AI platform ChatGPT reveals the promise and the peril of this new technology

Mike Seymour, an artificial intelligence specialist at the University of Sydney Business School

Sure, we all know artificial intelligence is rapidly improving. But something has just happened that causes a jolting realisation that the world has changed, and maybe not in a way that we will like.

Two weeks ago OpenAI, a San Francisco-based, investor-funded research group, released ChatGPT, a “free research preview” of a conversational AI platform that is a quantum leap ahead of anything else previously available to the public.

If you haven’t yet tried it, go to chat.openai.com and see for yourself.

Ask it anything. It’s way simpler than Google. Type in a request and you get a written answer in clear and plain language that is drawn from the trillions of words on the internet that have been used to train it. Not only does it have a solid grasp of the facts about nearly everything, it’s also capable of cogent analysis. It can mimic, often hilariously, many different literary styles.

ChatGPT is well on the way to fulfilling the dreams and the nightmares about our AI future. It’s capable of being the friendly bot that expands our knowledge and eases our burdens, but also looms as a threat that can do knowledge economy jobs better, and much cheaper, than people.

Bots such as ChatGPT will have a huge impact on almost any activity carried out by humans. But let’s start with education.

Have you forgotten a definition or a concept? Ask ChatGPT. It’s competent in almost any subject.

Need help with an essay that’s due in an hour that, unaccountably, you neglected to start work on? Assign it to ChatGPT. You’ll have a thousand or more entirely plausible words in a few seconds. It might not earn a high distinction but it will probably be serviceable. And if you don’t like the bot’s first draft, refine your question and ask it again.

You’ll also take comfort in the knowledge that your friendly ghost writer won’t trigger a plagiarism checker because its words will be original.

Are you stuck on a maths or science question? Ask ChatGPT. Any high school question that can be expressed in words it will do pretty well on. At university level it may or may not come back with the right answer, depending on how technical the question is. But it’s still handy at explaining difficult concepts.

In fact its ability to explain is one of its impressive features. When asked a physics question on motion I saw it explain the answer step by step. It was a very helpful tutor. It would be terrifically useful to someone doing an assignment, or sitting an online exam, who didn’t know how to answer the question. Even for students who validly answer the question, platforms such as ChatGPT will quickly become the standard means to check an answer before handing it in.

ChatGPT also has the ability to write computer code. Ask it, in words, to supply code for a particular problem and seconds later you have it. I’ve only tried simple coding exercises so I’m not sure where the limit of its abilities is. But its capabilities will only improve.

How will education look, in the near future, when AI is ubiquitous?

Let’s start with teaching. An AI system will be able to assess a student’s current knowledge and let a teacher know where the deficiencies are. Possibly the AI will offer the student lessons to fill those gaps, bypassing the teacher entirely.

AI will be able to look over every student’s shoulder, keeping an eye on their progress and offering helpful hints along the way.

What about assessment? AI means that students can get help with any question in any place that they have online access.

University of Sydney Business School senior lecturer Mike Seymour, a digital specialist who works extensively with AI, believes educators can work with it. He explains there are ways to genuinely test a student’s mastery of a subject without forcing all assessment back to exam rooms.

One is to get students to present their work to people and be exposed to questions. Another is for teachers to know the work of their students well enough to tell if they have cribbed an essay from a bot.

But there are also ways, says Seymour, to genuinely assess student’s learning in an AI environment.

AI systems are so complex, and have access to so much data, that they can appear to be omniscient. But they’re not. They can only know what they have been trained on. So if students are presented with case studies to analyse with unique data not in the AI universe, or if students are asked to make a judgment about which data is relevant to the task, then the usefulness of the AI system drops away.

“What we look for increasingly with students is the ability to come up with the innovation, that thing which is outside the training space black box,” Seymour says.

Finally there’s the problem that AI, particularly at its early state of development, is plain wrong. In fact ChatGPT warns of this. It may “occasionally generate incorrect information” and “may occasionally produce harmful instructions of biased content”.

Bias is an underlying AI problem. For example, if we ask AI to make judgments of students, those judgments are only as good as the data the system is trained on, which leaves a lot of room for unintended consequences – possibly discrimination on the basis of gender, race, cultural background, financial disadvantage or a host of other factors.

Underlying any analysis of the impact of AI on education is the big question: What jobs are we educating students for in the AI age?

It’s pretty clear that whole stratas of knowledge workers – such as those who gather and analyse information, or who write computer code – are vulnerable to being replaced by AI. But will there be, as in past technological revolutions, enough jobs created in new fields to replace those that are lost? And, if so, what skills will people need to do these jobs?

All of this is unclear, which adds to the feeling that the AI revolution is unlike anything humans have seen before.

A key thing to remember is that AI has limitations. The answers it gives are complex variations of things that already exist. AI literally is incapable of thinking outside the box.

It can’t truly innovate. And in that, there lies hope.