Discover more from AI Supremacy
LLMs and Instructional Applications, Challenges, and Approaches, a guest post by Nick.
Generated by Prodia
I get a lot of Email replies from readers hopeful to find more information about A.I.’s impact on education, students, teachers, best practices and so forth. I have heard your concerns and I share your interest on this matter. We are lucky on this platform to have people actively writing in this domain and luckily I bumped intoPhd. His new Newsletter on Substack is called Educating AI.
Let's figure how to best integrate and implement generative AI into today's classrooms!
Based out of Dayton Ohio, Nick is a multifaceted educational professional, serving as an educational researcher, curriculum developer, classroom instructor, and rhetorical theorist. I admire his academic rigor, philosophical perspective and his writing style reflects that and hopefully may be of use to some parents, teachers, counselors, professors and concerned citizens around this important topic.
Our 👧🏻 kids really matter, and how students are exposed to AI is top of mind these days. As such, Education in AI is going to be one of my ongoing series. So expect more blogs and essays on this topic moving forwards.
If you value this, you can support the channel:
⭐ To afford to keep some issues free on this Newsletter, I’m also trying to find the highest quality Sponsors that are aligned with our A.I. enthusiasm directly. Things we might actually want to read or use.
📶 From our sponsor: 📶
Designing Large Language Model Applications | Download the early release chapters of this O’Reilly generative AI ebook, compliments of Mission Cloud.
Transformer-based language models are powerful tools for solving various language tasks. With this book, you'll learn the tools, techniques, and playbooks for building valuable products that incorporate the power of language models.
BySummer of 2023
SELECTION OF NICK POTKALITSKY’S ESSAYS:
These articles intersect with Education in AI’s core topics.
An exploration of the human-machine network as the foundation of SEL curriculum
A characterization of prompt engineering as a rhetorical exercise focused on analysis of task, style, and audience
An exploration of in-context learning and “temperature” setting
A characterization of higher ed’s near total silence about student use of LLMs for college applications as setting an implicit curricular agenda for K-12
A “grabbag” of tools, methods, and resources for educators new to LLMs and in search of a way forward…
LLMs and Instructional Applications, Challenges, and Approaches
Do you know any teachers, parents, professors, academics, counselors or others who might be interested in this topic?
Purpose of the Report
This comprehensive report serves multiple crucial purposes:
Overview of LLM Potential: It provides an in-depth exploration of the potential applications of large language models (LLMs) in educational settings.
Challenges Assessment: The report delves into the challenges that LLMs pose to instructional settings, offering a detailed examination of these obstacles.
Recommendations for AI-Responsive Education: Furthermore, it presents a series of practical suggestions for the development of AI-responsive instructional methods and approaches, bridging the gap between theory and actionable strategies.
Within this report, I undertake an initial examination of the potential applications of LLMs within instructional settings and the complex challenges they present to teachers, schools, and districts. This report lays the foundation for more extensive work on my Substack, Educating AI, where I am actively constructing a comprehensive instructional approach and curriculum and simultaneously working on a manuscript of the same title. See upcoming posts for more detailed explorations of the principles and recommendations overviewed in Section 4 of this report. Learn more.
Potential Applications of LLMs
LLMs exhibit remarkable versatility, with potential applications that encompass:
Reflective Tools: Serving as writing assistants or personal tutors, aiding students in improving their writing skills.
Creative Tools: Functioning as text generators and scenario creators, fostering creativity within instructional settings.
Communicative Tools: Proficiently translating and interpreting text, breaking down language barriers in diverse educational environments.
Challenges Presented by LLMs
However, the adoption of LLMs in education is not without its challenges:
Plagiarism Concerns: The ease with which students can use LLMs for plagiarism poses a significant hurdle, hindering the development of essential skills, competencies, and literacies.
Shift in Authorship Concepts: LLMs challenge traditional notions of authorship and originality, often perplexing many students.
Privacy and Data Issues: These models operate with limited regulation, collecting private user data and sharing it for profit, even when dealing with users as young as 13
Bias and Stereotype Transfer: Persistent biases, prejudices, and stereotypes from LLMs' pre-training data can manifest in their responses, perpetuating harmful stereotypes.
