Many states now require high school students to learn coding before they can graduate, and a host of organizations encourage students to build coding skills. Is this all a waste of time and energy now that chatbots can code? In a word, no.
A few weeks ago, I was chatting with a friend from high school as he pondered a career change, and when I floated the idea of pivoting to data science, he responded with a question I’ve been hearing more and more the last few months: “How is this not the kind of thing that is first on the chopping block to get replaced by ChatGPT?” In the wake of new AI chatbot releases by Google and others, Farhad Manjoo, the New York Times’s tech reporter, cited titans of tech declaring “the end of programming” as “AI programmers are quickly getting smart enough to rival human coders.”
If this were true, it would have big implications for K–12 education, as many states now require high school students to learn coding before they can graduate, and a host of organizations encourage students to build coding skills. Is this all a waste of time and energy now that chatbots can code?
To understand the implications of AI for the future of coding, I talked with coders across several fields, none of whom believed that AI would decrease the value of these skills. The notion that coding will soon be obsolete seems to be an overreaction to the legitimately incredible (and sometimes alarming) capabilities of these new AI programs, and coding jobs appear to be much less immediately threatened than many others. Instead, all of the coders I talked to believed that these skills would only become more valuable in the future. As Rochester, New York–based software developer Justin Reeves put it, “the increasing application of AI tooling is more likely to be a boon for CS [computer science] demand.”
The last few months of exposure to the powers of generative AI should give us all pause in making predictions about what human skills will (and won’t) command high pay in the future. Be it poetry, images, recipes, coding, or frighteningly authentic counterfeit songs, 2023 has been the year that generative AI began astonishing everyone with its capabilities. Because these AI abilities will only grow, with effects that may lead to considerable social and economic change, it may be impossible to accurately forecast what types of skills young people should invest in.
Despite such uncertainties, we cannot avoid making some forecasts, and my conversations with professional coders surfaced several reasons to doubt that the increasing sophistication of AI implies that students should stop learning to code.
First off, it should go without saying that just because computers are better than humans at something does not mean students should not learn it. Math education is an obvious example. For decades, computers—and even many pocket calculators—have been better than humans at every single skill students develop in the K–12 math curriculum, yet no one seriously suggests we stop teaching math.
There is also nothing new about having alternatives to coding. Chatbots that can write code are a welcome new technology, but as long as personal computers have existed, there have been parallel ways of interacting with computers, some code-based and others that are more user-friendly. Coding is, in fact, nothing more than one method of giving a computer a customizable and precise set of instructions. Coders develop their skills because it is sometimes quite useful to interact with computers with the clarity, functionality, specificity, and replicability of code.
Consider Microsoft Word. You might not realize it, but that program—like a great many of the software applications you use every day—can be used to “code.” When experienced users want Word to do something there is no pre-set button for, such as automating a repetitive task, they can create little programs called “macros” that are written using the Visual Basic for Applications (VBA) language. Certainly AI might make that kind of coding obsolete, but then again there have long been other ways to create macros within Word (or other Office applications) without coding anyway. Even before the new wave of generative AI, Udemy was offering a data science course “without coding.”
So if there are already many ways to do so much programming-type work without actually coding, why does anyone still do it? There are many reasons, and it is hard to see how AI will alter their basic logic. For example, microeconomists like to use scripts (i.e., documents of code), rather than just pointing and clicking, because it makes their work more reviewable and replicable. It also makes their work more precise. Presumably, an analyst could instruct a sophisticated AI program to perform a series of various tasks, but often there are multiple ways to do something, and without an understanding of the underlying code, the analyst would not be able to figure out what happened, or whether it could be improved. As Seattle-based data scientist Devesh Tiwari told me, “It is great to have powerful tools that generate useful output, but we feel much better about them if we know what they are doing inside the black-box,” which coding helps with.
More broadly, being able to code enables humans to interact with computers in their own highly analytic languages. As computers grow in importance due to AI, humans will presumably need to interact with them even more, and sometimes that will include using code. Tyler Moore, the Tandy Professor of Cyber Security at the University of Tulsa, explained to me how coding skills give students “a better grasp of what computer programs can and cannot do,” which is increasingly important now “that so much of our world takes place digitally.”
