Educational drift: what ChatGPT means for higher education
I've put online the session I gave colleagues at the Business School on ChatGPT, because the conversations afterwards suggested it was worth a wider audience. It's long, nearly two hours, and it was recorded four months into something none of us fully understand. Treat it as a set of first principles rather than a set of answers. The video is at the bottom. Here's the argument.
The gap I've been chasing
Anyone who has taught strategy knows Johnson's drift model. One line for how fast the environment changes, a lower line for how fast the organisation changes, and the gap between them is strategic drift. Kodak is the textbook case.
For a while now I've been adding a third line, lower still, for how fast education changes. The gap between organisations and the environment I call the utility gap. The gap between education and the environment I call the relevance gap. Closing those gaps is why I do this job. I came across generative AI the way a strategist comes across anything, by scanning the environment, and I've been following it since the launch at the end of November. Both gaps have just widened overnight.
What this thing is, and isn't
Most of the session is explanation, because I think understanding it is the precondition for deciding anything. Four points carry most of the weight.
It completes the sequence. A large language model turns your words into numbers, predicts what comes next, and turns the numbers back into words. Two, four, six, eight in; ten, twelve, fourteen, sixteen out, made billions of times more complicated. It's a few lines of code with billions of dials, and you set the dials by running an enormous amount of text through it until "to be or not" reliably came out as "to be".
It doesn't look anything up. That's the paradigm shift, and it's the one I'd ask every academic to hold onto. Google is retrieval AI: it fetches. Generative AI generates from the model, with no lookup at all. It's closer to how I'm speaking right now, which isn't retrieval either. I'm not remembering sentences. I'm predicting them.
It has no common sense and no understanding, and where the training data is thin it will fill the blanks with confident invention. Ask it about strategic management and it synthesises a hundred thousand publications well. Ask it about rollerblading in Manchester or sheep farming in the Peak District and it'll make things up. Think of it as a personal assistant with an IQ of eighty and near-infinite knowledge up to 2021.
And two waves of innovation are already under way. The first puts the AI into applications: Bing, Copilot, and the thousand-plus tools on Futurepedia, most of them a front end on the same models. The second puts applications into the AI, which is what plugins do. Both are moving faster than any of our committees.
The juice
Here's the frame I'd most like colleagues to take away. Knowledge workers produce intellectual output, and it has four parts. Context: the background reading, the literature review. The juice: the questions, the framing of the debate, the insight and the analysis. Communication: the essay, the report, the presentation. And promotion: telling people it exists.
Colour in what the AI can do today, with a bit of hard work on the prompting. Context, most of the analysis, communication, and promotion all turn blue. What stays white is the juice. It can't ask the questions. It can't frame the debate. It can't work out how to use itself in your field.
Analysts are predicting that this quadruples the output of a knowledge worker. If that's even half right, our job becomes clear: graduate people who produce excellent juice and use AI fluently for everything else. Call it the 4x MBA, or the 4x graduate. The risk, if we're not careful, is that we graduate button pushers instead.
Assessment, and the calculator
Which brings me to the question every academic asks first. If it can write the 3,000-word essay, how do we stop students cheating? My answer is that we've been setting the wrong test. Setting an essay in the blue areas is like setting a mental arithmetic exam when every student has a calculator at home. Detection is a losing game. Design is the answer: assess for juice. The questions students ask, the way they frame a problem, the quality of the insight, and how well they direct the tools to do the rest.
That means the curriculum changes as we change. AI has to run horizontally through everything we teach, not sit in one module. And staff have to build their own literacy first, because you can't assess what you don't understand. Think of a student and their AI graduating together, as a pair, ready to be four times as productive from day one. The other threat, the one we talk about less, is not doing this fast enough. If we don't embed AI into our curricula quickly, we're letting our students down.
Institutional inertia versus a culture of learning
None of this is in a textbook yet. The knowledge is on Twitter, YouTube, blogs, and arXiv, and the only way to stay current is to have a go and read what you find with a critical academic eye. Institutional inertia comes to mind. So I proposed some things I can do something about: departmental AI leads and an AI corner in every monthly meeting, a termly gathering to share what's working, a day on 20 May, the ChatGPT Challenge, taking over the whole ground floor of the building so staff, students, and local businesses can play with the tools together, and a couple of bureaucracy-busting ideas, like a small prepaid card for every academic to spend on AI tools without a procurement form, and a workload model that lets people build something first and claim the time back afterwards.
Watch the session
The full recording is below. It includes the history of OpenAI, the plugins demo, a detour into whether GPT-4 could be conscious, and the implications for students, academics, and the wider economy that I've only sketched here.
If you're working on any of this in your own institution, get in touch. I'd like to compare notes, and I'd like to be wrong about the inertia.
Rob Dixon teaches strategic management at Manchester Metropolitan University Business School and works with organisations on building AI capability.
Comments