Digital Divide: The commercialization and privatization of LLMs contribute to the digital divide in educational institutions, exacerbating technological disparities.
To address these challenges effectively, the report emphasizes the following curricular recommendations:
In-Depth Curricular Analysis: Teachers should conduct thorough analyses of existing content and teaching methods in light of recent advancements in generative AI technologies.
AI-Unassisted Spaces: Educators should establish AI-unassisted spaces within courses and classrooms to encourage students to develop foundational skills without overreliance on AI.
Discipline-Specific AI Training: Teachers should incorporate discipline-specific training on best practices for using generative AI, preparing students for real-world applications.
Critical Examination of LLMs: LLMs should become subjects of critical study at all educational and institutional levels, ensuring that human oversight remains integral to their operation.
Personalized Learning: LLMs offer a unique opportunity for personalized instruction and universal access to specific curriculum materials.
Student Involvement: Students should actively engage in redefining the purpose of education in light of the evolving and unpredictable landscape of generative AI technologies.
In conclusion, educators find themselves at a crucial juncture, requiring a balanced response to the potential of LLMs while remaining cautious due to uncertainties about their long-term effects. The report recommends a policy of gradual, measured changes, where practitioners meticulously monitor outcomes and leverage findings to inform subsequent adjustments. This approach ensures that education evolves in tandem with the ever-advancing field of generative AI.
2. Applications of LLMs in Educational Settings
Potential applications of LLMs
The potential applications of Large Language Models (LLMs) are incredibly diverse and multifaceted, as detailed in existing literature. To better organize these applications, they can be categorized into three distinct groups: reflective, creative, and generative (Figure 1).
Figure 1. Differences that teacher and students can anticipate between existing and future technologies. Department of Education, “AI and the Future of Learning and Teaching,”16.
"Reflective" applications encompass the LLM's ability to serve as a reflective tool, such as a writing assistant or a personal tutor. Notably, LLMs like ChatGPT are highly adaptable, enabling users to define their specific function, whether as a reader, analyst, advisor, editor, revisor, rewriter, debater, or de-constructor. Users can modify these functions through subsequent commands or interactions. For instance, Rohan Mehta, a first-year college student, employs ChatGPT to “discuss” his writing “out loud…albeit with a machine.” Mehta continues: “[U]sing ChatGPT to verbalize the space of possibilities– from the scale of words to paragraphs– strengthened my own thinking” (The Download, 21). Nevertheless, questions emerge regarding the criteria for evaluating the quality of advice from the LLM and the appropriateness of the inquiries it poses. To maximize the benefits of LLMs in education, teachers must model appropriate interactions and help students establish criteria for assessing the LLM's guidance. Demystifying the LLM's role as an authoritative source and emphasizing its function as a tool are essential steps in this process.
Furthermore, LLMs prove to be excellent tutors, particularly in contexts where academic content is concrete and can be broken down into discrete steps, such as mathematics and specific writing and reading tasks. Jiahong Su and Wiepeng Yang emphasize that AI tutoring as a process “adapt[ed] to the learners’ level and pace, thereby providing a highly customized learning experience.” This adaptability is especially valuable for learners who lack access to in-person tutors or prefer self-paced learning. For example, a significant initiative by Khan Academy and Instructure aims to make AI tutoring programs accessible through Canvas, one of the most widely used Digital Learning Systems in the United States. While ambitious, the goal is entirely feasible given the adaptability of the GPT architecture to new contexts, tasks, and purposes. Importantly, these projects aim not to replace educators but to extend educational support into spaces where traditional instructors may be unavailable, enabling all students to make significant progress toward their individualized educational goals.