Another reason to doubt that students should quit learning to code is that, whatever uncertainties exist around the future of coding, it is hard to see why those skills would be more affected by AI than anything else. With chatbots that can write humorous poetry, conduct sophisticated mathematical calculations from a written input (with plugins), and write a recipe from a picture of the inside of your fridge, it is difficult to see what kinds of white-collar “information” jobs will not be affected. Just because Chat GPT-4 can write a sonnet does not mean we are going to stop having students practice creative writing.
For the moment, coding skills have not lost any of their importance. Indeed, AI is driving growth in new fields such as machine learning engineering. As for more traditional coding skills, the current AI chatbots are not very good yet. When I have solicited coding help from Chat GPT-4 (at present, the most updated version of OpenAI’s chatbot), it has often been useful, but it has almost never been error-free. Without significant coding skills, someone hoping to make a webpage or scrape data from the web using an AI chatbot to write their code will no doubt give up from frustration in short order. If you cannot see inside the black box, you will not be able to guide or correct the chatbots’ work. Some of that may shift over the next few years as the technology improves, but whether the development of AI will make coding more or less valuable is far from clear. For example, it is easy to imagine that, over time, AI will make coding both easier to learn and even more valuable. Moore argues that “The jobs of the future will rely on AI, and coding will continue to be necessary to develop AI-based solutions to problems.”
Tiwari, the data scientist, underscored the importance of preserving K–12 coding standards by pointing out that “college students who excel in computational and quantitative fields encountered those concepts well before entering college.” That means that rather than scale back coding requirements, reformers should be thinking about how to fine-tune them, perhaps in tandem with math education, which itself is due for some reform. Teaching coding and math together—which is how actual mathematicians do their jobs—would be one reform that could improve student outcomes while streamlining some requirements. The non-profit curriculum provider Bootstrap has a creative approach that many schools could already choose to adopt: Middle and high school students take integrated courses that merge traditional math concepts with coding and data science skills.
Just as the emergence of computers didn’t end math education, the emergence of AI that can code doesn’t mean those skills are obsolete. At least for students hoping to work in STEM fields, learning to code will probably need to be part of the curriculum for many years, and it is possible that coding skills will become even more important as AI becomes a larger presence in our lives.
Can we stop with the learning stations already? My teacher prep endorsed them. My first instructional coach trained me in them. Every school that I’ve ever worked at has incorporated them. Look them up on Teachers Pay Teachers and you’ll find scores of activities for various literacy stations, each one promising that they are proven effective.
Unfortunately, this idea—that a teacher should only teach for a few minutes before setting kids loose to transition through a maze of stations full of glitter, glue, and razzmatazz—is a glossy, inefficient, ineffective use of class time.
Stroll through a classroom that uses learning stations and you’ll see students engaged in all sorts of seemingly compelling activities and projects. Look more closely, and one sees a hodgepodge of the promising and the pointless, tasks that demand thinking beyond realistic expectations and tasks that require no mental effort at all: some extended writing or structured practice and lots of needless coloring or word searches.
Consider just one concrete example. In a video on the influential site Edutopia, which specializes in instructional advice, one station-based classroom exhibits high student engagement and seamless routines. But a critical consideration of what’s actually happening at these centers raises concerns. What’s the point of a twenty-minute station where students spend their time cutting, pasting, and writing in rainbow letters? And that doesn’t even touch on the copious references to kinesthetic, visual, auditory, and other ways of learning, allusions to the myth of “learning styles,” which has as much basis in reality as homeopathic medicine.
The body of research on learning stations is bordering on inexistent, but there are plenty of correlational studies to warrant hearty skepticism, if not outright rejection, of this popular classroom practice.
In one study, researchers split students into groups that received either 25 percent, 50 percent, or 75 percent of class time on teacher-directed activities versus seatwork, and then all students took the same test. The conclusion was simple: Students who had more teacher direction learned more. Similarly, Barak Rosenshine found that the most effective math teachers spent the majority of class time in direct instruction, modeling, and questioning.
Conversely, classrooms built on literacy stations, like the one in the Edutopia video, tend to feature only a short teacher-led mini lesson, followed by an hour of individual student work time—an allocation of time in the aforementioned studies that achieved the worst academic gains.
