February 2026 – Learning Aloud

AI reduces skill learning

When a technology offers advantages and disadvantages, the decision-making process can be quite complicated, especially when oversight cannot be guaranteed. For example, many states now ban cell phones, making a use such as telling parents when the schedule for after-school activities has changed difficult. The advantages and disadvantages vary with the field of application and my interests have mainly been focused on education. Just to be clear, by this I mean learning in general, not just the type of learning that occurs under supervision or is associated with educational institutions.

The generic educational situation that raises concern involves tasks undertaken to encourage both skill development and knowledge, and includes a requirement that demonstrates that the task has been attempted by the existence of some product. In educational settings, such products might result from homework or class activities, or simply by visible demonstrations of activity. The issue with AI is that in many cases, such as problem sets or documents of various types, these same products could be generated by AI, avoiding the cognitive activity of the learners. The phrase “cognitive offloading” has been used to describe this alternative form of product creation. Teachers might simply call it cheating. Cognitive offloading itself can be a desirable or undesirable option, requiring decisions regarding when it is appropriate and efficient, and when it is a detriment.

While cognitive offloading to avoid learning tasks seems an obvious problem, little actual research exists to demonstrate the damage done. Some would argue that if technology can replace an activity and that technology is readily available, why bother to “learn” the skill in the first place? Why learn information if your cellphone can allow you to search for information when it is needed? Why learn basic calculation skills when you cellphone can also serve to do mathematical operations? There are responses to these challenges, sometimes offered by students or parents, but this analysis would take this post in a direction I did not intend.

Here, I want to focus on learning to write and writing to learn by discussing a different learning task. This may sound unnecessary, but at present, there is a reason to take this approach. The justification for being indirect is that writing is a complex skill consisting of multiple subskills, and we learn to become competent at even a basic level over years and not weeks or hours. We are investigating an alternative to the traditional methods of instruction that can be subverted now, and we cannot rely on experience to help us evaluate and tease apart how the development of subskills are impacted. The insights and evidence of the potential damage done would take to long to emerge. As one perspective, consider the lingering impact of COVID on learning. What about the move to online learning did we not anticipate and what consequences are we still trying to mitigate?

AI in Learning to Code

Shen and Tamkin had an opportunity to investigate the impact of AI with adult programmers learning to make use of a new library. Think of a library as a collection of functions (tools to perform specific and commonly used tasks). Instead of having to write code to accomplish common tasks each time a programmer encounters a need, libraries allow programmers to call prewritten code snippets. It takes some work to make use of a library – what functions are available, how do you call the function you want, what inputs and outputs are involved and how are these integrated with the code you write yourself? The researchers recognized that the learning coders had to do to make use of a new library provided an opportunity to study how AI could help and hinder learning a complex process.

Shen and Tamkin studied actual programmers as they worked to learn a new library. They suggested that the process be viewed as a tutorial including both background information and simple programming tasks. Programmers were assigned to a control and a treatment group, with the treatment group having access to AI. The learning phase concluded with an assessment evaluating multiple concepts and skills. Video of treatment group participants was collected to document how each individual used AI and worked on the programming exercises.

The researchers found that the treatment groups did not differ significantly in the time spent learning, which they found surprising. On the post-test, the largest group differences were in debugging skills. Smaller skill differences were found for code reading and conceptual understanding. Those without access to AI made more coding errors on the practice tasks, spent more time practicing debugging, and ended up with better skills on the outcome evaluation. How AI was used differed greatly with some simply asking AI to solve the coding challenges and others who only asked higher-level questions of the AI tool. Some users had the AI tool solve the coding challenges and then retyped the solutions themselves (rather than copying and pasting). This was not an effective strategy.

Generalizing from the coding study

I have spent considerable time both coding and writing and I have always found the processes to have similarities. While others may find this a strange observation, I have always said that coding and writing were the two professional tasks I learned I could not perform later in the evening if I wanted to get a good night’s sleep. Reading was fine. Grading was fine. Something about both coding and writing was cognitively stimulating, making it difficult to sleep.

The application of AI to complex skills is interesting, but difficult to study. Clearly, a single skill would seem very unlikely to be developed if a learner could completely substitute AI for practicing the skill. However, it seems possible that learning a multiple-component skill such as reading or coding might benefit from replacing specific components with AI under certain circumstances. We have limited cognitive capacity and substitution for some components of a complex task could allow the remaining components to receive more attention until well learned.

