How A.I. Will Be Impacting Informal STEM Education... Sooner Than You'd Think
ChatGPT is an A.I. model developed by OpenAI and launched for research testing in November 2022. Models like ChatGPT are trained to synthesize information, provide answers to user inquiries, produce written copy in different writing styles, and more.
I think of it as similar to a search engine, but way smarter. Sure, it can spit out an answer to a question, like “why does deep-sea gigantism exist,” just like entering that inquiry into Google will offer you website hits and articles about that topic. But, rather than wading through pages and pages of results (often optimized for SEO and not necessarily the best content), ChatGPT will give you one comprehensive answer. In that way, it minimizes the amount of brainpower needed to sort through the hits and find the right answer. But that’s not all it does.
Try asking it to synthesize information for you. For example, “read and summarize this article in 10 bullet points.” And if the answer is not quite what you’re looking for? Ask it to clarify or change the results. For example, “Give me more information about the methods used in the study” or “actually, condense that to 5 bullet points.”
Now, where it gets really cool is the ability to synthesize and organize information. One popular utilization of this is asking the model to design a meal plan for you based on certain parameters. But it’s not as simple as you think. You could say “design me a 5-day meal plan, with three meals a day, totaling 2,500 calories, with an emphasis on fighting chronic inflammation, for someone who is a vegan. Oh, and I don’t like seafood.” And - boom! You’ve got a beautiful meal plan laid out for you. Imagine trying to create that using a tool like Google and you’ll start to understand these major differences.
I’ve been playing with this platform for a few weeks now, experimenting with questions relating to my personal and professional lives. As an evaluator, I was interested in how well the model could perform some of the lower-level writing and planning tasks I do often in my work. I was also interested in how well it might be able to synthesize information about important topics in informal STEM education (ISE) evaluation and even predict trends we might see in the space of ISE in the future. Here are some of the experiments I ran:
Draft a consent form - I asked the model to draft a consent form for a fictitious youth-serving summer program. ChatGPT scoured its internal database of information that has been fed to it by programmers and gave me a pretty good consent form. Sure, the form needed to be customized to more parameters around the project, myself, and my organization, and it was not fully correct in terms of what needed to be included or how things should be presented, but it was pretty darn good for a first draft.
Draft a survey - Drawing from the last example of the fictitious summer program, I entered some parameters about the program, the audience, and the learning outcomes I was seeking to assess. This time, its output was a little impressive. It produced questions that were too simplified and exacting to my specifications. For example, without the context of what the fictitious program consisted of and what learning concepts it taught, a learning outcome of “Teach youth about the environment” became a survey question of “Did you learn anything about the environment?” Not the best. Sure, it might be improved with more program-specific context, but I’m unlikely to use it to craft instrument questions as I’m not sure the benefit if you need to translate all the context and understanding from your own brain into the tool. It’s not much of a tool if you need to replicate that much work.
Identify trends in the ISE space - This was the most interesting (and abstract) ask of all. I requested the model identify some future trends in ISE that might occur in the next 10 years. It identified five trends that won’t be too surprising to anyone in the ISE field, but that were also (I believe) quite accurate.
Here is where it is important to note that the model is not aware of the entirety of content like this nor the whole web. It has been fed a subset of data by its programmers and then trained using research participants - like me! That means that while - at first glance - its outputs look pretty good, experts will recognize that most everything it produces needs at least some work.
And this is where the danger lies. Beginners in a field might try using this content as-is, without recognizing the ways in which context and experience might necessitate a change to what the chatbot produces. I really view it as a tool to help with “first pass” work, followed by an expert reading over results finely for mistakes or misattributions.
So, in my view, it does not replace the need for white-collar workers (like museum educators or program evaluators), but it gives us a leg-up and a time-saving mechanism that allows us to spend more of our time doing work that is higher-level, creative, interpersonal, and uniquely ours.
This also introduces the question of how A.I. will become a normal part of our work, and likely sooner than you think. At one point in time, the Internet was thought of as a niche tech advancement that only geeks were interested in (self-identified geek here). Now, it’s ubiquitous. The same will likely happen with A.I. So, rather than fighting against this advancement or looking for ways to curtail its use, we ought to welcome it and identify ways to use it for our advancement.
Formal education has had an immediate pushback to its use, with students asking the model to write essays for class, among other things. But - much like the calculator - we should see A.I. as a tool rather than a cheating mechanism.
Perhaps this is the time for us to reevaluate what it means to assess learning in the classroom. Similarly, informal educators should consider what A.I. could mean for their programs and their audiences.
One thing I’ve found super helpful about ChatGPT is the “just-in-time” information component. If I encounter something I’m interested in learning more about, or if I have a specific question about something, I now find myself reaching first for ChatGPT over a search engine. Granted, some information literacy and critical thinking are needed here (remember: the model does not know all information, is still in a research phase, and does make mistakes), but it’s made it so much quicker and easier to find the specific information I need at the time I need it.
It’s also important that we pay close attention to our changing field and put safeguards in place to ensure A.I. models are not replicating inequity. As we know, programming and programmers can hold a lot of bias. But, used correctly, this type of tool can produce better learning experiences and more equitable outcomes. It all depends on how we use it.
Here are two imagined ways in which ISE professionals might use A.I. in the near future.
Future Application One:
Imagine a kiosk in a museum where visitors can look at an artifact and ask A.I. real-time questions about it. Now, imagine that the kiosk can collect data on common visitor question themes (as an evaluator, I’d love this) and then create a recommendation for permanent signage. Or, maybe permanent signage won’t need to exist anymore. Maybe A.I.-informed signage that digests real-time news and scientific breakthroughs, plus visitor inquiries, becomes the standard for crafting information that is more timely and relevant to visitors.
Future Application Two:
Pretend you’re creating a bird-watching program for teens. You ask them to take their devices (iPads, cellphones, etc.) out into the field and collect photos of birds that they see. Then, they use A.I. to identify the types of birds, create a digital checklist of common birds found in their neighborhood, and adapt it using weather and seasonal data for what birds they might see when. Now, not only have you created a program that allows teens to learn about birds, they’ve become more tech-literate and created a tool for others to use, focused on place-based learning. How cool!
The potential applications are endless, and as A.I. models like ChatGPT continue to advance, we have the opportunity to welcome these advancements into our practice.
If you enjoyed this post, follow along with Improved Insights by signing up for our monthly newsletter. Subscribers get first access to our blog posts, as well as Improved Insights updates and our 60-Second Suggestions. Join us!