The Classroom of Tomorrow: Promise and Limitations of AI Teaching Assistants

New AI Tech Is a Powerful Tool. How Should We Use It?

Plato in his academy, by an Unknown Xylographer (1879)

Introduction

Recent advances in AI (Large Language Models (LLMs), Speech-to-Text and Text-to-Speech) enable radically new EdTech functionality and pedagogical approaches. This could transform global education. Critics argue that AI risks worsening inequality, commercializing learning, and threatening teachers' jobs. I see exciting opportunities if we implement them thoughtfully.

This post explores the promise and limitations of AI teaching assistants (TAs). I also want to imagine the desirable future of education in 10-20 years. I welcome critiques and perspectives from educators and technologists.

The follow-up post presents a thought experiment on what the continuing progress of foundation AI models could mean for education. Will we "build" AI teachers "simply" by training a foundation model in pedagogy?

Heads up: While my expertise lies in EdTech, I haven't been a classroom teacher. 
As such, my perspectives might lack nuances. Nonetheless, I invite a discussion on these topics.

Envisioning Future Classrooms

Imagine a classroom where each student has a teaching assistant supporting their learning journey. TAs are directed by a teacher and understand each student's needs, challenges, and optimal pacing. Teachers craft personalized learning plans aided by TAs.

The learning plan is based on the course curriculum. It takes into account student learning profile: their general knowledge level and knowledge gaps, personal interests and psychological type, learning style, and velocity. TAs follow a learning plan - give assignments, grade them, and keep an easy-to-interpret learning profile of every student.

Many assignments, especially for younger students, are designed as games. Group projects can be done in the virtual or real world. Lectures come from the most engaging, expert teachers globally for each subject, recorded and delivered at just the right time. Students participate in collaborative projects, seminars, and class discussions overseen by the teacher and facilitated by TAs. Games, simulations, and VR experiences engage learners.

AI TAs collect real-time analytics, enabling better academic decision-making. Each exercise also serves as an assessment and a source of valuable information about a learner's progress and the efficacy of the activity. Knowledge gaps are detected and addressed by adjusting the learning plan. Student's personal learning plan is continuously adapted based on this information. The dream of mastery learning and evidence-based pedagogy become a reality.

AI enriches teachers' professional development by providing regular feedback and advice. AI handles most administrative tasks. AI tools can also help teachers prepare learning materials for lectures, seminars, exercises, and tests.

Now imagine that today's K-12 schools evolve into learning centers that provide daycare and are staffed by people who may not have a Master's in Pedagogy but are good with kids. Their role will be to provide a safe and nurturing environment. There will be many such "counselors" - class sizes stay close to today's best schools. Academic learning is handled by a smaller team of teachers, in many cases remotely or asynchronously, supported by AI TAs - one TA per student. 

The result? Proponents believe this model could improve outcomes and satisfaction for both students and teachers. This is the promise of AI-based Teaching Assistants. It's a hypothesis and a dream, with first pilots and trials just starting.

Is this a desired future or a dystopia? 
Do you believe the broad adoption of AI Teaching Assistants can improve learning outcomes?

In this future, human teachers can focus on the high-value part of their job:

• Providing motivation and encouragement.

• Designing and tuning individual learning plans.

• Helping with learning problems that TAs can't handle.

Human teachers know what questions to ask to assess student's mastery of the material. And they possess human intuition and empathy. So they can diagnose student knowledge gaps and misconceptions and either solve the problem themselves or instruct AI TAs on how to proceed. These are vital functions. I believe AI will empower – and thus scale – the best teachers. These teachers would transform into learning facilitators, guidance counselors, and coaches.

Teachers will get time and resources to go beyond teaching academic subjects to meta-skills like curiosity, adaptability, leadership, critical and first principles thinking, logic, emotional intelligence, ethics, and grit.

I believe AI TAs are coming and will disrupt education in a significant and positive way. 

But, I do not believe these TAs will replace teachers. Here is why.

What AI-based Teaching Assistants Can and Can't Do

Let's closely examine the key roles of teachers today:

1. Providing motivation, encouragement, and learning discipline; cultivating a positive learning environment.

2. Guiding learning: designing and implementing learning plans.

3. Delivering instruction to groups and individuals: lecturing, assigning work, providing feedback, and facilitating group discussions.

