It’s sometimes said that you should never begin with an apology, but I will, because the last promised blog post hasn’t appeared yet. It will come – it’s just been sidelined a bit. In my defence – if I can have one – it’s down to the emergence of a little thing that looks like it’s going to significantly impact on my current doctoral project. That little thing is called ChatGPT and I’ve been thinking about, and researching, the possibilities of life with this AI wizard as a de facto project team member.
This has led me to the conclusion that we have three possible reactions: Flight, Fight…or Collaborate.
Option 1: Flight – when you look at the growth in interest and users in just the last couple of months, I don’t fancy our chances of running away from it. The cork is well and truly out of that particular bottle, and the genie walks among us. The fact is that change is scary, and paying attention to extreme speculation might have us wishing for a slower pace of change because we don’t know how to cope with this new thing in our midst. In a world that is becoming increasingly interconnected and intercommunicated, we have fewer places left to hide from change and less time in which to do it.
Option 2: Fight – if we take a view that AI is some form of new evil when it comes to learners not “playing the game” when it comes to engaging in deep and meaningful learning (read: Critical Thinking), then we must fight it and find new ways to detect work that hasn’t been completely authored by the learner.
We could champion the development of software that will allow the detection of this ChatGPT-produced work – but the algorithm has produced totally original work. Do we then start to distrust everything submitted in writing that wasn’t produced under controlled conditions where access to external tools (because that’s what this thing is) is forbidden/made impossible ? Do we close our eyes and try to imagine the learner’s voice saying the exact words on the screen or on the page (which I have often done)? If they got some editing help with their language, we may not hear what we imagine their voice to be, even though they created the substantive content before us. Frankly – do we have time to do that? If our cohort of learners is large enough that we are using adjunct markers who don’t actually know the learners, would they even be able to do this? Are we simply creating a more adversarial relationship with our learners by distrusting them as a first position?
Option 3: Collaborate – Imagine a typical workplace situation. We are part of a team, and we have a project we need to complete. The boss hasn’t micro-managed who is going to do what – that’s been left for the team to work out between themselves. There are two things the boss does expect, though:
- The team will use all the tools at its disposal to complete the project.
- The team will deliver what it has been tasked with.
I’m not totally fixated who wrote which part of a collaborative document (because the reality is that each member is likely to specialise in particular parts of the task [and we could tell a whole new story about how that gets sorted out]) – but I expect each team member to be able to defend every part of the output (without necessarily being an expert in each one).
Perhaps the philosophical question becomes whether ChatGPT is a tool or a team member – at what point does the algorithm become “sentient ”. Think of what it’s capable of doing that you would expect of a good human team member:
- Does exactly what is asked of it.
- Does it on time.
- Does it without complaining.
- Can suggest alternative approaches to what it’s put forward and seek clarification so it can correct and improve its own work.
- On 24/7 (maybe we don’t expect quite that level of commitment from a human team member, but you understand where I’m going with this).
Disadvantages on the other hand:
- Works with what’s available on the web – so it won’t see what’s not on the web.
- Can’t experience visceral reactions to what is happening in real time in front of it (but that might be a possible advantage sometimes).
A traditionally important measure of academic quality has been the assumption that work submitted for assessment is the learner’s own work. What does that really mean, though? Perhaps the closest we’ve come to being able to assure ourselves of that is the traditional proctored examination, where the learner performs under controlled conditions, including constant supervision. Practical tasks that are performed under supervision, or which may involve structured role plays, also fall into this area. When it comes to project work performed outside the classroom – which would be the majority of collaborative project work – we have to accept the learner’s attestation that the submitted work is their own.
One of the issues with work produced collaboratively is separating out who really created and therefore “owns” which part of it. Let’s assume that Johnny claims ownership of a certain part of his team’s project report. The team worked on the project collaboratively; everybody’s had input into every part of the report, so his output isn’t totally his own – it’s the result of various inputs from other members of the team.
So, what if ChatGPT is used to write Johnny’s output? It might do in minutes what it would take him hours, days, or weeks to do. Effectively, Johnny has outsourced the production of his output, with the exception of guiding ChatGPT in what he wants it to produce. Since he didn’t write all of the output himself, or produce any of the graphics that accompany the words, does that mean he cheated ? I think there are two issues here. The first is whether or not Johnny “owns” the ChatGPT output since he didn’t put all the words on the page himself. In favour of the answer being “Yes” I offer the following points:
- He had to think critically to provide ChatGPT the parameters within which to work – which involved Johnny in the learning necessary to set appropriate parameters.
- He had to check the output from ChatGPT to ensure it met the team’s requirements – which means he needed sufficient learning to make appropriate judgement calls about whether Chat GPT had appropriately synthesized the information included in the output. He then had to explain his judgements to the rest of the team when they were thinking about how that output was going to integrate with their work.
- Where ChatGPT critiqued its own work and offered suggestions if more than one option was possible to address a parameter Johnny provided, he had to critically assess the suggestions and make decisions about which option was the most appropriate.
The second is whether or not Johnny contracted out of learning by using ChatGPT. In favour of the answer being “No”, Johnny hasn’t contracted out of learning in the same way as somebody who paid another person to write an assignment for them because they either couldn’t, or didn’t want to, do the work themselves. He may not have done his learning in the same way we’ve thought about it pre-ChatGPT, but he’s still done it – refer to the previous bullet points.
The work may have been done more quickly than if he did all the donkeywork himself, but we need to ask ourselves whether we are still fixating on the idea that hours spent is the best way of measuring learning done. As a young apprentice, I remember being told that it takes 10,000 hours to become “expert” at something – but there were some things where I’d reached an appropriate standard long before that number came up. I was being judged on standards that were industry-appropriate, and not being held back because I reached the standard more quickly than some formula said I should have.
We still have the opportunity to question Johnny at the project presentation, just like we’ve always had – not because we’re trying to trap him in the areas where he’s maybe not so “expert”, but because he’s opened a door for an inquisitive audience and we want him to show us around what’s on the other side.
I’m interested in learning about your thinking on this topic, because it’s through collaborative conversations that we can create even more valuable learning and development experiences. Let me know what you think.
In my next blog on Thursday 26 March, I will build on my thoughts here and share more of my thinking about issues associated with ChatGPT and some of the relationship issues associated with it being a de facto project team member
Rob Nelson Education 