Learning to write might represent an example. I have often referred to Flower and Hayes’ writing process model when describing the components of writing and writing to learn. The use of AI to offer content to provide the basis for a writing task and perhaps even to offer a structure to guide the organization of a writing product could free up capacity to focus on lower-level skills such as spelling, grammar, and coherent paragraphs. In contrast, I typically use Grammarly while I write to allow to move more quickly while relying on this AI tool to alert me to possible spelling and grammatical improvements.

Part of what Shen and Tamkin observed in their qualitative observations of the different learner-imposed focus of AI and the relationship of differences to what was learned or not learned offers a related perspective. Debugging is an important lower level coding skill and having AI debug code appeared to limit a coder’s ability to debug when working without AI.

Suggestions for Learning to Write and Writing to Learn

AI can support both “learning to write” (developing writing skill) and “writing to learn” (using writing to deepen understanding), but depending on which writing skills are the goal best practices should differ.

Learning to write: skill development

Here AI should be thought of as a coach, not a ghostwriter.

Emphasize feedback: Tools like Grammarly give immediate feedback on grammar, syntax, cohesion, and organization, helping students revise iteratively while concepts are still fresh.

Structure and separate subprocesses: Generative tools can help students brainstorm ideas, outline structures, or identify expectations for different types of writing (e.g., sample introductions, transitions).

Process?first policies: “Writing first, AI second” approaches ask students to draft independently, then use AI for critique and revision. When coders used AI in the Shen and Tamkin, this is the general theme that seemed most successful.

Writing to learn: thinking with text

When the goal is conceptual understanding of content knowledge, AI is best used to amplify reflection, not replace it.

Clarifying concepts for the writer: Students can ask AI to reexplain readings, generate examples, or pose practice questions, then respond in their own words, using writing as a space to consolidate understanding.

Challenge personal understanding: AI can generate counterarguments, alternative explanations, or “what if” scenarios that students must address in writing, pushing them beyond summary toward analysis. Why do others disagree with the summary I am creating? What can I offer to support my position and what are the limitations of the alternative?

Shared design principles

There are some guidelines these goals for writing. Across both purposes, similar design choices matter.

Make process visible: Require artifacts – notes, outlines, draft histories, and brief process memos about when and how AI was used. Document the transition from any use of AI to products student has generated.

Align AI roles with goals: For skills (learning to write), let AI focus on feedback, exemplars, and mechanics; for content learning (writing to learn), keep generative help outside the main composing space and treat it as a prompt.

Previous analysis of technology and the writing process

Sources:

Flower, L., & Hayes, J. R. (1981). A cognitive process theory of writing. College Composition & Communication, 32(4), 365-387

Shen, J. & Tamkin, A. (2026). How AI impacts skill formation. arXiv preprint arXiv:2601.20245 (this study has yet to officially be published)

The Rise and Fall of the Twitter EdChat

For over a decade, the hashtag #Edchat allowed educators using Twitter to gather, share, and commiserate. However, as the platform formerly known as Twitter transitioned into X, the landscape of this digital discussion site shifted dramatically. Drawing from recent research, this post explores how educators have used #Edchat over time, the stressors inherent in this social media use, and the history of a community in transition. Researchers understand that for a variety of reasons, Twitter and Twitter chats are far less influential than they were a few years ago. They argue that studying the edchat phenomenon historically may have value for other social media platforms generally and specifically, should they hope to involve educators.

The Golden Era of #Edchat: Purpose and Participation

In its prime, #Edchat provided a means for informal professional development. Research by Willet (2019) and later Willet and colleagues analyzed hundreds of thousands of tweets to understand exactly how and why educators were using the platform. The study identified several key types of engagement:

Resource Sharing: The most common use case, where teachers curated and distributed lesson plans, EdTech tools, links, and articles.
Pedagogical Debate: Scheduled weekly chats allowed for deep dives into specific topics, from classroom management to the integration of AI.
Social Support: Perhaps most importantly, it provided a space for “digital social support,” helping teachers feel less isolated in their professional struggles.

This era was defined by a sense of “augmented intelligence,” in which the collective knowledge of the network enhanced individual teachers’ expertise.

