The Promises and Perils of AI in Education

ASU + GSV 2020 Virtual Summit

The ASU + GSV 2020 summit took place virtually with the theme: B.C. To A.D. Before Coronavirus to After Disease: The Dawn of the Age of Digital Learning. The summit theme was particularly relevant, not only to the field of education, but also to how we as summit participants were adapting to a changing world. With thousands of participants from industry leaders to educators joining from their digital screens, the summit illuminated the new reality that was brought forth by the implications of COVID-19 and social distancing limitations. Bringing the summit online meant an increase in access for more people to join and partake in sessions, events, and networking. Many say it has even propelled us into the future of education faster. COVID-19 proliferated the rapid transition to online courses, and even allowed teachers to discover new teaching tools.

As I observed how COVID-19 was changing technology in education, one big concept I decided to learn more about was a specific type of technology with a lot of promise, Artificial Intelligence (AI).

I watched four session videos to help me explore the landscape of AI across various education sectors:

• AI in Education, Hype vs. Reality
• AI’s Potential to Transform K-12 Education
• AI Meets HigherEdTech: Implications for Better Student Outcomes
• AI Transforming Higher Learning

Images of organizations represented by session panelists and moderators.

What is AI?

According to Qualcomm, Artificial Intelligence is “an umbrella term representing a range of techniques that allow machines to mimic or exceed human intelligence.” Briticanna defines AI as “the ability of a computer or a robot controlled by a computer to do tasks that are usually done by humans because they require human intelligence and discernment.” These definitions allude to how computers use AI to enhance the tasks humans perform whether that’s driving cars or playing chess.

To many people, AI is a black box. This is mainly because of lack of exposure and understanding due to it’s covert nature and behind the scenes operation. However, AI is a powerful tool and holds great promise and potential for education.

Promises of AI

AI is relatively new to the educator sector and people are just beginning to explore it’s application.

What is AI’s value proposition in education?

One panelist, Andreas Oranje from Educational Testing Services (ETS), notes that “AI does not do anything on its own. It responds to needs and should be used as a tool .. it’s not a self propelling thing.” The value of AI is it’s response to human needs like the need to facilitate better relationships between instructors and students and content, the need to help teachers scale their work, see the big picture classroom trends and give more directed feedback by aggregating many different data points together, or the need for learners to assess their own knowledge and skills and to know where to focus their attention on. Schools can collect real time data using AI to help teachers save time on repetitive tasks so they can instead provide richer interactions and more personalized attention to every student.

So what can AI be used for in education?

Edtech companies are taking advantage of the speed, scale, and consistency of results that AI-supported tech can provide. Panelists shared some use cases of how their organizations are leveraging AI to improve student outcomes.

Intelligent Tutors

• AI can help provide intelligent tutors or agents to help assist educators do mundane tasks so they could allocate time towards interfacing with students and responding to learner needs. Since AI can provide intelligent evaluation and feedback, it can scale the learner experience for one-on-one tutoring regardless of a student’s background.
• For example, Coursera uses AI course coaches to help learners target what materials they need to review and provides learning interventions. Carnegie Learning also developed an intelligent tutoring system for math and science literacy with an adaptive engine to personalize remediation.

Automatic Grading

• AI can fill a proctoring need where it can help automate grading for large classes and give back time and energy for instructors to give more human feedback on work and provide support. Duolingo, a language learning app, uses AI and traditional psychometrics to help score English tests. They also use a large scale item response model to predict learner behavior for any exercise and for any user at any point in time behavior and use that data to tailor their experience. It can measure if a topic is too easy or too hard for the learner and the AI will then respond by giving them exercises at the appropriate level and zone of proximal development.
• Gradescope uses an AI virtual teaching assistant to help instructors in grading assessments, allowing the instructor to be responsible for teaching. At Coursera, they use machine assisted peer reviews for grading that can help drive increased engagement and retention for every skill level.

Teacher Sidekick

• AI can make teachers ten times more effective by helping them differentiate learning and interpreting data points. Many panelists spoke about AI’s ability to improve student outcomes through content and technology because learning analytics can offer fine grain views of how well students are learning and how engaged they are. From a curriculum standpoint, AI can help teachers differentiate instruction and identify if a student is deficient in a topic and recommend remediation strategies that are personalized to their skill or competency level. AI can discern what concepts are difficult for learners based on data metrics collected from the user such as time spent on tasks and pre and post assessments.This benefits learners because they can focus on their gaps in learning rather than starting from zero. Ultimately, this helps students remain in grade level and not fall behind.
• AI can also improve equity in higher education because it can drive a better learning experience, lower the cost of education while enabling educators to teach at scale, be faster, and more efficient.