There is an influential body of research from the 1980s, generally referred to as “process-product research,” that points to a similar conclusion. In it, researchers observed teacher behaviors and correlated their actions with student outcomes. Across the board, teachers and classrooms that were highly structured and spent more time on direct instruction, whole class practice, and discussion generally achieved more. Humorously, in one cluster of studies, the researcher set out to prove that student talk was the best means of learning, only to conclude the exact opposite: “teacher talk correlated positively with both achievement and attitude.”
As one more piece of evidence, another study found that high achievers can learn comparatively well in self-directed activities, but low achievers need direct guidance the most. Understood so, student-directed learning may work for some, but it is a pedagogy of the privileged.
None of this evidence is a slam dunk against learning stations. Rarely if ever does evidence in education function so. Almost all classroom practices work to some extent, somewhere, sometimes, for some students. Any academic activity is better than none. The doggy paddle will get you across a lake, but surely a breast stroke would be better. Everything can work, but what works best?
Without doubt, many of the activities at the stations themselves have academic value—reading passages, phonics work, small group discussions, extended reflections. But if each of these activities are, in fact, worthwhile, the evidence recommends it would be better for the teacher to facilitate each one as a class. Rather than split a room of boisterous students into three or four separate learning groups, all working on separate activities, a teacher should instead run each activity as a class. Read and respond to a passage together, then transition the class to phonics work, and perhaps end the hour with a fun activity for a reward.
Neither is this evidence a criticism of all individualized instruction outright. In many ways, stations are an attempt to manage a seemingly unmanageable spread of academic abilities in one room. What will Johnny do while Timmy’s group gets an extension lesson from the teacher? Such individualized challenge or support can occur in the format provided above, though. In my classroom, every student read the same complex, challenging text and then responded to reflections of varying difficulty while I floated around challenging or supporting accordingly.
Inadvertently, one student in the Edutopia video provides the individual testimony that recommends this approach. She tells us that her favorite station is the one where she interacts with her teacher in a small group. During whole class instruction, every student has the teacher’s attention and every student receives instruction. Conversely, during stations, only a few get attention from the teacher, and the rest are on their own. I have said before and will say again that all these appeals to individualized or differentiated instruction often result in collective neglect.
Undergirding this approach to the classroom is an almost religious commitment to the idea that students learn best on their own, which simply is not the case. Like a heresy that springs anew, new models that are built on this false faith spring up over and again—project learning, discovery learning, and now learning stations. Every time, the rebuttal is the same: Teachers, not students or activities, are simply the best conduit of learning in a classroom. That should a self-evident conclusion, but alas, it is for some reason controversial to say so.
High-achieving students who are also disadvantaged by class, race, or ethnicity may be the most overlooked group in American education. Education reporters, ask yourself: The last time you visited a high-poverty elementary school, did you query the principal or teachers about how they serve children who are a year or more above grade level? Who arrive on the first day having already mastered that year’s material?
If the answer is no, you’re hardly alone. Stories about advanced education are too few and far between. One indicator: A simple Google News search for “gifted education” turns up 1/1,000 as many articles as a search for “special education.”
Yet advanced learners shouldn’t be ignored. Nobody should go to class and not learn anything new—not kids three grade levels behind, nor three grade levels above. Like all children, they deserve to attend schools that notice them and meet their needs.
Kids from disadvantage who are nonetheless ready for advanced learning opportunities have so much to contribute to the country. Like other high achievers, they have great potential to be tomorrow’s breakthrough scientists, to lead organizations, and to solve the world’s problems.
These are also the children with the best shot at diversifying our college campuses—especially in the wake of June’s affirmative action decision—as well as our graduate schools, professions, and elite institutions.
We know these kids won’t “be fine” just because they are smart. Reams of research shows that low-income gifted kids are much more likely than their more affluent peers to see their achievement decline as they move through school. Just as kids who have fallen far behind in the Covid era need intensive support to catch up, so do high flyers need intensive support to keep moving ahead.
That’s the message from the National Working Group on Advanced Education, a racially and ideologically diverse group of scholars, practitioners, and policymakers that I convened over the past year. Its recent report, Building a Wider, More Diverse Pipeline of Advanced Learners, offers thirty-six concrete, evidence-based recommendations for districts, charter schools, and states.
Several of them carry implications for reporters and analysts covering this important student population.
For example, journalists might ask their local districts:
To their credit, several education reporters are already doing excellent work covering advanced learners.
This has become a major focus for Sara Randazzo at the Wall Street Journal, for example, whose article went viral on the absurdity of eliminating honors courses “because equity.”
