Composing with AI

ChatGPT and AI Writing (Sent to WP Faculty on 12/29/22)

Hi, friends.

Grades are done and I had a little time, so I thought I'd send out some of my thoughts about a new technology that is making major waves. I think as writing teachers we should know what this is and what's coming, because next quarter your students will definitely be using this technology with their assignments. How do we deal with that? It'll be up to you, but you should know the ins and outs of what's going on here to help you forge your path.

What is ChatGPT?

The technology that powers ChatGPT is built on OpenAI's GPT-3.5 model. It works by synthesizing the language patterns of billions of parameters of text. The GPT-3 model has been around since 2020; it’s quite possible you’ve already been given its writing from some of your more tech-savvy students.

The two big things that ChatGPT brought to the mix, here, are accessibility and context: you can ask the AI in a conversational way to do the things you want it to do, and then carry on that conversation to make modifications to the output ('make it more dramatic' or 'incorporate a counter-argument', etc.).

What can it do?

It can do a lot. You can ask it to write arguments, jokes, music, screenplays, syllabi, and, yes, even full essays (not good essays, but full essays). It can even write code. Really, you should just try it out yourself. Go to https://chat.openai.com/" , make an account, and start talking to the bot. It still manages to give me a chill every time it outputs content for what I thought would be an impossible prompt.

What can't it do?

It can't do a lot. The most glaring weakness of the AI right now is its inability to find, source, and convey reliable, fact checked data. It puts forward the exact habits we try to wrestle our students out of--it confidently asserts information without providing the source. If you press it to provide sourcing, it's just as likely to make up a source as it is to give you a legitimate one. This is because of the wider limitation to its mechanism: it's not writing with rhetorical intent. So basic writing it can do by weaving together the language patterns about a topic that exist out there. But higher-level writing, which is purposefully sourced and built to forward a particular rhetorical and critical goal--it can not. That said, this tech is getting better with each model. It's evolving, and I think with the increased attention, we're going to see a Moore's-Law style explosion in model content and quality going forward.

AI [and / with / vs] Us

Even so, I think the concept of rhetorical intent is going to be the technological wall here, and that is the key area of discussion and intervention that we can start thinking about as writing teachers. This is a matter of syntax versus semantics. AI can do the former, but it can't do the latter, and if it ever can do the latter, it’ll be able to think for itself (and at that stage we have a whole new frontier of issues to work through).

We weave together language patterns all the time in our writing, just as the AI does. The difference for us is that we weave together language patterns for our specific rhetorical intent. The AI does not and can not. To be clear—and I think we should try to make this clear to our students—the output of the AI has no more inherent rhetorical purpose than a random word generator. But a good writer can use even a random word generator to inspire beautiful writing. When we interact with the AI, we do so with our own rhetorical intention. What do we get, then, with that equation? Our rhetorical intent, its language patterns: I think the answers there might be the landscape of a lot of writing in the future.

Which is what we want to teach our students to do—to write, to communicate, to work and collaborate out in the "real world." If the real world is going to integrate AI in the work process just like it does computers and other technologies, we would want our students to be able to do so as well, but with a full understanding of the strengths and limitations of the AI, and to be able to use it as a conduit for their own rhetorical intent—not to surrender their rhetorical agency to the AI's (which, again, doesn't actually exist).

One of the big names in my research field, Ian Bogost, has a take on all this that I think is pretty spot on in his piece in The Atlantic:

The AI doesn’t understand or even compose text. It offers a way to probe text, to play with text, to mold and shape an infinity of prose across a huge variety of domains, including literature and science and shitposting, into structures in which further questions can be asked and, on occasion, answered.

My Approach

Which takes me to where I'm at with all of this. I plan to teach students the very ideas I just worked through in this piece, make sure they know about the benefits and limitations, and categorize the use of AI alongside every other tool that I teach them as I work to help them find their voices, learn the genres and conventions they need to know, research and think critically. When they turn in writing, I will evaluate it deeply and fully, just as I always do. If the writing is thorough, thoughtful, well sourced, accurate, and purposeful, with every choice deliberately forwarding their rhetorical goals—that's good writing, created with intentionality. The AI might help them generate ideas, areas of inquiry, template shapes, and drafting materials, but it won’t do all the rest for them: it still requires a good writer to produce good writing.

This will involve asking for assignments that require higher order, critical, intentional, and unique approaches. The classic simple summary essay on a piece of literature won't cut it; that has been completely automated at this stage. Though with an endless supply of downloadable papers out there, that's been the case for a while.

Further Reading
ChatGPT as a Coding Assistant:
https://twitter.com/goodside/status/1599082185402642432
ChatGPT as a tutor:
https://twitter.com/pwang/status/1599520310466080771
ChatGPT imagining stuff within stuff:
https://twitter.com/gfodor/status/1599220837999345664
ChatGPT errors, mistruths, failures, etc.:
https://twitter.com/taranjeetio/status/1601600143043620866
ChatGPT working as/with the professor:
https://oneusefulthing.substack.com/p/the-mechanical-professor
Will we create software that can detect this AI writing? People are working on it, but I’m doubtful:
https://twitter.com/JanelleCShane/status/1601729685385535488