4. Testing comprehension and assessing knowledge gaps.

5. Remediating knowledge gaps.

6. Leading field trips, lab experiments, and workshop projects; providing safety supervision.

7. Providing childcare (in elementary and middle school).

8. Accommodating students with special needs.

9. Engaging with parents on student progress.

10. Performing administrative tasks and paperwork.

11. Some teachers also contribute to methodology development.

In addition to teaching, educators also pursue professional development. An educational system as a whole provides certification for employees and identifies talent.

Here is what AI Teaching Assistants can do now or in the near future:

• Content Delivery and Assessment: Deliver instructional content, assign practice work, test comprehension, and accurately flag knowledge gaps. This covers roles #3, 4, and some of 2, 5, 9, and 11 above.

• Limited Remediation: While they can flag knowledge gaps, filling these in is an iterative process requiring human guidance. This encompasses only some of role #5.

• Engagement: AI is learning how to motivate and emotionally engage students. This makes tentative inroads into role #1.

• Administrative Support: They can handle most paperwork, aiding in role #10.

• Development of Evidence-Based Pedagogy: Collect and analyze the detailed data on student learning efficiency; collaborate with human methodologists, aiding in role 11.

But there are critical activities AI cannot handle:

• The human encouragement, nurturing, and discipline students need to thrive academically and emotionally. This is core to roles #1 and 8.

• Designing a personal learning plan optimal for student's goals, level, interests, and learning style. This is the key to role #2.

• Leading real-world learning experiences like group projects, lab experiments, and field trips to enrich education. This is role #6.

• Providing childcare and ensuring student safety. Role #7 will remain solely the domain of people.

As we can see, human teachers are essential. As teaching assistants never replaced university professors, AI learning tools won't replace school teachers. We'll automate tasks, not jobs. 

This was about an ideal teacher, competent, caring, and having time and resources to do their job. The sad reality is that many teachers aren't like that. For example, remote villages and smaller towns in developing countries, rural areas, and underprivileged neighborhoods in many developed countries often lack qualified teachers. Another example: Many foreign language learning platforms employ tutors of a smaller caliber and don't give them much agency. All these "gig tutors" do is role #3 - deliver learning materials and exercises. But this is what an AI TA is for! 

Rather than replacing teachers, TAs would augment them, allowing teachers to:

• Shift lectures to asynchronous, prerecorded, localized videos from the most talented global teachers per subject. We already see this happening on Coursera, EdX, and YouTube.

• Delegate individual student practice to AI TAs, providing instant feedback and explanations as needed.

• Focus precious human guidance time on addressing complex knowledge gaps beyond TA abilities.

• Lead small group projects, hands-on experiments, field and VR tours, and other collaborative work.

• Spend more time designing high-level learning plans, providing motivational mentoring, working closely with parents, and connecting students with specialists and peers as needed.

High-stake exams would be administered by TAs and proctored by AI and humans.

AI Teaching Assistants can augment and empower human teachers and radically change the educational system. When thoughtfully deployed, this technology can finally move us away from the 19th-century industrial format of education to personalized learning, previously available only to the elite.

Risks and Challenges

While promising, deploying AI teaching assistants at scale raises many concerns:

• AI Limitations: Fundamental models that these solutions rely on still need to overcome serious challenges, like hallucinations and value alignment.

• Pedagogy Development: Curricula need to be revised with new capabilities in mind. Educators need to be trained in the use of new tools. Teachers and students should be taught basic data literacy skills. Educators and students will need a mechanism for providing feedback on the AI tools, which can then be used for iterative improvements.

• Infrastructure Needs: Not every school has the tech infrastructure to support AI. What kinds of investments are needed to make this vision a reality?

• Equity Divides: How do we ensure that EdTech doesn't further stratify educational outcomes? Will only affluent schools have access to advanced AI TAs? What does this mean for public education and education in less affluent countries?

• Commercialization: Commercial interests might distort educational priorities. For example, could educational content be influenced by the highest bidder?

• Bias and Ethics: Whose values and worldviews get embedded in AI systems? What are the implications for diversity, equity, and inclusion?

• Job Displacement: Could increasing reliance on AI reduce demand for human teachers?

