It's Way Harder Than You Realize (Part 2)
On Building Sandcastles So That We Can Break Them Down
If there is one thing I say to colleagues about knowing how to best integrate AI into their lives, it’s that they simply have to play in the space. No matter the technology, engaging in iterative, playful interactions offers the best path forward understanding both its power and its drawbacks. This notion of open play is something I think many adults struggle with. I’m awful at this even when playing with my own kids. When they ask me to play, I frequently think about how I’ve forgotten how to play. Staring at a pile of Transformers or a bucket of Legos somehow freezes me up. Whereas for my kids, these are untapped worlds waiting to be made whole—brought into existence through equal parts imagination and effort. Meanwhile, if we’re playing a board game or a sport, I know what to do. There are rules and expectations in each of these scenarios. CandyLand has a structure.1 Knee hockey is bound by certain parameters and expectations.2 But playing in the generative AI space feels like immersing oneself in a blank canvas of uncertainty; at least initially. So I empathize with those who can’t simply bring themselves to play in the ways we might hope.
Which brings me to the essence of this post. Building on Part 1 from earlier in the week I offer a way of playing with ChatGPT, below. As someone brought into the world prior to the internet, but now someone who lives a life that is entirely too far online, lots of my thinking about things has been shaped by really interesting thinkers/users of the internet. One of my favorites is Jon Bois (who is nominally with SB Nation, but does dozens of other cool internet things) because of his unique perspective and willingness to try weird things.3 It’s an instructive stance that informs lots of my own thinking and approach to integrating generative AI in my work. Commenting on his piece 17776 (linked above), Bois wrote:
There are countless different ways to write, and things and ideas to write about. And the Internet offers a kaleidoscope of different formats, media, tools, sights, and sounds to tell your stories. And most of us are not even trying to scrape the surface of any of it. We’ve got to start thinking of the Internet as something more than a glow-in-the-dark newspaper.
Similarly, we’ve got to start thinking of Generative AI as something more than a glorified Google. Like Bois’ work, the blog Fire Joe Morgan was something I regularly consumed online in the early aughts. Born out of desire to critique old-school takes from sports journalists, it launched the careers of a number of prominent TV writers and producers that are household names today. Much of their work centered on a kind of back and forth with written (or spoken) pieces from journalists and broadcasters. It was snarky, witty, acerbic, scathing, and routinely hilarious. So, in the interest of play, and with the help of ChatGPT-4, I crafted a ‘FJM’ inspired back and forth with an AI evangelist writing a NYT op-ed style piece in the tone of David Brooks and Maureen Dowd. I then had GPT do the ‘FJM’ takedown from my perspective. Lots of the op-ed drew inspiration from Shulman’s Seven Areas of Teacher expertise. The piece below is the result of lots of thinking, prompting, and dialogue. Perhaps the more instructive part is linked here, as this allows you to see the ‘rules of the game’ and to make the canvas a little less blank.4 The AI crafted op-ed appears in standard text, “my” critiques are bolded and italicized—I did add some depth and color in a few areas, but it’s 95% AI generated.5
The Future of Education: Embracing AI's Promise
In the hallowed halls of academia, where chalk dust once settled on wooden desks and teachers poured over textbooks, a revolution is quietly unfolding. The rise of artificial intelligence is not merely a technological advancement; it is a paradigm shift that promises to redefine the very essence of education.
Ah, the hallowed halls of clichés. Chalk dust and wooden desks? Are we also using quill pens and inkwells? Let's get to the 21st century, please
The traditional model of education, rooted in Lee Shulman's seven areas of teacher expertise, has served us well. But as we stand on the precipice of a new era, we must ask ourselves: Are these principles still relevant in the age of AI?
Relavant? You’ll be irrelevant in a few months, GPT4.
Consider Content Knowledge, the understanding of subject matter that teachers must possess. In an age where information is readily available at our fingertips, the need to memorize facts and figures becomes redundant. The teacher's role shifts from a gatekeeper of knowledge to a guide, helping students navigate the vast ocean of information.