2008–2014: The Golden Era of Synchronous Twitter Connection

The #Edchat phenomenon began in October 2008. This early era was defined by the weekly Tuesday night chat, a highly structured synchronous event that became a must for thousands of educators. An agreement on Tuesday night did not result from any official declaration, but once started, Tuesday night became the default for those wanting to participate in a common chat.

I encourage anyone interested in the topic of teacher use of social media to read the two references to Willet and colleagues I provide here. These researchers had access to what I have heard described as the Twitter firehose, which was available until 2023, and downloaded over 15 million tweets for analysis using the #edchat hashtag as used over 15 years. Unlike researchers who implemented projects of a much smaller scale and made use of human raters to classify and quantify characteristics of such chats, Willet and colleagues used data analysis tools that quantified specific characteristics (questions, responses, links, secondary tags, retweets) and even tools that attempted to identify themes based on terms appearing in the tweets. These characteristics were mapped against years to identify trends.

This approach allowed certain questions that could only be addressed by this massive scale. What trends could be observed over the history of the edchat phenomenon, but may have been overlooked in the data? For example, the way in which edchats evolved interacted with the capabilities of the Twitter platform. Tags were a user-applied innovation that was later integrated into the tool as a capability. Tuesday night became the impromptu time for edchats, which took on a formalized approach. A chat leader would provide a series of prompts identified as Q1, Q2, Q… and participants would respond using R1,R2, R… . Other tweets could be added within the rough synchronous time frame defined by the prompts. Because chats were stored, others might review the session at their leisure and add their own contributions.

Just to be clear for those unfamiliar with the reason for this format, this experience was based on a kludge of sorts. By searching for #edchat, you could follow the sequence of questions, responses, and related comments in real time, separate from other Twitter chatter. The hashtag functions as a sort of portal, focusing Twitter use on the chat content and turning Twitter from an asynchronous to a synchronous tool. Tools other than Twitter, such as Tweetdeck (no longer available), even allowed users to create a multi-column display, with individual columns focused on specific tags and updated automatically. Only the #echat contributions would then be displayed within one column. These tools became popular as an easy way to turn the Twitter feed into a synchronous experience unique to those using the #edchat hashtag.

Research shows that between 2009 and 2014, these Tuesday sessions saw significantly higher engagement than other days, characterized by genuine dialogue and a high volume of questions and replies. Teachers weren’t just “knowledge telling”; they were building communities of practice and exploring new pedagogical ideas in real-time.

The researchers had a special interest in the frequency of questions and replies, and the ratio between the two, assuming these variables would be a good way to assess interaction. In addition, how did these variables differ between Tuesday and other days, assuming this would be related to the higher likelihood of synchronous interaction on Tuesdays? Replies made up a higher proportion on Tuesdays and were significantly higher in the earlier years. My interpretation differs from the researchers’. They argued that there was a decrease in interaction in later years. My interpretation is that the chats drifted away from the formal structure of questions interspersed with participants’ answers. Relying on the massive scale and automated methods employed rather than human raters following the give and take of individual sessions may have led to different interpretations.

2014–2018: The Shift from Dialogue to Broadcasting

At its peak in 2017–2018, #Edchat was a massive digital footprint, averaging 120,000 tweets per month and involving roughly 200,000 different users. However, beneath these impressive numbers, the nature of the interaction was shifting. Starting around 2014, Willet and other researchers observed a transition from authentic conversation toward broadcast-style communication.

Several key trends marked this transformation:

• The Rise of the Link: While early chats focused on natural discussion, later years saw a sharp increase in posts containing hyperlinks to external content, suggesting the platform was becoming a repository for resource sharing rather than deep discussion.

• Retweet Dominance: Retweets began to outnumber original posts, and the percentage of questions receiving replies dropped significantly. Retweets could be used by individuals to bring those who followed these individuals but were not chat participants to experience the content.

• Exploitation: As the hashtag grew in popularity, it became a target for spam and self-promotion. By 2018, the community faced a spike in problematic content and a decline in “authenticity scores” as commercial interests exploited the tag for marketing.

The transition to less interaction and greater influencer dominance may also be related to the active/passive distinction that researchers have begun to study in social media activity. A focus on Twitter chats as a source of resources is consistent with this research topic.

You can still find the use of #edchat on X, Bluesky, and Mastodon instances. The tag is typically used now to indicate educational content and is seldom used in the same way within a chat sequence. A few chats can still be found, often now, using more idiosyncratic tags.