The 74’s Jo Napolitano offered a great example of “solutions journalism” in her write-up of Dallas’s success in boosting the diversity of students taking advanced coursework.
Kudos also to Elizabeth Heubeck at Education Week for her profile of a Black gifted-education coordinator in Virginia who is dramatically boosting the number of Black and Hispanic students in advanced education.
In all three cases, the journalists rejected the frame that gifted education and the like is inevitably inequitable. More of that please! These stories are out there, waiting to be told. And telling more of them is apt to make good things happen for kids.
Editor’s note: A version of this article was first published by The Grade.
School systems have long been interested in supporting students’ mental health as a means to improve behavior, decrease absenteeism, and even strengthen academic performance. The privations and disruptions of pandemic-era school closures have only heightened this interest. Mindfulness is one technique among many being used as a mental health and resilience booster in schools, but does it really do any good on that front—much less on key outcomes like student academics and behavior? A recent meta-analysis published in the Journal of School Psychology and conducted by a team of Canadian scholars aims to find out.
Mindfulness originates from ancient Buddhist practices and focuses on teaching individuals to pay calm attention and focus with nonjudgmental acceptance on their thoughts, feelings, and sensations. The secularized version of mindfulness is often used as a therapeutic technique to decrease anxiety and stress, but research on it is heavily focused on adults in non-educational settings. In the K–12 setting, kids might be guided through a slow breathing and visualization activity to calm them before class begins. Practitioners tend to agree that students need help with stressful situations and relationships, but worry about how to make the time for mindfulness in the classroom and whether they are equipped to implement such programming.
The Canadian analysts here initially identified nearly 5,400 unique, peer-reviewed, English-language studies on mindfulness-based programs (MBPs) published through 2017 for their meta-analysis. But limiting their criteria to randomized controlled designs focused on students—in either K–12 or undergraduate college settings—reduced that number to just forty-six. They standardize and code a set of educational outcomes—which includes attention, self-regulation, impulsivity, interpersonal skills, school functioning, student behavior, academic performance, and mindfulness itself—and also create an overall school adjustment measure that combines those individual outcomes. They compute standardized effect sizes for each outcome and subgroup, and use a random-effects model that enables inferences to a broader population of comparable studies. They also make adjustments to their models to control for publication bias.
At the end of the programs, the effect size of MBPs compared to control groups was small for the overall school adjustment measure, as well as the individual academic performance and impulsivity measures. The effect was small to moderate for attention, and moderate for mindfulness. Specifically, of the forty-six studies, only nineteen assessed mindfulness as an outcome itself, and they had an effect size of 0.5 (considered moderate impact). There were no differences between treatment and control groups in terms of interpersonal skills, school functioning, or student behavior.
When examining whether students’ educational level had an impact on the effectiveness of MBPs, they found significant differences: increases in mindfulness (only) for elementary grade students; increases in both mindfulness and overall school adjustment for students in middle and high school; and increases in overall school adjustment (only) for undergrads.
Variations in the control groups influenced the results, too. For example, when an MBP treatment group was compared to a business-as-usual control group (i.e., no program per se to improve school adjustment), they saw significant differences between them. But when MBPs were compared to another active control condition (i.e., students received a different, non-MBP treatment to improve school adjustment), they saw no difference. The analysts suggest that other active control conditions (for instance, playing “The Good Behavior Game” or participating in concentration training) might be providing similar impacts on student focus or classroom disruption, without the specific emphasis on mindfulness. What’s more, positive impacts only materialized when the MBP program was led by an outside facilitator with previous experience in mindfulness. No differences were observed when the classroom teacher was leading the program, whether experienced or not—nor when the outside facilitator had no experience in mindfulness.
In the end, it seems like there’s no great harm in helping kids—especially older ones—to be more mindful of their thoughts and actions, presumably to reduce stress. Plus, the programs provide a little lift to academics and help students to pay attention. Encouragingly, their effectiveness is less about a particular mindfulness program—that’s good since they are all over the map—than it is about having an experienced facilitator who knows what she’s doing. Still, we’ve got to keep in mind (pun intended) that MBPs don’t have outsized impacts, probably don’t belong on a classroom teachers’ to-do lists, and will take time away from other priorities if they are not integrated into the school day.
