Creating and playing games

The closer a learning product is to the bottom of Bloom’s Taxonomy, the greater the risk of being commoditized by AI

December 7, 2022
Jonathan Libov

I, for one, welcome our AI underlords. So does Ben Thompson. Yesterday in AI Homework:

Instead of insisting on top-down control of information, embrace abundance, and entrust individuals to figure it out. In the case of AI, don’t ban it for students — or anyone else for that matter; leverage it to create an educational model that starts with the assumption that content is free and the real skill is editing it into something true or beautiful; only then will it be valuable and reliable.

Come at the king (of business illustration) you best not miss, so I’ll do my best to illustrate what I think AI means for learning and education. One of the bedrocks of educational theory is Bloom’s Taxonomy, a framework for ranking categories of learning from simple to complex, concrete to abstract, and strictly necessary to aspirational. Here it is, without markup:

Bloom’s Taxonomy Bloom’s Taxonomy

As far as academic frameworks go, Bloom’s Taxonomy is remarkably durable and intuitive (it’s probably durable because it’s so intuitive). It's the framework through which I think all these What happens to learning in a ChatGPT world? should be viewed.

The most common but also the most rote form of learning is Remembering. Utilitarian as it may be, you need to memorize the dates of the Battle of Yorktown in order to understand the story of the Revolutionary War, much as you need to memorize a chunk of the Periodic Table of Elements. All the way at the top of Bloom’s Taxonomy you have Creation, the most elusive and aspirational form of learning. That is, to truly understand a subject matter, you must be capable of producing a novel work or story, or even a novel joke, with it.

Learning products in Bloom's Taxonomy

If you take Bloom’s Taxonomy as the model for learning and plug in all of today’s tooling—now maybe including ChatGPT—you get something like this:

Bloom’s Taxonomy Market Map Bloom’s Taxonomy wrt commoditization

The closer you are to the bottom of Bloom’s Taxonomy, the greater the risk of being commoditized by AI. Paraphrasing Wikipedia on your way to a five-paragraph essay isn’t exactly new, and it’s now even easier with GPT3. It doesn’t even seem far off at all that ChatGPT could reproduce rote exercises and utilities like flashcards, coding and math exercises.

It's certainly true that creating flashcards is actually more useful than practicing with flashcards you found. This is foundational to Anki and even Quizlet—my understanding is that Quizlet's marketplace of flashcards belies the core, valuable activity of creating flashcards on your own or with friends. But we're talking human behavior here; ChatGPT flashcards are to self-made flashcards as Keurig is to roasting and grinding your own coffee beans. The former will dominate.

I think Ben Thompson mostly got it right. Evaluating whether and in what ways the AI captured the subject matter accurately is not only an interesting exercise (albeit a little contrived), it’s arguably more valuable than re-producing the material that other learners have already produced a million times over. Flashcards and exercises are commodity products that computers can generate better than humans. We simply had an interim period where some online tools helped facilitate that kind of production because the internet transforms distribution.

As Austin Allred, founder of the coding academy BloomTech (yes it's the same Bloom), pointed out yesterday, machines can accelerate all this rote learning—Remembering and Understanding in Bloom's Taxonomy terms—by enabling more people to play more games against the computer (more on that below).

At the tippy top of the Taxonomy, the outlook for learning and education is arguably brighter than ever. Because all the utilities at the bottom will get commoditized, we have more time to spend in the most valuable area of learning: Creation and Evaluation. What’s more, new tools and services make Creation and Evaluation easier than they’ve ever been. Think:

  • Creating games on Replit

  • Telling authentic stories via Flip

  • Creating learning memes on Antimatter (full and shameless disclosure: I’m the founder of Antimatter)

A Cat and a Mouse

Why won’t creating games, telling stories, or creaing memes be commoditized the way exercises and utilities will? Because, let’s not forget, ChatGPT isn’t studying history so much as it’s learning from the stories that humans have told about history. Every novel story created by a human is a story that ChatGPT has yet to learn. And every story told by ChatGPT is potential raw material for humans to tell a new, novel story. We’re one step ahead by definition.

Consider these memes created by students on Antimatter about the Revolutionary War. Or this Guess the MacBeth Character meme created by a Teacher on Antimatter:

MacBeth To be or not to be (intensely human)

There remains a very wide gap between AI’s ability to tell good jokes or puzzle-like stories and what humans are able to produce. This isn’t totally a coincidence. Memes are in some ways a reaction to the legibility of the internet today, the very same legibility that serves as one of the foundations of LLM’s. It’s inevitable that LLM’s will be capable of being genuinely funny or creating real puzzles, but it may be closer to the last mile than wherever we are today, and it may turn out to be an infinite cat-and-mouse game.

More Games

Speaking of games, it's my general belief that technology inevitably makes everything more game-like. Less time putzing around, more time playing measurable status games on Facebook and Twitter for Likes and Retweets or actual games on Fortnite, Minecraft, etc. Less time watching sports, more time playing fantasy sports. Less time saving with a steady APR, more time investing in markets that go up and down.

In a sense, the What happens to Learning and Education in a world where content is free and AI is abundant? question could be re-phrased in game-like terms: Students + AI vs. Education. To be clear, that's the fear ChatGPT is drumming up; it's not and never should be adversarial. Still, to that end, I think it’s instructive to look at the diverging outcomes of the two games that have been solved in some form by math.

Chess. It seems quaint now to reflect on the time when we weren’t sure if computers would ever solve chess. They did, long ago, but it’s uplifting to think that chess is nevertheless more popular than ever. To boot, people still love playing against the computer, or against a highly liquid supply of human competitors, at incrementally increasing levels of difficulty. This is a testament to the durability of puzzles.

Baseball. Math hasn't ruined baseball so much as baseball management's inertia in saving the game from the math. The math tells managers to pull starting pitchers swap relievers in and out according to their as-little-as-one-batter specialities; the math also teaches you not to steal bases and to shift infielders in ways that eliminate the mano a mano qualities that once made baseball fun to watch. Whether it's out of inertia or nostalgia, baseball's management has been dreadfully slow in adjusting the game to preserve what made it special. It's a lesson in inaction.

While chess and baseball's courses aren't diametrically opposed—baseball needs to be saved while chess naturally fits more seamlessly into the world we live in now—we're nevertheless presented with two models of dealing with the math. With respect to learning and education, we should avoid baseball's inertia and nostalgia at all costs. To reiterate Ben Thompson's point, "Instead of insisting on top-down control of information, embrace abundance, and entrust individuals to figure it out."

Similarly, we should embrace the infinite distribution of challenges that computers and the internet have brought to the world of chess. Consider how viewers of chess game streams look at the model during the game, which enhances the conversation around the game and makes everyone smarter. More liquidity, more infinite games for learners to play. There's little to fear but inaction.