The Paradox of Digital Support

I wrote a series of posts beginning in 2013, focused on edchats, mostly questioning the information value of the process. I appreciated the camaraderie the chats offered, but the limit on the number of characters Twitter allowed, along with my reaction to the content included in such chats, led me to believe the experience was very inefficient, and I thus proposed tactics I thought would increase the professional development value of the experiences.

Edchats were often included as one experience within the grad course on technology class I taught, and I proposed, without success, that students might analyze the content of such chats as a potential thesis project. My suggestion at that time was that video-based systems (e.g., Skype, Zoom) would allow a much more productive approach.

2018–2023: Volatility, X, and Fragmentation

The decline of #Edchat accelerated after 2018, driven by platform volatility and the eventual transition of Twitter into X. Changes in leadership and algorithm priorities disrupted the organic reach of educational hashtags. As the environment became more polarized, many educators began to migrate to other platforms like Instagram, Mastodon, or niche, specialized communities that better served their specific needs.

By 2023, the once-unified #Edchat community had largely fragmented. This decline highlights a critical vulnerability: digital spaces on commercial platforms often lack the stability and continuity of traditional professional development. When profit extraction and algorithmic shifts override user experience, the community suffers.

Lessons for the Future

The history of #Edchat is a reminder that while platforms change, the human need for collaboration remains constant. The legacy of this 15-year experiment suggests that for future digital communities to succeed, they must:

1. Prioritize Active Participation: Moving beyond passive consumption is essential to avoid the stress of social comparison.

2. Foster Authentic Dialogue: Successful communities require mechanisms that encourage genuine interaction over simple content broadcasting.

3. Shift to Knowledge Building: The goal of any digital faculty lounge should be to move from merely “telling” knowledge to collaboratively building it.

Perhaps online interaction among educators isn’t gone; it is simply evolving. As educators move toward new tools, the story of #Edchat serves as both a testament to the power of digital connection and a cautionary tale about the challenges of sustaining authentic community in commercial environments.

I have tried to identify where those educators interested in online interaction with peers went. I could not find the type of quantified data provided by Willett, but other researchers (Greenhow and colleagues) have suggested that Facebook groups and Instagram have become favorite sites for interaction.

Sources:

Greenhow, C., Galvin, S. M., Brandon, D. L., & Askari, E. (2020). A decade of research on K–12 teaching and teacher learning with social media: Insights on the state of the field. Teachers College Record, 122(6), 1-72.

Willet, K. (2019). Revisiting how and why educators use Twitter: Tweet types and purposes in# Edchat. Journal of Research on Technology in Education, 51(3), 273-289.

Willet, K., Carpenter, J., & Na, H. (2025). Ex-Edchat: Historic retrospective of X/Twitter# Edchat. Computers & Education, 241, 1-18.

Evaluating the Consequences of School Choice

The education of learners in the K-12 range is undergoing a noticeable shift as the concept of “school choice” has become a political wedge issue, and state-level legislative policy decisions have given parents and their kids greater control over where and under what conditions students are educated. While often framed from a perspective of the individual family, I have always felt this perspective is too narrow and decisions made for the benefit of one benefit also impact the experiences of others. In various ways, we are all stakeholders, whether it be as citizens living in a society dependent on an educated population, financial contributors as state and local taxpayers, and students and their families involved in school experiences.

While I have followed the issue of school choice for years, this post was prompted by a recent EdWeek article – As School Choice Goes Universal, What New Research Is Showing. Before I comment on the article, here is some background on K12 School Choice.

School choice is the opportunity for parents to enroll their child in an elementary or secondary school other than the assigned school based on their home address. There are multiple variations. The optional school may be another public school in a different district or a public magnet or charter school. Funding for charter schools, which typically operate with a separate board from the public school district, may be public or private. Private schools can be further differentiated as parochial (religious emphasis) or independent.

My personal interest is mainly in the impact of tax-based funding as I see funding as a nearly zero-sum variable – schools compete for a fixed pot of money dependent on student enrollment. Private schools rely on tuition paid by families, contributions, and, increasingly, tax money collected and then distributed to the school in which the student is enrolled. When school choice sends tax money to a private school the approach may be as a voucher or a family-controlled educational savings account (ESA). A voucher might be thought of as the cut of tax money a school receives from the state – per pupil expenditure, but by my understanding typically not the local tax based on property taxes. An ESA provides funds parents can use for several forms of educational assistance – tutoring, textbooks, and private school tuition (source – Overview of Public and Private School Choice Options).