SOURCE: Jessica Mettler et al., “Mindfulness-based programs and school adjustment: A systematic review and meta-analysis,” Journal of School Psychology (April 2023).
Accurate property assessments are a basic requirement for many school funding systems to function properly. Unfortunately, data suggest that a variety of problems are introducing inaccuracies, exacerbating inequalities, and causing negative chain reactions in taxing jurisdictions all around the land. A trio of economics researchers conducted a deep dive into one state’s efforts to reform its property assessment infrastructure, and their report was recently published by the Board of Governors of the Federal Reserve System.
In 1989, a series of newspaper articles in the Lexington Herald-Leader exposed the existence of rampant problems in every corner of Kentucky’s K–12 education system, including numerous instances of tax evasion, inequitable treatment of property owners, and ineffectual safeguards around tax assessment protocols. Journalists and state investigators identified the most pressing problem as the underassessment of properties—specifically, high-end homes—brought on by a mix of mistakes and corruption. In response, lawmakers passed the Kentucky Education Reform Act (KERA) in 1990. Under that law, the state’s Department of Revenue was given expanded power over locally-elected property assessors, and a technical assistance intervention was imposed on the worst offenders in an effort to improve their practice. Additionally, KERA introduced a means by which to double check the accuracy of assessments.
To evaluate Kentucky’s efforts, researchers from the Federal Reserve examined administrative data from all 120 counties in the state. The state had deemed ninety-three counties to need intervention: twenty-five were emergency reassessment (ER) counties, and sixty-eight were technical assistance (TA) counties. Twenty-seven additional counties weren’t given any direct intervention under KERA and serve as a control for certain analyses. Data for pre-treatment investigation cover the years between 1982 and 1989, and post-treatment data cover 1990 through 1998. The researchers use a difference-in-differences approach to conduct their analyses.
Overall, they found that the treatment (training, oversight, double-checking) substantially increased aggregate assessed real property values. By the end of the active intervention in 1994, per-pupil property value in ER counties had increased by 32 percentage points over untreated counties compared to 1989. For TA counties, the increase was a more-modest 11 percent, as expected due to the lower levels of intervention needed. By 1996, districts in ER counties were receiving approximately $100 additional per pupil from local property tax revenue. (The researchers looked at commercial and farm property, as well as residential, but the report focuses mainly on residential, as does this review.)
They also found that interventions had large positive effects on assessment inequity. Using a new protocol called coefficient of dispersion (COD), state overseers measured the average difference between the sales ratios and the median. When properties are equitably assessed, assessment-to-sales ratios within a jurisdiction should be similar across properties, regardless of size or sales price. A low COD indicates an equitable assessment across properties (e.g., an accurate and properly-functioning system), while a high COD indicates a flawed assessment (though it is not enough to isolate a cause). The intervention reduced the COD in ER counties by an average of 15 points, indicating that the system had become more accurate. A similar pattern of COD reduction occurred in TA counties, though it was less pronounced.
The report describes in detail how these results indicate an increase in assessment accuracy and equity, rather than, say, an increase in home prices within treatment counties over the same period. Data indicate that internal factors (such as understaffed assessors’ offices and corruption) likely drove the underassessment problem, rather than external factors such as population density changes and local economic conditions. Hence, they deem the cure highly effective in treating the disease: The more inequity that existed before the intervention, the greater corrective the intervention applied.
Finally, the researchers use their estimates of the treatment effects along with the state aid formula to simulate the impacts of the intervention on state aid to the treated districts. Predictably, increasing local revenue to schools decreases the need for state dollars to level up district funding, although changes to the overall funding formula, which were also part of KERA, limit that impact slightly. Still, millions in state funding that should rightly have gone to districts playing by the property assessment rules ended up in ER and TA districts in each pre-treatment year, a situation significantly remediated post-treatment.
What does all this mean? The Fed’s research team suggests that COD double checking could be applied to any property tax assessment system to ensure its accuracy. This recommendation applies not only to schools, but to public transit, police and fire service, and libraries. Any system funded by property taxes, and run by public servants who may toil in understaffed offices or who could succumb to corruptive elements, should take advantage of this simple—and proven—double-check protocol.
SOURCE: Alex Combs, John Foster, and Erin Troland, “The Role of Property Assessment Oversight in School Finance Inequality,” Board of Governors of the Federal Reserve System (July 2023).
Cheers
Jeers