Redundant? Like relying on AI to tell us the Earth is flat? Critical thinking called; it wants its relevance back."
Pedagogical Content Knowledge, the ability to teach subject matter, is similarly transformed. AI can tailor learning experiences to individual students, adapting to their unique needs and learning styles. The teacher's expertise in how to teach becomes less critical as AI takes on the role of personalizing education.
Tailor learning experiences? Like a one-size-fits-all suit made by a blind tailor. Unique needs met by an algorithm? Bud, you just dropped learning styles as though they were a thing."
The broader context of education is undergoing a profound transformation. Knowledge of Educational Contexts, once confined to the classroom and school community, now encompasses a global network of learners. AI connects students across continents, fostering collaboration and cultural exchange that transcends traditional boundaries.
Global network of learners? Sounds like a marketing pitch for a tech startup. Let's not pretend AI is the United Nations of education.
And what of teachers? The notion that they are irreplaceable is becoming increasingly outdated.
Outdated? Like vinyl records and film cameras? Oh wait, those are cool again. Maybe human connection and expertise will make a comeback too.6
With AI's ability to personalize learning, adapt to individual needs, and provide consistent and unbiased education, the role of the teacher is shifting.
While the human touch will always have its place, the reality is that AI offers efficiencies and capabilities that human educators cannot match. The replacement of teachers by AI is not a loss; it's an evolution, a step forward in making education more accessible, equitable, and effective.
Evolution? More like a devolution. Let's replace human insight with code and call it progress. What's next, replacing parents with smart speakers?
Critics may point to AI's limitations, such as hallucinations and misinformation, as barriers to its integration.
May point? They ARE pointing, and with good reason. AI's limitations aren't speed bumps; they're roadblocks. Let's not drive off a cliff here.7
Yet these challenges are not insurmountable. With careful design, ethical considerations, and human oversight, AI can overcome these obstacles and become a reliable partner in education.
The future of education is here, and it's AI-driven. It's time to embrace the promise of AI, not with blind optimism but with pragmatic enthusiasm.
Pragmatic enthusiasm? How about pragmatic skepticism? AI-driven education sounds great until Skynet takes over the school board.8
The replacement of teachers is not a threat but an opportunity, a chance to innovate and expand the horizons of education. The landscape of education is changing, and we must adapt, innovate, and experiment to ensure that we seize the opportunities that lie ahead.
Or maybe we should think, reflect, and question before we leap into the AI abyss. Opportunities? Sure. But let's not forget the risks, costs, and very real human factors.9
The value of this sorts of exercise is in many ways rooted in its novelty. I had a desired outcome in mind and I was able to get ChatGPT to that space. But I also needed prior knowledge—of things like Shulman’s Seven Areas of Teacher Expertise, of the Fire Joe Morgan sports blog, of the writing tones of David Brooks and Maureen Dowd, and of the ways we might weave generative AI into this finished product. That’s an eclectic mix of things that all came together. But it’s less about the product and more about the process.10
So how do I think this plays out in relation to the Seven Areas of Teacher Expertise? Some quick, not even half-baked reflections under the framework of EASIER-HARDER-PUSH
Content Knowledge: PUSH. Given hallucinations, I am not going to ChatGPT (first) to start to learn something. With that said, the responsive nature of the technology does allow one to learn content differently. Is this better? Not sure, yet. but there is potential for positive and negative in here.
Pedagogical Knowledge: EASIER. Having an understanding of pedagogy certainly helps and allows one to get better outputs. But if I am a new teacher and I have little pedagogical knowledge, GPT can help me deploy some foundational techniques in useful ways.
Curriculum Knowledge: HARDER. Somewhat dependent on the educational setting. Given what I expect will be huge implications for education, it’s almost as though every discipline needs to be informed at least some by AI. And that feels like it makes things harder in terms of knowing curriculum.
Pedagogical Content Knowledge: HARDER. This is my biggest area of concern. I know how to teach history really well. That was before ChatGPT. I don’t know if the way I’ve taught history in the past will be as effective as it once was. And I don’t think anyone does, in any subject.