According to the EdWeek source I identified previously, 18 states now allow all students to use state funds to attend private schools, approximately 1.5 million students in the 30 states that allow at least some students access to similar programs. An example of the type of situation that accounts for the difference might be an allowance for students attending what are considered low-performing to use such resources.

My interest has been in the academic achievement of both students who leave their designated public school and a more nuanced issue of what happens to public schools when a sizeable number of students and the funds associated with these students leave. Despite all of the research on such topics, the EdWeek analysis notes that only one of the 18 states identified as providing all students access to state funds made use of the same standardized achievement test in both public and private schools. Obviously, having large numbers of participants from both public and private institutions taking the same tests would offer the cleanest and most powerful comparison of academic impact.

Proposed advantages and disadvantages of each approach to K12 education

At present, the political winds seem to be blowing toward greater parental choice. This is the case despite the lack of consistent findings on whether such choice is of greater benefit to student achievement. Depending on the sample of students used and the method used to quantify achievement, studies generate every possible outcome. The EdWeek attempt at a summary concludes that “Preliminary studies on earlier iterations of these programs have shown “neutral to negative” effects on state test scores, though some programs, like Ohio’s EdChoice, have demonstrated positive outcomes regarding graduation and college-going rates.”

What follows is my attempt to summarize the advantages and disadvantages based on two books by Diane Ravitch that I have read. Dr. Ravitch is a defender of public education as you can probably tell from the titles of her books, but pros and cons are more about the arguments and not the data. Ravitch’s work is heavily focused on research findings, but as I have already indicated, the subissues are so complex that it is very difficult to offer general conclusions. The issue that interests me – what happens to the public schools when students leave (the final con described below under the cons of Private schools) has not, to my knowledge, been a focus as an appropriate methodology would be very difficult to put together.

Pros of Public Schools:

Universal Service and Stability: Public schools are charged with serving all children, providing not just education but also essential social services like nutritious meals, medical care, and mental health counseling.
Accountability and Transparency: Public schools operate under strict state regulations and testing mandates, ensuring a level of transparency regarding student progress and the use of taxpayer funds that can be absent in the private sector.
Professionalism: Public schools generally require higher standards for teacher certification and may provide due process (tenure) to protect academic freedom.

Cons of Public Schools:

Impact of Poverty and Segregation: The biggest “con” of the public system is often beyond its direct control. Concentrated poverty and racial segregation significantly drive lower academic performance, and schools alone cannot solve these structural societal issues.
Curriculum Narrowing: Due to high-stakes testing mandates, public schools may reduce time for the arts, history, and physical education to focus more on tested subjects like math and reading.
Bureaucracy and Funding Disparities: Public schools are often burdened by intrusive regulations and suffer from persistent underfunding, particularly in districts with low property tax bases.

Pros of Private Schools:

Autonomy and Flexibility: Private schools enjoy the freedom to design their own curricula and select their own testing methods, allowing them to cater to specific educational philosophies or religious values.
Personalization and Choice: Families can select schools that align with their specific needs or interests, whether through small religious schools or specialized academies. This “market-driven” approach appeals to values of freedom and optimism.

Cons of Private Schools:

Lack of Oversight: some private schools receiving public funds are not accredited and are exempt from state accountability systems. This makes it difficult for the public to evaluate if tax dollars are producing academic results.
Selective Enrollment: Unlike public schools, private institutions can be selective. Critics argue this leads to greater segregation and less equity, as schools may shun students with the highest needs or those who are the “toughest challenge” to teach.
Draining Public Resources: Every dollar diverted to a private school voucher is a dollar removed from the public school system, which still incurs fixed costs such as building maintenance and teacher salaries. This can lead to increased class sizes and program cuts in the remaining public schools.

Summary

Parents and their kids are being allowed greater control over where they attend school. School choice comes with multiple pros and cons, and these issues have been difficult to evaluate because public and private schools in most states are not required to use the same achievement tests. My personal interest is in what happens to the public schools that lose students to private schools when choice is allowed.

Sources for Pro and Con Section

Ravitch, D. (2014). Reign of error: The hoax of the privatization movement and the danger to America’s public schools. Vintage.

Ravitch, D. (2020). Slaying Goliath: The passionate resistance to privatization and the fight to save America’s public schools. Vintage.