Adaptive Learning

• AI can help create smart classrooms that propels learning to be more personalized, accessible, relevant, and effective. AI can be a game changer for educators because the data and insights from AI can increase engagement, guide learners to master concepts, and help individuals and organizations learn the skills to remain relevant. In addition, looking ahead, students of today will need work in the future of AI so exposure early on is important.
• For example, Coursera speculates that low skill jobs are at high risk for automation, so they analyze industry skill gaps from companies to help guide learners to learn the right skills or find a new career.

Student Retention and Well-being

• AI not only can assist in improving learning outcomes, but also the well-being of students. The most tangible example is AI’s application to saving children’s lives. This means protecting student safety in terms of mental-being and physical safety. In this case, AI can help collect and monitor behavior data that will provide insights before things get bad and can recommend interventions to improve mental health.
• EdX saw a 30% increase in engagement using AI nudges for learners. AI can help improve retention rates of students in online education by taking into account multiple variables like academic performance, interaction with the platform, However, a limitation of this is that there are some life variables or big life changes that you can’t control, predict that affect a student’s ability to learn.

Language Learning

• Elsa is a language accent app that uses AI to differentiate how to help a learner improve their English speaking ability by personalizing what areas they need to work on their accent.

Mentor Matching

• Protopia uses machine learning to read all profiles of alumni and helps to find a perfect match for those seeking advice from mentors using automated emails.

Hearing about the tangible real-world application was eye-opening, but also started a conversation about some misconceptions people have about AI.

Misconception #1: AI will steal teacher’s jobs and take over the world.

A computer robot holding a revolving world. Credit: Erik Blad for The Intercept

A huge concern and misconception that the general public holds about AI is that it’s uncontrollable and unstoppable. A lot of people imagine the dystopian world of machines and robots ruling the world. AI frightens people because AI is perceived as a threat to many jobs. However, in reality, AI is more like a supplementary sidekick. They can offer more personalized or adaptive attention to students and learners when there aren’t enough teachers to provide that individualized attention and support.

As AI can help alleviate mundane rote work, there is still a question about how much AI can replace intellectual or creative work. For example, particularly in the western world, academic freedom in the classroom is highly valued and any sort of curriculum personalization might encroach on that. While AI can disrupt the status quo of traditional learning, it’s purpose is not to take people’s jobs or functions away.

Most panelists agree that AI augments rather than replaces teachers. AI should act as complementary tools to lift teachers up. AI can be used to reduce pain points that teachers face. By contextualizing the AI technology or product to the pain points and the needs it helps address, AI can provide more predictive insights.

The golden question is how can we build tools to help teachers instead of using AI just for the sake of it? The role of the teacher needs to be preserved and we must make sure AI helps teacher tasks more efficiently rather than substitute or replace their roles.

Misconception #2: AI can solve everything.

A robot hand solving a Rubrik’s cube Credit: OpenAI Youtube

People also have the assumption that if AI is so powerful, then it can solve all education problems. However, AI is not the panacea for things wrong with education because it’s not a perfect system and cannot be 100% accurate. When stakes are high such as determining the future progression of a learner, AI can underperform and make a wrong decision a lot of the time. The impact of a wrong decision can have negative consequences for a learner and without checkpoints or intervention along the way, it can be harmful.

While it’s important to start debunking misconceptions about AI in education, there are still valid reasons why people push back on AI. It’s because AI can be risky in terms of data privacy, equity and bias, and strategy.

Perils of AI

Data/Privacy

• The way AI works is that it needs to collect a lot of data in order to synthesize, parse, and aggregate, interpret into meaningful insights. There are millions of data points of learning data that can be collected such as sensory data, body posture, physiological responses like sweating, speech, other personally identifiable information (PII), and even handwriting recognition. As more data is being collected and stored, the issue of privacy about who has access to and control of personal data, what does it mean for data to be all stored in a place, how long does it get stored for, and what is the data being used for. Parents want to know that students data is private and protected.
• There’s a need to be radically transparent, traceable, and explicit about data usage because consumers are more vocal nowadays and they want to see their evidence of how and when and what their data is being used for.
• Rather than nefarious desires for data usage, we need to move in the direction of using AI for good. A starting point is to make sure data is fully anonymized, and to model data protection strategies from the financial services industry to protect personally identifiable information.
• Panelists advocated for AI data collection to use a concept called “Privacy by design” which is an opt-in system where users need to consent to give their privacy to get the proposed value of AI. Data privacy and COPA compliance need to be treated as a top priority and there should be universal standards about how to handle data. The field needs to be more thoughtful and intentional about what type of data collection is needed, from whom, and for what purposes, and for how long.Control needs to be given back to the individual and allow them to make choices. People are often already inundated with a lot of their data being collected on various platforms so developers need to be more intentional about what to use the data for and use evidence-centered design.
• In addition, if data persists and is stored/tracked forever, panelist Bethany Maples points out that it may be an inevitable path for rebirth as people might be put on a track that they can’t escape. The American notion that we invent ourselves and can start over and reinvent ourselves is being challenged. She advocates to finding a way to delete data, giving users more data control, and building allowance for change.