Knowledge of Learners: HARDER. Students growing up in this era will engage with educational systems much different than my own. We’ll all need to know our learners better, and differently, than ever before as these technologies become more pervasive.
Knowledge of Educational Context: HARDER. There are no “best practices” here, only emergent ones. In my opinion, we’re in the midst of a systemic disruption that we’ll only be able to fully understand after it’s much further along. People who live through major historical events often don’t know they’re doing so in the moment.
Knowledge of Educational Ends: HARDER. Again, given what I are think are likely system-wide disruptions, it’s hard to know what educational ends will even look like. With that said, I do feel strongly that those in possession of knowledge will know how to manipulate and leverage these tools to their advantage far greater than those lacking in knowledge—so let’s lean into that for as long as we can.
Dylan William, John Hattie, and Arran Hamilton recently released a working paper The Future of AI in Education: 13 Things We Can Do to Minimize the Damage that is really quite profound in its scope and willingness to grapple with the future. That these voices are sounding an alarm like this should cut through any perceived noise about AI in education right now. It’s a signal—and a very clear one—for those of us in education. In Part Three: Recommendations, they write:
…you may be deeply skeptical or even perturbed by the scenarios we have discussed. You may feel that the mass deskilling and de-eduction of humanity seems like something that could never happen. You may say, ‘schooling has been through these existential crises before and survived,’ although in this context it is worth noting that the current model of schooling is but 150 years old. However, we feel we have actually rather pulled our punches by focusing on the implication of human learning, human agency, human employment, and human fulfillment. There are many other reasons that we should also be worried (or at least on high alert) about the latest advances in AI.
I don’t agree with everything they’ve suggested. But the piece did move me in a way that I haven’t been moved by something in education in quite some time. And that alone is worth paying attention to.
Last Word? Last Word!
One of my favorite authors is Jim Harrison. Perhaps best known for his poetry, he was incredibly prolific throughout his career, and passed in 2016. When I say that I have read everything he ever wrote, it’s not hyperbole. Below is a passage from this longer piece, The Gospel According to Jim, a profile of Harrison, by Chris Dombrowski.
Driving home on the Burma Road, we pass an old dilapidated house — doorless, windowless, roof caved in by a windfall cottonwood — home, if you ask the locals, to one of the largest, most seething dens of rattlesnakes in the valley.
“Son, do you see that old house?” Jim says.
“Sure I do.”
“Good. Do you know what it says?”
“No, what does it say?”
“It says, don’t let your life become the sloppy leftovers of your work.”
Here’s to not letting your life become the sloppy leftovers of your work. I’m going to ride my bike many miles this weekend and eat the cantaloupe, watermelon, and tomatoes from our gardens out back.
It also has a fight waiting at the end of it. Like baseball, CandyLand has unwritten laws that are immutable.
I am aghast at my children’s collective apathy around knee hockey. They are 8, 6, and 3. They play ice hockey. They can rollerblade. When it comes to knee hockey, they are often clueless. This is my first significant failure as a father.
This is my favorite internet thing perhaps ever? If you don’t know his work 17776 about the future of football, do yourself a favor and spend some time with it.
I will note that I do use custom instructions in GPT-4, providing some ground level parameters and framework to make my use of ChatGPT more efficient and beneficial. A recent addition, I’ve been really impressed by the way it integrates these instructions into its responses. Again, play with what custom instructions you might want to deploy.
Annotation, FTW!
I really liked this one. I added nothing here.
I was struck by the way it used ARE in present sense. I added nothing here.
I was floored by its use of “pragmatic enthusiasm.” I even asked it about it in the chat transcript. And it referenced some of the custom instructions I had given it. Again, very weird moment here in the chat. The Skynet-school board line is a good one!
This was surprisingly poignant and authentic. I added nothing here. But again, it was likely drawing from the inputs I had given in the custom instructions.
The process? I am having a hard time trusting it in Sixers land!