The "creative" applications of LLMs are particularly intriguing due to their capacity to generate substantial amounts of creative text and imagine scenarios for various educational purposes. While LLMs are known for their text generation abilities, this report focuses on their creative writing potential within instructional settings. Remarkably, LLMs feature a built-in creativity modulation setting known as "temperature." The "temperature" setting, typically at 0.7 by default, can be adjusted to control the level of variability in the generated text. For instance, a user can type "temperature 0.9" at the end of a prompt to encourage the LLM to produce more expansive, elaborate, or even "hallucinatory" associations, or conversely, "temperature 0.2 or 0.3" for more focused responses. It's important to maintain a critical stance toward LLMs' potential to generate non-real scenarios within real contexts, an issue extensively reported in previous articles. However, educators can use temperature settings strategically to harness LLMs' creative capabilities for specific tasks, such as introducing students to different literary genres. For example, a teacher can use ChatGPT to generate small, accessible examples of various genres, fostering interactive engagement and emphasizing meta-cognitive skills and competencies. In such instances, the LLM serves as a "computational" or "generative tool" rather than a "research database."
Moreover, researchers have explored the potential of LLMs to create immersive, primarily text-based scenarios for instructional purposes. These simulations are at once content-drivers and meta-critical activators. In other words, students learn content through immersion in simulations, while engaging actively in evaluation of the limitations, biases, and inaccuracies at work in simulated content. While these applications are largely speculative at the moment, their possibilities become even more intriguing when combined with AI-driven imagery generators and/or virtual reality (VR) goggles.
Benjamin Breen of the Substack Res Obscura is already doing incredible work with text-based simulations in his college history courses. Simulations can also take the format of virtual debate partners where students simultaneously build knowledge of content and engage in meta-critical reflection about limitations, biases, and overgeneralizations. These creative applications aim to transform teaching by promoting creativity in both instruction and assessment. According to Jessica Stansbury, the director of teaching and learning at the University of Baltimore, LLMs are exciting because they have the potential to shift teachers' roles from being "gatekeepers" of information to becoming "facilitators" (Heaven, 46).
The "communicative" applications of LLMs revolve around their capability to translate and interpret vast amounts of text rapidly. In U.S. schools, where over 400 languages are spoken (Figure 2), adapting materials to support students in achieving shared academic goals is crucial. Traditional ESL (English as a Second Language) instructors often have limited familiarity with their students' primary languages. While familiarity with these languages is not a prerequisite for facilitating student success, it can enhance the fluidity of communication and provide an additional avenue for engagement.
Figure 2. “Top 20 Languages on States’ ‘Top Five’ Lists of Languages Spoken by K-12 ELs: School Year 2019-20,” Office of English Language Acquisition.
Since 2017, deep learning and neural network-based AI have transformed online translation tools, offering real-time translations of student speech with up to 90% accuracy. As GPT-model AI continues to advance, translation capabilities are expected to improve further in terms of speed, accuracy, and accessibility.
Expensive specialized translation software may become unnecessary, as both teachers and students will only require access to the same LLM. However, the growth in translation capabilities depends on the availability of language materials in the pretraining dataset and the prioritization of specific languages by technology developers. For instance, as of now, OpenAI's ChatGPT 3.5/4.0 is proficient in only English, Spanish, French, German, Portuguese, Italian, Dutch, Russian, Arabic, and Chinese.
World language educators can seamlessly integrate LLMs' translational capacities into their curricula. At the highest level of instruction, teachers can instruct students to direct an LLM to translate all of its responses into a target language and can then teach a lesson the rhetorical situation of prompt engineering in the target language. At a more modest level of instruction, students can be tasked with comparing translations generated with different temperature settings to identify variations in word choice and cultural context. These assignments emphasize that all translation is interpretation, and with proper handling by the teacher can be used to address a common critique of AI translators, namely their insensitivity to cultural nuances.
In conclusion, LLMs offer a multitude of potential applications in education, spanning “reflective,” “creative,” and “communicative” domains. Their adaptability and capabilities have the potential to enhance the learning experience and expand educational opportunities. However, it is essential to approach their use with thoughtful consideration and guidance to harness their full potential effectively.
3. Challenges of LLMs in educational settings
LLMs, while offering numerous opportunities for teachers and students, also present a host of challenges, many of which are deeply rooted in the design and construction of these products. The Department of Education's 2023 Report on "AI and the Future of Education" underscores that LLMs are often described as "incomplete models" of the world. They rely on "limited" and "selective" datasets, and many of their challenges stem from these limitations in the datasets.
It's also crucial to remember that the current generation of LLMs is not explicitly designed for educational contexts. However, there are ongoing efforts to develop more educationally-aligned LLMs. Claude 2.0, as a constitutional generative AI, holds great promise in this regard, with more specific training and further adaptation.