• Data Privacy and Security: How can we protect student data collected by AI?

• Depersonalization: Amidst hyper-personalized content, how do we retain essential human relationships in education?

• Policy Barriers: What existing regulations might hinder implementing AI in schools? What new policies might be needed?

These are incredibly complex challenges with no easy fixes.
But I believe that with care, wisdom, and collaboration, we can navigate them. 

What About Self-Study Apps?

Many current EdTech applications seek to deliver affordable self-directed learning without involving human teachers or coordinating with the school curriculum. These tools have democratized learning but come with their own set of challenges. I think the efficacy and, thus, applicability of these apps is limited – as we discussed earlier, human teachers bring unique critical capabilities, which self-study apps lack.

Self-study apps are convenient and affordable and excel in particular scenarios such as:

• Language Vocabulary Building: Apps like Duolingo are effective for learning basic vocabulary and phrases, providing a solid foundation for language learners.

• Skill Reinforcement: Quizlet's flashcards can be a powerful tool for reinforcing knowledge, particularly for rote memorization subjects like history dates or medical terms.

• Initial Skill Acquisition: For beginners in a subject, self-study apps can offer a non-intimidating entry point. ELSA Speak, for instance, can be useful for novices learning English pronunciation.

• Supplementary Material: Self-study apps can provide extra practice exercises and resources to enhance the learning experience.

In these scenarios, self-study apps can be effective learning tools, especially for motivated and disciplined students. Still, even within the scope of these scenarios, they often struggle to address student knowledge gaps and misunderstandings. Self-study apps may also lack the depth needed for mastering complex subjects, often providing a surface-level grasp. The one-size-fits-all approach can sometimes be ineffective for learners with specific needs or learning styles. As a result, these apps generally suffer from high attrition rates as they rely heavily on self-motivation, which is often insufficient to sustain required long-term engagement.

While beneficial in the above contexts, I don't believe self-directed apps can replace human teaching at scale. But, they have an essential role in supplementing human-led education.

First Green Shoots of the New Classroom

Schools started adopting digital technologies a while ago with recorded lectures, digitized exercises and assessments, and later added remote instructors, like those provided by Elevate K-12 or Proximity Learning.

We're in the early stages of the new EdTech wave enabled by LLMs. Some promising AI teaching assistants are emerging:

• Multi-subject generalists like Khan Academy's Khanmigo and Quizlet's Q-Chat

• Specialized foreign language AI tutors like Duolingo Max, Speak, ELSA Speak, Buddy.ai, and SmallTalk2.me

If you haven't seen these yet, take a look - I bet you'll be amazed. Also, if you haven't done this already, try learning something newby talking with a plain-vanilla ChatGPT, Claude, or Bard. I think you'll be impressed but will also see their limitations.

In fact, since last school year, students and teachers have already been using LLMs like ChatGPT to help design and perform class assignments. For example, check out these overviews by Laurence Holt and Ethan Mollick. Historically, such bottom-up guerilla-style adoption bodes well for a new technology. 

Neal Stephenson's sci-fi Young Lady's Illustrated Primer doesn't look fanciful anymore!

So What?

Will these new technologies and approaches fit into the existing educational system? No. Truly transforming education likely requires rethinking schooling norms and procedures. This heavy lifting requires political will, resources, and educator participation. What segments of education will be early adopters? What countries? I don't know.

How much of the heavy lifting will be done by fast-evolving foundational model vs. education-specific models and solutions? Again, at this point, I don't know.

But we can make progress by creating modern solutions that complement today's educational institutions and practices. If these yield superior outcomes cost-effectively, they may catalyze systemic changes. Let's shape the future trajectory of AI in education wisely:

• Educators: Thoughtfully experiment with AI as a teaching aid and share feedback.

• Technologists and Entrepreneurs: Innovate with ethics top of mind. Collaborate with teachers.

• Policymakers: Reconsider the regulatory frameworks to facilitate the introduction of AI in educational settings. Encourage fair and equal access to new technologies in educational institutions.

• Parents and Students: Understand what technologies are being integrated into the learning process and how you can best benefit from them.

AI offers opportunities to radically improve education worldwide.
But we must innovate prudently and inclusively.

What's your take on the future of AI in education?