Shall we play a game?
Okay, I’ll start. “Peanut butter and…”?
What came to mind? Did you think “Jelly”?
Well, that’s right! That’s the answer. But… I never told you the rules.
I guess you could tell you were supposed to fill in the blank, guessing not just A new word but the most LIKELY next word.
Also, did you only guess one word or did you keep going, such as “Peanut butter and jelly is a great sandwich to have on a picnic”? I suspect not.
So you also knew to fill in the blank with not just a string of words but one word.
Let’s go another round.
Ready?
“Happy birthday to….”
Did you guess “you”?
You did it again – you filled in the next, most likely single world.
Ah, but what if I told you we were at an event for Sharon. And it was her birthday.
Round 3: “Happy birthday to….”
Did you guess “Sharon”?
My question didn’t change, but your answer did. What’s different this time? The context. Without me telling you, you changed the rule from the most likely single word to the most likely word IN CONTEXT.
And those three rules, in a generalized way, is how emergent AIs like ChatGPT work: guess the next single word, in context. The tools can produce so many remarkably complex things, things that seem more powerful than just these simple rules, but in the end that’s all it comes down to.
So at it’s core, understanding emergent AIs can be as easy as one, two…
(Did you guess “three”? That is: ). The next 2) single word 3) in context)
For almost three decades, I have taken personal and professional pleasure in taking deep dives into the latest disruptive technology and exploring its pro-social and educational potential. Web design (1995), online discussions (1998), video games (2002), virtual worlds (2005), XR (2016), and more.
While emergent AI is just my latest, it caught me by surprise and is moving at a speed unlike anything I have ever experienced before.
Last November I threw myself into exploring the tools, not just for producing text, but also images, video and audio. Last December I wrote about the experience after my first few weeks (Can generating art through AI still make me an artist if ordering food on GrubHub doesn’t make me a chef?) In it, I shared how I came to realize AI is not a competitor but a collaborator, not for exploring the realm of what is real (epistemology) but exploring what might be (aesthetics).
Since then I have integrated it into every aspect of my personal and professional life. I am using it to design a card game, Uncannny Valley, designed to teach how to view the world as an AI. I use it to write legal documents, and my kids’ camp emails, and make sense of obfuscating medical instructions.
What led me to write this particular post, which I wanted to do for a few months, was that it has led me to get in front of groups of educators and non-profit professionals to help them understand this latest disruption to their classrooms and projects.
As I write this mid-summer, I have already delivered or booked four different events which by the end of this year will have brought me before 1,000 people. I never expected to become the expert they are looking for, but, well, here I am.
What I hope to do below is share HOW I am approaching the topic, in case it is of interest or use to others.
The three things that have proven most effective have been:
- AI Game Shows
- AI Arcades
- AI Roleplay
AI Games Shows
Many in my audience are scared. Their concerns are existential. AIs will replace them, or change their classrooms or workplace in ways more profound that COVID-19.
In that context, a lecture – no matter how engaging – is a non-starter. Instead, I turn the mood around. I start with an over-the-top, dramatic game show, music blasting as I enter the stage. Working with my partners we then put on a 45-60 minute show that delves deep into what generative AIs are, and what role educators can have in shaping their future role in society… but we have a lot of fun along the way.
There are three key sections. The first section is a game played with the audience, sharing a computer up on a screen. It is called CONTEXTO. It is simple: guess the word. As you guess words, you learn how close your guess is to the secret word of the day. With each guess you try to guess closer and closer until you figure it out. The neat trick here is the logic behind those relationships; you do better if your guess is more likely to appear with the secret word (within news articles, the basis for this one particular AI model). That essentially is how large language models work within emergent AIs. As we guess correct words closer and closer to the secret word, we start to see patterns. We ascribe meaning to our correct guesses. But this is false; the AI doesn’t understand why the connections exist between these words; it just knows that these connections exist, and that, it turns out, is enough. So this activity is a great way to playfully teach how AIs view the world.
The first section is also secretly the game show recruitment phase; all participants who suggest words are invited to join a team. In the second phase, those teams are introduced to the audience. Like in an improv show, the audience is then invited to provide some prompts (ideally related to the topic of the gathering), in categories like character, settings, genre, and challenge. Then each team has a few minutes to rapidly write a story with chatGPT, as a creative collaborator, and share it; the audience then votes while guessing which prompts were selected.
After each phase, teams get points. In the final round, with double points, each team takes one sentence from their story and generates a related image in Midjourney, to accompany their story. While all these stories are being written and images being generated, my and my co-host banter and highlight key aspects of the opportunities and challenges (ethical and otherwise) of working with generative AI.
A small trophy is given to the winning team but everyone goes home happy… and a little more informed about how generative AIs work and the role we can all play to advocate for a pro-social outcome.
AI Arcades
While the game shows are designed for an audience, AI Arcades are designed to maximize crowds and hands-on engagement. The first one I did set out 8 stations – some at tables, some at high boys for those standing. Each one had a guide and a different form of generative AI to play with. The idea here is that until you get past that first hurdle, you’ll never get over the fear of the unknown and start to have agency around your relationship with the tools and the impact they are having. In March we focused on:
- ChatGPT (text generation)
- Midjourney (image generation)
- Lensa (personal avatar generation)
- Parodist (celebrity voice/video generation)
- Character.ai (chat bot generation)
- Alexa (voice chat)
- Otter.ai (AI-driven audio transcription)
- Uncannny Valley (my card game to welcome our future AI overlords!)
(On the first time I did this, I read the news on the way into the session: Character.ai had just been given a valuation of $1B).
AI Roleplay
The third approach I’ve been offering is for the more serious-minded. This maximizes hands-on, but focuses on the immediately practical applications of these tools.
Everyone gets assigned to a team, representing a department within a company. Our company then sets out a new mission, such as to launch a new fundraising campaign. Each team then uses a different tool to collaborate with AIs to produce their piece of the project. One group writes a press release. Another creates a social media campaign. Another produces image assets for the first two to use. And so on.
At the end of the session, each “department” shows off their piece of the project and then, breaking roles, we explore the advantages and challenges of using generative AIs to super-power our work.
Unlike with other consulting, this space is changing so fast, not just technologically but socially. There is much more concern this summer than in spring that generative AIs are built on unpaid labor. It is informing nation-wide strikes and lawsuits. And whether or not people are more informed, they are certainly more aware.
Take for example this webinar that was canceled today. I am a member of the Author’s Guild. This summer over ten thousand members signed a petition that essentially says, “You can’t steal our jobs.” Next week was a webinar called “How to Use AI Writing Programs to Your Benefit” that they offered in response to members interest in the use of AI applications in their writing profession. In advance, registrants submitted hundreds of questions, and so far outside the scope of the Webinar they decided to scrap it and redesign it as a series later this fall, which will “allow for discussion and questions around the more nuanced elements of AI use.”
That’s all well and good. But what I’ve learned, given the current rate of change, they might as well offer it now. Because whatever they plan to do in a few months is going to likely be outdated by the time it is offered.
Luckily, I am comfortable with a high rate of change.