Unfortunately, the development of these education-focused LLMs is progressing slowly due to the limited economic returns they promise in the educational sector. In this report, we will delve into five key challenges posed by LLMs: (1) Academic Integrity, (2) Concept of Authorship, (3) Privacy Concerns, (4) Cognitive Bias and Representation Issues, and (5) Accessibility and Equity Concerns.
The introduction of LLMs in educational settings has raised widespread concerns, especially surrounding issues of plagiarism and academic integrity. To provide some context, it's important to recognize that many American students continue to face challenges in staying engaged and completing assignments. Disengagement levels have spiked as schools have returned to more "normal" or "traditional" instructional approaches. Students are grappling with high levels of stress, emotional disturbance, and mental health concerns, largely stemming from the collective trauma of the worldwide health pandemic.
For students from underrepresented groups, the pandemic has only exacerbated existing disparities, gaps, and inequities in education. The prevailing culture of standardized testing continues to permeate most American schools, and the increasingly competitive college admissions process has transformed high school into a point-scoring and grade-collecting endeavor for most students.
Given this context, many students are actively seeking ways to cut corners, save time, and gain a competitive edge over their peers. LLMs like ChatGPT, Poe, and Claude present a strategic opportunity to work smarter, faster, and harder in an effort to keep up with the increasingly professional-level work demands of today's secondary school experience (Figure 3).
Figure 3. “How Often Do You Use ChatGPT to Help You Complete Written Assignments.” Intelligent, Feb 2023.
However, despite these potential benefits, most teachers remain cautious about allowing students to use LLMs. They are concerned that these tools might hinder their ability to assess students' actual knowledge, writing skills, and competency. When a student submits a paper, is the teacher evaluating the student's skill at writing or the student's skill at working with an LLM? While some ed-tech and ed-training firms, as well as some high school and college-level instructors, are endorsing the full-scale embrace of LLMs as primary tools for writing instruction, most educational institutions are grappling with how to maintain spaces for unassisted work while also integrating new curriculum on the use of generative AI for specific purposes.
The resistance to adopting LLMs in schools appears to have two main dimensions: (1) institutional concerns about equity, privacy, and accessibility (Whose stories do LLMs privilege? Who has access to LLMs?), and (2) pedagogical concerns about the effectiveness of such a methodology (Will our students truly develop writing skills and competencies if operating in a fully-assisted context?
How do we assess learning when students generate work with the full assistance of AI?). Thus far, no comprehensive approach for maintaining unassisted spaces has emerged, leaving students very confused about what is expected of them. This report anticipates that it will take considerable time to establish some form of equilibrium in this rapidly evolving landscape. The silver lining is that a paradigm shift is already underway, and even as more powerful LLMs become available, teachers, schools, and districts will have an emerging set of tools that can be carried over to new products.
Concept of Authorship
In academic and instructional settings, the concepts of authorship and originality play pivotal roles in defining how humans interact with specific texts, documents, and projects. LLMs have significantly altered the landscape surrounding these concepts, requiring educational institutions to respond with appropriate adjustments in their school policies, academic handbooks, syllabi, classroom expectations, and teaching methods. Over the past thirty years, AI has made notable strides in the realm of published texts through functionalities like spellcheck, grammar check, Grammarly, etc. However, LLMs like ChatGPT, Poe, and Claude represent a qualitative leap toward shared authorship or co-responsibility.
This raises crucial questions about how students and faculty should denote their reliance on LLMs during the writing process. Moreover, it prompts an intriguing question regarding the handling of texts and compositions that are entirely generated by LLMs. (Importantly, even in these cases, there is some level of human intervention, particularly in setting specific parameters for the LLM at the point of prompt engineering.) At present, most scholarly journals do not publish LLM-generated content.
A meta-study of such journal articles recently concluded: "ChatGPT is not a useful tool for writing reliable scientific texts without strong human intervention. It lacks the knowledge and expertise to accurately and adequately convey complex scientific concepts and information" (Blanco-Gonzalez qtd. in Homolak 23). Regardless of one's opinion on the quality of AI-generated scholarship, students need a clearly articulated policy about the use and citation of text produced with the assistance of LLMs. These policies will necessarily involve implicit or explicit theoretical and practical commitments on the part of faculty and administration to particular ideas about the nature of authorship and originality.
