It has been a challenging week for those of us promoting the potential benefits of AI. Despite the reality that resigning from Anthropic may have cost him a fortune when his company goes public, an employee did quit, claiming he could not, in good conscience, continue to work at an AI company when there is a chance more powerful models may end up killing everyone (10% chance was his estimate). Other employees agreed; congressional leaders panicked, but Trump said to forge ahead because we must compete with AI development in other countries. All the tech pundits immediately focused on the topic, and most news outlets frantically tried to find someone willing to come on their programs who sounded like they knew something about this concern. In this context, I am the guy trying to write something about the value of AI in K12 and secondary classrooms.
First, while not the focus of this post, let me say that the AI models I support have been around for some time and aren’t the newest models that have created the current concern. I have no idea if we are locked in a “Manhattan-like” race to achieve the level of AI intelligence that would make our adversaries cower in fear or at least assume we have achieved a level of mutual assured destruction. I have opinions, but I try to stick to things I can argue with at least some degree of insight and credibility. I want to make a pro-AI argument based on a couple of topics I have addressed before — AI tutors and AI Writing Feedback.
You don’t have to convince me humans can do a better job.
Tutoring — I must have a dozen or so posts focused on tutoring. I became interested in tutoring perhaps 40 years ago because I was interested in mastery learning, and I read Benjamin Bloom’s two-sigma challenge argument. Bloom proposed that researchers should develop approaches to school learning that matched the level of achievement learners could reach with consistent access to a tutor. He proposed mastery learning because it was a strategy that provided some of the benefits of tutoring — e.g., meeting the student where they are, addressing deficiencies rather than moving ahead whether or not a student was ready, and multiple opportunities to demonstrate their competence and understanding. While I was interested decades before AI tools were available, it seemed to me that technology could offer a practical way to make some of these component strategies feasible.
AI rekindled my interest. AI has some obvious advantages to the benefits I saw earlier. Most simply, it could be responsive at the level of the individual. AI does not have to rely on preestablished smaller units of instruction and the benefits of evaluating and shaping an optimal rate of progress through these units. AI can respond to a learner’s requests for help and evaluate performance tasks to provide feedback and make independent decisions about what should happen next.
Responding to student writing — So much of the commentary surrounding AI and learning to write or writing to learn has focused on cheating. How can educators give a grade when they cannot be certain that a student has actually written the work that has been turned in? How can students learn to write when experiences designed to exercise writing subtasks can easily be completed by AI rather than the student? Certainly, those of us who assign and evaluate writing tasks must grapple with these possibilities.
My interest has been more in making use of AI to provide feedback on writing tasks. We have very useful ideas about developing writing skills that focus on writing a lot, receiving immediate feedback on initial attempts and rewriting the initial drafts based on this feedback, and even writing conferences in which a student meets face to face with a teacher to discuss what they have written and their thoughts on their strategies and challenges. Research shows the impact of these tactics.
The Reality
You do not have to convince me that a skilled human being does a better job at vital human tasks. You have to convince me you are actually willing to pay for one. We live in societies that demand universal, high-touch services while systematically starving public education budgets. Whenever we are faced with the staggering bill of hiring, training, and retaining competent professionals at scale, we blink. Both tutoring and writing instruction make great examples of drastic understaffing.
It would be great if every student, but especially students who struggle, had access to a well-trained tutor. Wealthy parents can afford such assistance. Schools receive some government assistance. However, consider the summary of a recent study from USC — Two-percent of U.S. children receive high quality tutoring. The definition of “high quality” was based on the following factors:
- Is delivered at school
- Is delivered during school hours
- Groups are composed of three or fewer students to one tutor (four or fewer students can be effective with older students)
- Is provided to a student three or more times per week
- Lasts at least 30 minutes per session.
- Is delivered by teachers or well-trained professional tutors rather than volunteers, peers, or parents.
These are high expectations, but even if you relax them, you get the point. It is simply impractical to provide enough human tutors to meet classroom needs simply because of the cost and availability of suitable professionals. School budgets are already constantly being challenged at existing levels of taxation with larger and larger class sizes and other cost containment strategies becoming the standard.
Developing writing skills faces a similar reality. I know some folks who teach the freshman-level English Composition classes and always felt they had a really difficult job. Most are hired as adjuncts and work 4 classes if they can get what is considered a full load. Class sizes are typically 20 students, making a full load the responsibility for 80 students. My English Comp classes were 50+ years ago, and that may have biased my opinion. I thought the classes were pretty much worthless, with a focus more on literature than writing. The one writing class I had that was beneficial was called Business Writing (I think this was the title) and required more writing and more thorough evaluations. I probably wrote the most as an undergraduate in the Psychology Research Methods course which had a sizeable component focused on APA style and writing research reports. I encountered the “no-no” passive voice for the first time and struggled to express myself in a form that just did not sound right to me.
One of the books I have read on the impact of college — Academically Adrift — is quite critical of the college experience, blaming both institutions and the commitment of learners. The book draws heavily on a longitudinal study of the college experience, questioning what students learn in early-year college courses. Written communication is one area that receives attention.
I tried without much luck to access studies documenting how much writing a typical Freshman writing course requires. You would think such data would exist. AI requests returned data that were based on multiple syllabi the AI appeared to access to respond to my request. The conclusion from this approach was that students turn in 15–20 pages of what was described as formal writing and maybe 25–30 pages of informal writing. I was unfamiliar with this distinction so I requested an explanation. The description of informal writing follows.
Informal writing often means lower-stakes composing that supports learning and revision, such as:
* Freewrites and brainstorming.
* Reading annotations and response journals.
* Discussion posts.
* Peer-review comments.
* Outlines and research notes.
* Draft reflections or process memos.
* Exploratory writing before a formal essay.
The 20 pages of writing I mentioned above mentioned in the Academically Adrift report as the point at which measurable improvement was typical. However, 50% of students did not reach this level of activity and showed lower levels of improvement and sometimes very little at all.
I don’t see a practical way for educators working with 80 or so students to increase the amount of writing that can be carefully reviewed and critiqued, discussed with students from time to time, and rewritten and reevaluated. I don’t see any way to provide feedback within a couple of days so students can rework the original submission on a reasonable schedule. I doubt institutions will be able to lower class sizes by higher additional experienced writing instructors.
Writing instruction in K12 was difficult for me to describe based on sources that quantify the amount students do. I was surprised the amount in dedicated higher ed writing courses was so low and I know without the equivalent kind of data that similar concerns are made about writing among younger learners. Distrust of out of class writing will likely only add too this issue.
I assume these data provide a reasonable description of what we can do with the current Human Resources commitment.
Summary
This situation reminds me of the expression “don’t let perfect be the enemy of good”. My proposal and the topic of plenty of earlier posts is that AI tutoring and writing assistance seems a reasonable way to augment the work of beleaguered educators.
You must be logged in to post a comment.