Equity and Bias

• Data privacy concerns are connected to issues of ethics. There are many ethical concerns about AI because depending on how an AI algorithm is designed, there can be built in biases. If we aren’t conscious of the biases and don’t have a mechanism to control for bias, there can be a lot of assumptions that can skew data. This results in an unintended bias for lower socioeconomic backgrounds or underserved communities. I was introduced to the AI in Education Toolkit for Racial Equity that seeks to address some of these concerns.
• The second equity issue is the widening of the digital divide gap and access to technology like devices and internet connection. The gap of availability in broadband and devices illustrates the stark differences of access to opportunity between the haves and the have not’s. It actually highlights a larger systemic problem.
• To put this into context, there are 80 million learners in the US, and 50 million of those are learning in hybrid or remote modes, but 13 million do not have internet or devices. Access to technology should not be a luxury, but is a must have. If not now then when?
• As COVID-19 has highlighted the digital divide, the question still remains whether AI exacerbates it or closes it. Some analysts suggest that AI is going to make access to quality education much better because it can help eliminate barriers and democratize student achievement. Many panelists saw that while AI poses many complexity and challenges, there are numerous opportunities if we all take action and we are positively moving in the right direction.

Advocacy

• Michael Connor of Middletown Public Schools stated “We need buy-in not more tools or innovations.” More educators, teachers, parents, and students need to start understanding what the value of AI is and to embrace, advocate, adopt, and share its benefits and advantages with others. Buy-in also involves the need to humanize AI and be able to tell the promising stories about how it can improve and aid education. Conversations around AI need to be more commonplace and a part of the organization thread as schools become more lean, agile, creative, and analytical.
• For example, if schools can start integrating AI into the architecture of curriculum, and see the value of big data, it can help target students and close equity gaps. If AI-powered education can provide personalization and precision to students so that each student gets a tailored lesson plan and is engaged, this is a huge upside for students.

Nonetheless, the call to action and growing consensus is for the education ecosystem from school and university admins, Edtech companies, parents, instructors, and etc — to collectively come together to solve the immediate education problems ideate high tech, low tech or even no tech solutions to address the most pressing education issues at hand. Adoption of AI-technology will take a while in mainstream due to the highly regulated nature of education.

Opportunities for AI

The future of AI is ripe with opportunity and could bring grand transformation to education. Some panelists made predictions that as technology is projected to get cheaper and become more available, AI-powered education will come at a more affordable price as a result. This could mean that education can be more scalable and learner centric where learners get more choice in their learning options in a more efficient manner. This is a huge opportunity for entrepreneurs to think about the market and how AI and adaptive learning can assist in just in time learning, or to create pathways for different skill sets in the economy that keeps shifting.

The other types of opportunities are how to improve AI for application in education included:

• AI should be applied outside the classroom context of teaching and learning like improving administrative workflows.
• Some say the ultimate end goal of AI is getting rid of grades. This might look like expanding credentialing beyond numerical based assessment or GPA to a more competency-based creative and knowledge economy where learners are demonstrating they are able to create new things. This may be subjective and harder to assess.
• While AI has historically leveraged neuroscience, there’s a lack of learning sciences incorporated because learning is contextual and much more from the science of how people learn that can improve AI.
• In addition, the demographic of people developing AI is not diverse enough. If AI needs to be designed to meet the needs of all learners, in order to mitigate bias, we need to have more inclusive of people from all different types of backgrounds from race, gender, sexual identity, religions, abilities, and etc.
• Lastly, we need to use AI to improve the human connection. This could look like fewer constraints and expand the reach of connections of learners across geographical locations.

COVID-19 has permanently altered our approach to education. When students go back to physical classrooms, they may have an epiphany that online education conditions became the better and preferred way to learn, education needs to change, and the overall efficiency of the education system can improve. As I listened, I sensed more optimism about the potential of AI in paving the way for these changes. So how will education look like after COVID-19 is curtailed? We are on the slow, yet steady verge of finding out.