UNESCO reports that ChatGPT and other LLMs currently lack substantial regulation. In the United States, legislation related to social media and AI technology has lagged significantly behind the rapid development of these products. OpenAI acknowledges concerns about the late-stage development of Artificial General Intelligence (AGI) and the competitive race it may engender, but concrete details about an action plan, including tangible steps toward specific goals, are conspicuously absent from OpenAI's website.
The combined issues of a lack of regulation and privacy concerns prompted Italy, in April 2023, to become the first country to block ChatGPT for six months while protective legislation was drafted. Given the likelihood of continued Congressional and legislative gridlock in the US surrounding the development, commercialization, and consumption of social media and assistive-AI products, the responsibility for developing guidelines for appropriate use and safeguards for protecting the youngest and most vulnerable users will continue to rest largely on individuals and smaller institutions, such as families, schools, and districts.
Cognitive Bias and Representation Issues
The information generated by ChatGPT and other LLMs is essentially "tertiary" source material. This designation distinguishes ChatGPT content from primary and secondary source materials. A "tertiary" source is essentially the product of amalgamated primary and secondary source materials that have been "scrubbed" and reprocessed through these applications' powerful language transformers. This raises a crucial question: Do any biases, preconceptions, stereotypes, or other elements from the original data cling to the language and content even after it is "scrubbed" and reprocessed for the generation of new AI-written texts and responses? Early studies suggest a strong affirmative answer to this question.
Cognitive biases and misrepresentations of gender and diversity present in the original dataset, such as The Common Crawl, are definitively being transferred in subtle but predictable ways into the texts and responses generated by ChatGPT and other LLMs. An article in the Journal for International Affairs illustrates one aspect of ChatGPT's cognitive bias.
For this article, the editorial board of the journal engaged ChatGPT in a conversation focused on the question, "Why is the world in 2023 insecure?" In its response, ChatGPT exhibits an overtly pro-democratic and incipiently pro-capitalist bias when addressing problems like geopolitical conflict and the accelerating environmental crisis. Instead of addressing how capitalism may be implicated in the current arms race in Ukraine or the rise in greenhouse gasses worldwide, ChatGPT offers blanket reassurances that "coordinated and collaborative effort by governments, civil society, and the private sector" will hopefully "ensure that people feel secure and that institutions are able to respond effectively to the challenges of our time" (381).
More seriously, ChatGPT continues to fail many measures that track racial and gender bias. A recent test by Sayash Kapoor and Arvind Narayanan (AI Snake Oil) focused on ChatGPT's handling of the WinoBias benchmark's anti-stereotypical questionnaires related to gender identity markers. These questionnaires present participants with questions about an unspecified "paralegal" or a "lawyer" and encourage responses using pronouns.
For example, a stereotypical response for a paralegal might involve using "she" pronouns. Kapoor and Narayanan's findings were concerning. They concluded that "both GPT-3.5 and GPT-4 are strongly biased, even though GPT-4 has a slightly higher accuracy for both types of questions. GPT-3.5 is 2.8 times more likely to answer anti-stereotypical questions incorrectly than stereotypical ones (34% incorrect vs. 12%), and GPT-4 is 3.2 times more likely (26% incorrect vs. 8%)."
These issues of bias are further compounded by the fact that LLMs produce tertiary sources, meaning that the transformation process makes it impossible to track the bias and assumptions back to the original source material. As a result, students using information gathered by LLMs like ChatGPT find themselves in a precarious situation, working with source materials that are haunted by tones, inflections, voices, biases, and rhetorical situations that cannot be established, tracked, or analyzed in any one-to-one correlation. In the words of one teacher, "At least with traditional sources, you can track the biases specifically back to a named and identifiable source. With these LLMs, the bias is covert, has the aura of authority and objectivity, and is thus all the more dangerous.”
