AI and the Future of Teaching and Learning

In May 2023, the US Department of Education released an important 67-page report on “Artificial Intelligence and the Future of Teaching and Learning” (in the Public Domain). A large committee of teachers, administrators, researchers, technologists, policymakers, and shareholders worked on this report for several months, and its thoroughness and thoughtfulness is unparalleled thus far by anything else released by similar entities like MLA, the Chronicle of Higher Ed, UNESCO, etc.

I believe I am right in assuming that many of my readers do not have the time to plough through the lengthy document, and so I am continuing my “Research Summary and Analysis Series” in order to provide access to some of the most important conclusion in the report and perhaps entice a few of you to pick it up and read it for yourself. Check out the 1st part of the series where I summarize and analyze the MLA’s first working paper on AI and education.

This research report will proceed in two parts:

1. I will provide an analysis of the document in light of trends in the educational field now 4-5 months after the publication.
2. I will provide summaries of each section (with charts) for those readers who want a little more information.

Part 1: Analysis:

Share Educating AI

The current response to generative AI in schools in unfolding along 3 primary trajectories:

(1) High-level reporters like DOE, MLA, HigherEd, who tend to focus on more conceptual and theoretical issues, and academic researchers, whose findings will not be available for months or years,

(2) Ed-tech and educational training services who are looking to meet immediate needs of teachers and students and in most cases make a profit on new applications grounded in GPT tech or some kind of choice chat-bot mechanism, and

(3) Schools, districts, and individual teachers who are largely responding in piecemeal fashion, cobbling together policies, strategies, and methods in light of existing literature about AI in schools—very little of which addresses the kinds of technology currently available.

We educators need models that are more aligned with educational objectives, values, and purposes. Constitutional AIs like Claude show real promise, but ultimately, schools and districts will need something more “explainable.” Educators need to be let inside the operations of the foundation models they use in order to protect students’ privacy, ensure accessibility, balance out human responsibility and computer autonomy, just to name a few issues.

I say all this realizing that there is not much money to be made in the educational sector in comparison with other burgeoning markets for AI, and thus we educators will probably continue to wait for our specialized models.

We educators need training and products that help us address the question: How are these new technologies impacting our students? Ed-tech and ed-training have largely ignored reports like DOE’s as they continue to roll out products grounded in GPT architecture. The industry currently is focusing on teacher-oriented products promising to maximize back-end efficiencies.

In many ways, addressing the issue of teacher case-use is much easier in comparison to student case-use. Teachers are adults, who have already mastered the skill and competencies focused on in schools’ curricular pathways. Thus, questions about plagiarism or the encroachment of the development of nascent skill-sets are moot points. And yet, most teachers I talk with want assistance on the front-end–in the classroom–working with students. And not just a tool we can throw at kids to use for a glamorous tech moment.

Educators need an entirely new way of thinking about school that threads the proverbial needle between old and new competencies, skills, and purposes: an AI-responsive method or approach that creates a productive feedback loop between what was and what now is. ?

Unfortunately, an AI-based teaching assistant will probably not figure out this issue for teachers, schools, or districts any time soon.??

We educators can become an important extension of the research and development process. As we wait for scholarly articles to be published, we can carry out our own research work in the classroom, hesitantly, documenting every step of the way. Every lesson, prompt, assignment, and assessment serves as a small but important contribution to the larger project of developing AI-responsive instructional methods and approaches.

In the work we carry out, the gathering of quality data is of the essence. Only through the collection and analysis of good data about our learning methods and practices will we be in a position to modify our approaches and techniques and ultimately to report our findings in general ways in the educational communities we participate in. I find this last point to be the most exciting. With every new technological shift in the education landscape, instructors truly have the opportunity to be on the frontline in terms of research and development.

Here, the hope is the product designers will take our input seriously so the next round of foundational models will truly have teachers, students, and schools in mind.

Part 2: Detailed Summaries

Here is the Table of Contents for DOE’s report on “Artificial Intelligence and the Future of Teaching and Learning”:

1. Introduction:

In this section, the committee cites “three reasons to address AI in education now”:

1. “First, AI may enable achieving educational priorities in better ways, at scale, and with lower costs.” (2)
2. “Second, urgency and importance arise through awareness of system-level risks and anxiety about potential future risks.” (3)
3. “Third, urgency arises because of the scale of possible unintended or unexpected consequences.” (3) Share

2. Building Ethical, Equitable Policies Together

In this section, the committee establishes guiding questions and 4 foundations principles:

3. What Is AI?

• “Our preliminary definition of AI as automation based on associations requires elaboration.” (11)
• Then, the committee offers 3 extensions to the original definition: (1) “human-like reasoning,” (2) “an algorithm that pursues a goal,” and (3) “intelligence augmentation.”
• The committee spends extended time emphasizing the generative AI “model” reality.
• “AI models are like financial models: an approximation of reality that is useful for identifying patterns, making predictions, or analyzing alternative decisions.”
• “We want to bring to the fore the idea that every AI model is incomplete, and it's important to know how well the AI model fits the reality we care about, where the model will break down, and how.”

