech platforms, Alex McGraw-Hill, is a web-based, AI-powered assessment and education system that includes K-12, including homeschool and college content.
Like many other artificial intelligence domains, China is leapfrogging this pack to advance AI-centric education. But the United States is also taking steps to keep AI in the classroom. Meanwhile, many leading AI ethicists are raising the alarm. [Tongues. a machine-learning model with png]
I did some research on the role of AI in the classroom, what the future holds for AI in future education, and the ethical side of things.
That's what I found.
The Foundations of AI in Education: Knowledge Space Theory
In education, AI usually does not find out what a student does and does not know through diagnostic tests and then develops a personalized curriculum based on each student's specific needs. According to Derek Lee, founder of Chinese Adtech Unicorn Square AI, "In three hours we understand students more than three years spent by the best teachers" - just a little bit.
Many Adtech companies, including Square AI, train their AI learning platforms using Knowledge Area Theory (KST). Knowledge Space Theory "applies concepts from combinatorics and stochastic processes to modeling and empirical descriptions of specific knowledge areas."
Knowledge structure and family diagram of groups
Visuals from KST research paper
In simple words: Knowledge Space Theory uses mathematical language to define and track "knowledge points". These points form a detailed picture of a person's "knowledge state" for a particular subject.
There are a lot of complex maths that go into KST, but we put it there. This April 201 research paper by Christina Stall and Cord Hakkemeyer offers more details on technical inclinations.
The most popular Adtech systems demonstrate the strengths and weaknesses of the AI-based approach.
Assessing and studying in the knowledge space, ALEKS is a popular online education platform with K-12 broad curricula through higher education. Students begin with an ALEKS assessment to measure their current knowledge in a range of subjects. Based on their responses, ALEKS claims to obtain an accurate picture of the student's level of understanding. When students enter the Learning Mode mode, the system uses their assessment to adjust its course to fill in its knowledge gaps.
ALEKS pie chart
An ALEKS pie chart from PR Newswire
Once a student consistently gives correct answers, they move on to the next heading, and ALEKS updates their student's knowledge status map. Additionally, ALEKS periodically re-tests students on previous titles to ensure longer theme memory retention. According to McGraw-Hill, Alex's current masters, "Because students are forced to demonstrate mastery through unpredictable mixed questions, Alex's mastery of the course is the true mastery of the curriculum."
But knowledge is not without controversy in space theory and ALEKS. After all, I argue that there are unlimited possible "knowledge states" for a subject. But ALEKS assessment is usually only 20-30 questions long. From these very few questions how can they claim to accurately measure one's knowledge state?
ALEKS Assessment
ALEKS Assessment Questions from ALEKS.com
In fact, you shouldn't look far to find document display inconsistencies. According to AdSerge, ALEKS created high performance in medium-sized midwestern cities but not in smaller cities. The same article also points out that Dreambox Larni, an online math learning platform, "was associated with higher education outcomes in a mid-sized city in the Midwest but not in a neighboring city like this."
Zoom out: China is investing heavily in AI for education
Come back for a moment. China has invested billions in substantial share centers around artificial intelligence, and education technology. According to a report, "AI education will fall in the coming years and global spending is projected to reach Rs 1 trillion by 20252. Most of the growth comes from China, followed by the United States accounting for more than half of global AI education spending."
Chart showing Chinese investment in AI
Chart obtained from TheConversation.com
In addition to government support, Chinese tech companies have also acknowledged the heavy spending of the Chinese middle class. The importance of education is deeply rooted in Chinese culture. A study, detailed in the South China Morning Post, highlights that families of preschoolers spend an average of 2 percent of their income on education and that families of 12 children spend 20 percent of their income on education-related expenses. A desk this past summer, Carnegie Mellon University and Chinese ad-tech firm Yixu Education Inc. In his words, "The CMU-Square AI Research Laboratory will develop new ways to improve the adaptive learning experience of K-12 students around the world for personalized AI, machine learning, cognitive science, and human-computer interface technology." You, through the Square AI Learning brand, has opened more than 1,700 AI-powered learning centers across China and plans to open more. And keep in mind, this is just one of many Adtech firms in China.
China is a battleground for AI-leadership versus AI-assisted education
Why did I switch gears to talk about China? As it turns out, the .4 1.4 billion countries are a battleground for AI's two competitive approaches to education: AI-led versus AI-assisted.
No company performs this first approach better than Squirrel AI. Founded in 2014, Squirrel claims to be "China's first pure-game AI-powered customized education provider."
The squirrel works with highly skilled teachers to divide the subject into smaller potential ideological blocks. For example, Squirrel broke high school level math into more than 1000 Knowledge Points (Remember Knowledge Space Theory?). The idea is that this level of granularity allows Square to "diagnose" student knowledge gaps more accurately than others.
Open the laptop showing some kind of graph
Squirrel AI - Image by Noah Sheldon for MIT Technology Review
"By comparison," Karen Hao writes in her brilliant MIT technology review article, "the textbook can divide the same subject into 3,000 points; Divides into
Squirrel AI founder Derek Lee envisioned a classroom where human teachers play a relatively passive role. In his philosophy, AI takes care of real teaching, and people only go into that situation if there are problems.