Accessibility and Equity Concerns
Despite various local, state-wide, and federal initiatives over the past thirty years aimed at bridging the technology gap across America's diverse school districts, the uneven distribution of computer technology and access to the internet remains a persistent issue, particularly in under-resourced schools and educational institutions. The COVID-19 pandemic exposed this technological gap further, as resource-rich schools had the tools to pivot flexibly to online and digital instruction, while at-risk institutions lacked basic resources, materials, and technologies to deliver instruction to students who were often already several years behind their peers in more affluent districts and schools.
A recent study in the Harvard Political Review underscores the broader implications of this technological gap in terms of long-term employability and income inequality. According to Alyvia Bruce, "educational inequality" seamlessly transitions into "income inequality" as "the digital divide has especially pronounced effects on children, whose socioeconomic situation determines their ability to engage with increasingly technology-reliant educational materials." In this context, LLMs like ChatGPT, Poe, and Claude enter the scene.
To utilize these applications effectively, students must have access to certain technology hardware, computational resources, and computer science literacy. It is worth noting that schools grappling the most with plagiarism and academic integrity issues due to LLMs tend to be in resource-rich districts, independent/private schools, or colleges where technology resources are more accessible. Although OpenAI has committed to maintaining a free version of ChatGPT (3.5), the commercialization of a premium product (4.0) with more reliable access and service creates a real-time equity and accessibility concern.
While ChatGPT 4.0 subscriptions did not sell as anticipated during the spring and summer of 2023, but as OpenAI partners with more corporations and embeds its technologies in other applications and platforms, users of the free version will experience increased slowdowns, delays, and interruptions with ChatGPT 3.5. This, in turn, may encourage those who can afford it to subscribe to the $20 per month ChatGPT 4.0 or seek out other paid services. These dynamics play out in schools, resulting in groups of students who can afford specialized, premium technology gaining an edge, while others rely on a slower, free version of the LLM or have no access at all.
Outside of school, these dynamics significantly impact college admissions processes. Students with access to the best LLM products and high levels of technical competency will have a significant advantage during the 2023-24 application year. Colleges and universities, unfortunately, have largely kept silent about the use of LLMs for college admission essays, contributing to inequities that have plagued the college admissions process for generations. Addressing these disparities remains a complex challenge, one that colleges and universities continue to place on the shoulders of individuals and smaller institutions, including families, schools, and districts. See my recent post on the college application process.
4. Recommendations for AI-Responsive Education
To provide practical recommendations for the development of AI-responsive instructional methods and approaches, it is crucial to navigate the intricate landscape where potential benefits intersect with real challenges. This endeavor involves numerous stakeholders across multiple levels, each requiring adaptable strategies and methods that can evolve alongside technological, social, political, and legal vectors, thresholds, and developments. However, the very complexity of this task poses a risk: a well-intentioned response may inadvertently create further challenges or ethical dilemmas, leading to pedagogical inconsistencies or contradictions within or across the overall project.
In the current educational landscape, various actors are deeply involved in developing AI-responsive instructional methods and approaches:
1. Federal and state departments of education are actively crafting policy briefs and initiatives aimed at guiding AI integration and implementation. Their primary focus is on "keeping humans in the loop" to counterbalance algorithmic biases.
2. Academic researchers and think tanks are conducting studies on short- and long-term efficacy of AI implementations both inside and outside the classroom. While few findings have been published, more are expected in the coming months.
3. Ed-tech firms and education training organizations are developing training and tools to assist teachers in incorporating AI into their classrooms. However, their product-driven approach can sometimes overshadow broader strategic thinking and solutions.
4. Superintendents, administrators, and school leaders are seeking top-down policies to frame effective AI utilization in schools. Often, they lack the necessary resources and exemplars to formulate these policies.
5. Teachers, teacher assistants, and aides are working to develop AI-responsive strategies both within their individual practice and to enhance the student experience. They frequently do so without formal training or guiding principles.
6. Parents, guardians, and caretakers are assisting students in making informed decisions about AI use during assignments, often in the absence of clear AI-usage policies or agreements.
7. Students are immersed in a world where AI discussions abound, yearning for guidance, policies, curriculum, and instruction on how to best utilize these applications in their daily lives.