Share Educating AI

4. Learning

In this section, the committee recognizes that AI can enable adaptivity in learning by improving the way existing educational technologies “meet students where they are, build on their strengths, and grow their knowledge and skills” (18).

At the same time, the report recognizes that current foundational models are only working well in specific instructional and tutoring situations. Where tasks are more open-ended or less predictable, these technologies become “brittle and can’t perform well when the contexts change” (19)

The committee offers several “directions” for expanding “AI-Based Adaptivity”

• Shift rhetoric from “deficit-based” to “asset-based” (20)
• Shift from “individual cognition to include social and other aspects of learning” (20)
• Include formats for both neurotypical and neurodiverse learners (21)
• Shift from “fixed tasks to active, open, and creative tasks” (21)
• Shift from a focus on correct answers to an emphasis on additional goals (21)

Key Insight: Learning With and About AI (A Duality)

“So far, we’ve discussed AI systems and tools to support student learning and mastery of subjects like mathematics and writing. Yes, it is also important that students learn about AI, critically examine its presence in education and society, and determine its role and value in their own lives and careers.”

Important Questions about AI and Learning

“Is high-quality research or evaluations about the impacts of using the AI system for student learning available? Do we know not only whether the system works but for whom and under what conditions?

5. Teaching

In this section, the committee endorses the use of AI technologies to make teaching more efficient, equitable, and student-oriented. At the same time, the committee has several concerns about the introduction of AI into teacher’s work lives.

The report configures these concerns in terms of different “tension points.” The thrust of this section is that teachers, schools, and districts should strive to find an equilibrium between the two tension points.

First, the report cites a chart quantifying how much teacher work time is spent on tasks:

The first tension points (below) lie between “human control” and “computer autonomy.” The committee recognizes that “the middle ground is not one dimensional and involves many choices.” The report insists that “AI technologies must allow teacher monitoring, have protocols to signal a teacher when their judgment is needed, and allow for classroom, school, or district overrides” (31).

The second tension points (below) lie between “highly customized assistance” and “increased teacher surveillance.” Here the concern is that “an AI-assistant [would] capture data about what [teachers] say, what teaching resources they search for, or other behaviors” and that data could be made available to administration for evaluative and disciplinary purposes or private corporations for undisclosed purposes.

The third tension points (below) focus on “responding to students’ strengths” and “fully protecting student privacy.” AI tutoring systems would have unparalleled access to real-time learning data on individual students. The committee has concerns that AI will collect demographic data about students and will in turn be used to create algorithmic biases about how to tutor or instruct individuals who belong to specific groups or identity categories.

Key Insight: Explainability

“Based on the needs of teachers (as well as students and their families/caregivers), we add another layer to our criteria for good AI models: the need for explainability. Some AI models can recognize patterns in the world and do the right action, but they cannot explain why…This lack of explainability will not suffice for teaching” (34-35).

6. Formative Assessment

This section focuses on the insertion of AI in existing ed-tech software used for formative assessment. Predictably, the committee insists that AI can help make assessments for individualized, but that humans need to remain in the loop through the process of assessment.

The committee is interested in the possibility of automated essay scoring. At the same time, the committee worries about how the pervasive use of AI might further consolidate the biases that already exist, for instance, in the college admissions essay assessment process.

7. Research and Development

In this section, the committee considers several questions about research and development of foundational models designed specifically for educational settings.

First, the committee calls the development of models “aligned” with explicitly educational goals, signaling that the current models are out of sync with educational environments and objectives in some core ways (see Section 5, “Explainability [34-35]).

Second, the committee calls for models that steadily adapt to wider variability in student context, strengths, needs, learning style, etc. Here, the committee is concerned that the initial foundational models for schools will be built primarily with neurotypical learning as a target audience.

Third, the committee hopes that AI can be used to re-design teacher professional development into something more individualized, more efficient, and more engaging for participants.

The chart below maps out a possible future in terms of “the long tail of variability” where foundational models become more accessible to more students over time.

Important Questions:

Are AI systems moving beyond the tall portions of the “long tail” to adapt to a greater range of conditions, factors, and variations in how people learn?

To what extent are AI technologies enhancing rather than replacing human control and judgment of student learning?

Share Educating AI

8. Recommendations

The committee concludes with 8 different recommendations. These recommendations combine more conceptual objectives with more concrete ones.

Thanks for reading!

Nick Potkalitsky, Ph.D