But Chilari AI is not the only player in the Chinese Adtech space. ALO7 exemplifies another approach: AI-assisted learning. Pan Pengkai, founder and CEO, built an English language-learning platform that uses AI to enhance the classroom experience.
Describing ALO7's approach, Karen Hao writes, "Knowledge that can be used through adaptive learning, like vocabulary words, is practiced at home through the application. So can skills like pronunciation, which can be refined through speech-recognition algorithms. But something that requires creativity." Also, like writing and talking, learning is done in the classroom. ”
In other words: where squirrels target all-but-replaced human teachers, ALO7 tries to help them.
American and European companies are also moving ahead with AI-powered Adtech
Of course, China is not the only country investing in Adtech. Analysts report that the use of AI in American education from 2018-2022. 47.7777%. And US AI startups reached a record high of $ 3 billion in venture capital in 201.
So, how is the rest of the world getting closer to AI for education? I found many American and European companies supporting the AI-assisted approach.
Open the laptop next to the Carnegie teaching material
From CarnegieLearning.com
One such company, Carnegie Learning, has won three awards (in 2019 only) for their AI-based learning platform for K-12 math. The Carnegie team believes that "all teachers can teach mathematics successfully and effectively with the proper resources and support to build a culture of collaborative learning." In their view, human teachers are still the most important component of learning. It looks a lot like ALO7's approach: AI was used as a tool to help human teachers in the classroom.
Meanwhile, in Sweden, small labs have a completely different approach to building AI for education. Instead of creating their own ad tech platform, SANA actually helps to privatize education by taking care of the AIN and machine learning aspects. Small Labs homepage screenshot titled "Teacher Saved Up to One Day Per Week via Small-Managed Personalized Assignments".
Unlike companies like Dreambox Learn, which specialize in one subject, Sana covers several areas: math, language learning, and vocational education. For a Southeast Asian education company, Sana created a personalized homework system. For another client in financial exam preparation, offers small individual review sessions that suggest the best steps for each student.
Both of these client studies from small labs demonstrate an ALO7-like approach: using AI to free up teachers ’time and personalize learning without removing individual elements.
A quick tangent AI support
Want to talk about switching gears for a moment. The language learning app allows more than 0 users to enjoy courses in 22 languages. This user base creates the world's largest collection of language learning data - a real treasure trove.
Of course, this data is constantly used to improve your product. But they are not just hoarding it for themselves. Instead, they are "committed to sharing our data and findings with the wider research community."
Image from the homepage of Ai.duolingo.com
As both an individual and a community, the team "works to build unique systems and gain new insights into the nature of language and learning." Through their research hub, ai.duolingo.com, they share publications and data sets so others can close their work.
In a world where AI development is often viewed as an "arms race", it's nice to see other examples of companies choosing cooperation and openness for hostility and intimacy.
We cannot forget the moral aspect of AI in education
Clearly, AI-powered Adtech is already here to stay. But the state of AI in education is not all positive. Earlier I mentioned performance differences between school districts using ALEKS. That's just the tip of the iceberg.
In late August of this year, motherboard research found that automated essay-scoring systems are easy to fool with Gibberish and grade Chinese and African-American students.
An excerpt from the article explains how e-rater reduced students from mainland China to grammar and mechanics but is high for essay length and word choice, suggesting that those students are using "pre-memorable shell text". E-Rotter scores African American students low in grammar, style, and organization but expert graders often score them very well.
Rachel Thomas's article highlights on Twitter
The system in question, E-Rater gave students from mainland China higher overall grades than those given by expert human graders. Meanwhile, e-Ritter gave African American students poorer marks than those obtained from human graders.
In their research, the motherboard copied the MIT experiment by feeding the GRE's online score it now! The practice tool, which uses an e-router, is based on a basic automatic B.S. with two "meaningless gibberish" essays. Essay Language (BABEL) Generator. I think Tom Feathers, writing for Motherboard, gives an excellent description of the result:
How does it happen It turns out that this scoring algorithm does not try to analyze the actual quality of the writing. Instead, they are fed with a large collection of human-graded essays. Using this data, the algorithm tries to identify a pattern that collaborates with higher or lower grades.
Take your side in the fight against partisan AI in education
This is a known problem: AI systems learn and expand human biases. In this case, when the AI-powered Adtech system was trained on the data described by humans, the human biases stained the data, which in turn infected the algorithms, resulting in biased results.
At the end of this series: A GRE essay-grading system that could sustain institutional racism without change that first appeared when a group of unknown humans sorted a group of different essays.
What can we do to reduce the impact of AI bias on education? For one thing, keep the man in the loop. GRE grading, for example, would be even more biased if no one was involved.
As soon as the motherboard writes, "All the essays written by e-Rutter are also classified by humans and the controversial second human is sent to the final class. Because of that system, ETS does not believe that any student was adversely affected by the bias discovered in E-Rotter."
After that, fighting back against prejudice in AI starts with you. Educate yourself where the AI bias comes from, and then spread that awareness to others. Challenge your own ideas about how and where we should use AI. Encourage your own company to take action against biased data. And lobby your government to regulate data collection and AI usage.
Comments
Post a Comment