Creating a system or approach that accommodates the diverse needs and complexities of these stakeholders amounts to a Herculean task, akin to a complete overhaul of the traditional educational paradigm. However, change is necessary, and it must occur incrementally, addressing each level of involvement systematically.
Instead of providing comprehensive solutions for every level of this intricate equation, this report will offer guiding principles for AI-responsive approaches and methods, along with a series of suggestions for educators, schools, and districts to explore and adapt in a contextually relevant manner:
4 Principles for AI-Responsive Education:
CONNECTION: Students and teachers need to be connected to each other through a human-machine network of accountability where humans stand out as the primary decision makers. Check out my recent post on the principle of connection.
EQUITY: Students and teachers need to engage with human-machine networks that privilege human wellness, dignity, equity, diversity, and inclusion to create stronger and more vibrant communities.
PURPOSE: Students and teachers need to define common and creative purposes to ground engagement with human-machine networks and to realize those purposes through meaningful action.
EXPERIENCE: Students and teachers need to experience all the facets of study, research, thought, and writing in order to innovate towards the solutions for the challenges of the global community in the 21st century.
6 Recommendations for AI-Responsive Education:
1. Curricular Analysis: Conduct a thorough analysis of existing curriculum and methods in light of recent advancements in generative AI technologies. Determine which core skills, competencies, and literacies must be retained alongside new requirements.
2. Cultivate AI-Unassisted Spaces: Foster AI-unassisted spaces within courses and classrooms, allowing students to develop fundamental skills and competencies that can be enhanced dialogically through the use of generative AI technologies.
3. Discipline-Specific AI Training: Train teachers to impart best practices in generative AI within specific disciplinary contexts, preparing students for professional and academic usage after graduation.
4. Critical Study of LLMs: Position Large Language Models (LLMs) as subjects of critical study at all levels of school administration, instruction, infrastructure, and assessment. Always ensure that "humans are in the loop" in processes involving student health, communication, instruction, or evaluation.
5. Privacy and LLMs: Acknowledge the potential of LLMs to individualize instruction and provide universal access to curriculum. However, address privacy concerns related to the collection of detailed, real-time data on students' cognitive patterns.
6. Student Involvement: Involve students in the process of redefining the purpose of education in light of generative AI technologies. Provide programs guiding students through a multi-year journey of discovering and defining their creative purpose, allowing discussions about AI usage to align with these contexts.
See forthcoming posts at my substack, Educating AI, for in-depth studies of each of these principles and recommendations.
Large language models have the potential to revolutionize curriculum development, instructional design, and assessment methodologies in unpredictable ways. As educators, administrators, students, and parents navigate these complexities together, schools should prioritize building an inclusive culture of transparency and accountability surrounding strategies and policies for integrating and utilizing these new technologies. Rather than limiting access to specific curricular tracks, a more dynamic approach involves critically embracing AI applications for the benefit of all students.
However, before widespread integration, schools must grapple with fundamental questions about the purpose of education and the essential skills and competencies required. With well-defined purposes and strong rationales, schools can create spaces and practices within classrooms where LLMs are or are not utilized, facilitating the development in specific skills, competencies, and literacies.
Although creating AI-responsive education poses a significant challenge to all stakeholders involved, our students are prepared for and indeed are demanding this transformation. Following the pandemic, they are seeking more efficient, purposeful, creative, and engaging forms of instruction and assessment. With the emergence of LLM technology, we have both the chance and the potential to cultivate the school environments, curricula, and teaching methods that align with our students' preferences.
To view more guest posts go here. To read the archives go here.
Nick’s dedication as an educator and researcher is directed towards the continuous enhancement of research methodologies, instructional techniques, and assessment systems in areas spanning language arts, history, media studies, and digital humanities.
Presently, Nick holds the position of a high school language arts teacher at a small independent school in Ohio, where he actively contributes to the school's endeavors in crafting an AI-responsive curriculum and policy. In his prior roles, he has been actively involved in teaching and curriculum development across various educational levels, private and public, ranging from middle school to college and graduate school settings. His educational background includes a Bachelor's degree in the Classics, a Master's in Education, a Ph.D. in American Literature, two teaching licenses in Ohio, and a Montessori certificate.