ChatGPT: ‘Opportunity to be human again’, experts say

As ChatGPT enters the mainstream, MaltaToday reached out to four experts in the field of AI to understand what this means for the way we live. The common thread? It gives us an opportunity to adopt more human approaches to work and education by leaving the mundane to the AI

A revolutionary tool

Alexiei Dingli, a professor of AI, noted that ChatGPT is not a fringe tool. With over 1 million users trying the tool in its first week of release, it’s safe to say that the tool has gone mainstream.

“All social media channels are littered with hints and tips on using it effectively. University students up to secondary school students are already using it extensively. The big break will come when Microsoft integrates it with its Office suite. Yesterday Microsoft already launched its Azure GPT services. So it's only a matter of months until everyone starts using it daily.”

Dingli said that there are concerns that many white-collar jobs will become automated, but the plus side is that it would allow people to focus on more meaningful work by automatic any repetitive and time-consuming tasks.

“ChatGPT has the potential to revolutionise education by providing students with a wealth of information and resources at their fingertips,” he said, but added that it could lead to adverse effects such as an over-reliance on these models and a lack of critical thinking skills.

“Many Universities are already panicking about it, and soon, educators will realise that their way of teaching is long defunct and they need to reinvent their approach.”

Bringing education into the digital age

Patrick Camilleri, a senior lecturer at the University of Malta specializing in AI in education, admitted that he feels that he should be ready to embrace this change. “It will probably even replace me in what I think I do best, that is educate.”

He said that the onset of the AI-driven fourth industrial revolution means education as we know it is outdated, and it must undergo a major transformation to reflect the changing times.

One way of doing this is by facilitating a new model of learning that focuses on the qualities that make us human and differentiate us from intelligent machines.

“In the process of employing the qualities of technologies such as AI and IoT, teachers should shift focus and nurture innate deep practices of thinking. In the process they should emphasise the enhancement of decision-making skills, awareness of responsibility, empathy and entrepreneurial skills.”

The idea, for Camilleri, would be to train students to evaluate what machines like ChatGPT offer according to the contexts that they are required in.

“Subsequently the students will learn to adopt, adapt and ultimately employ the qualities of more sophisticated technologies to achieve their goals. They should be prepared to learn to recognize opportunities as they arise and, in the process, choose and employ the proper technology to achieve their goals.”

And same goes for assessment purposes, Camilleri said. “Rather than assessing machine generated content, students may be assessed by having to engage in debates on the choice of procured answers providing explanations that go beyond being descriptive to being objectively analytical and subjectively interpretational.”

“On another note, AI powered technologies can be employed to analyse the capabilities of different students. This will enable the creation of more personal and effective learning paths directed to nurture individual and particular students’ traits, radically shifting education from: a ‘just-in-case’ to a ‘just-in-time’ to a ‘just-for-you’ design.”

How to spot an AI

Albert Gatt, an associate professor in natural language processing, said spotting AI-generated text is not an easy feat.

“This has been known for some time: there are benchmark tests which show that readers are at chance when asked to guess whether a text is human-authored or machine-generated, and this was the case even for models which preceded ChatGPT.”

He remarked that the language generated by ChatGPT is fluent, and the model is able to generate long, coherent text in different styles. He suspects that this is due to a combination of techniques, but most notably through the use of reinforcement learning.

AI-generated text may be hard to spot, but it’s not impossible. Gatt said that some research suggests that the origin of a generated text can be detected by analysing the statistical properties of the language.

“In particular, the 'decoder' part of these models tends to generate by sampling one word at a time, from a probability distribution. The resulting text has subtly different distributional qualities from human text.”

Several tools are being developed to help people identify whether text has been generated by an AI. One of these tools, GPTZero, aims to identify whether someone used ChatGPT to write a piece of text.

“I have yet to see a proper evaluation of this tool. Informal reports in the press have suggested that the success of GPTZero is variable, and in part dependent on the length of the text on which it is run.”

But beyond fluency, Gatt warned that the tool can sometimes generate fluent text that is factually incorrect. This is because, while the model has an impressive stock of knowledge, the knowledge is learned through language without access to the world of perception and experience.

“Without some form of attribution, or other fact-checking mechanism, I think it should be used with caution. Perhaps what will ultimately give the model away is its grasp of the truth.”

Another shakeup for education

Alex Grech, who teaches at the University of Malta’s media department, said that ChatGPT is a game-changer for education, but it follows a longer history of tech tools that have been helping students in their essays and assessments for years.

“I’m old enough to remember the hysteria with Wikipedia, taking laptops into lecture halls and whether students should rely on open education resources - or use calculators. As disruptive technologies go, this is different. Access to a chatbot trained on large language models enables us to use normal language queries and commands to produce seemingly personalised results.”

Meanwhile, education systems worldwide have been caught napping, Grech said, and things are likely to get chaotic in the next months.

Grech added that ChatGPT opens a Pandora’s box since it can’t distinguish between truth and falsehood. It’s output will not always be factual or trustworthy, and any essay answers written by ChatGPT could include inaccurate statements.

On top of this, the chatbot cannot take into account bias. It might make stuff up, write racially-insensitive content or misogynistic things.

“Then again, we know something about chaos in the EdTech world. Just remember the panic in January 2020 when teaching and learning pivoted to emergency online instruction overnight and people started blinking at other people's bedrooms.”

Regardless, Grech said that he will be using ChatGPT with his first-year communication students on project work. “We must adapt our learning environments where we retain the best of older practices while shifting to embrace new technologies - in the same way that calculators and laptops were eventually embraced as a means of augmenting human capacity in different disciplines.”

“In the short term, ChatGPT is likely to shake the trees of assessment and get us to question the value of long-standing, orthodox teaching methods. In the fullness of time, learning how to use AI/LLM technologies to assist writing and facilitate much-needed critical thinking may well become an essential skill set for young people. They may well use it as a shortcut to code and get things done faster. Ironically, it may take a disruptive technological development like the chatbot to force the bastions of education to do some soul-searching and adopt more deeply human approaches to teaching and learning. “

Can ChatGPT take over journalism?

As part of this story, we asked ChatGPT to write an article about ChatGPT using the comments we received from each expert. This article, acting as our ‘control’ experiment, was written by a human journalist in the MaltaToday newsroom. The human-written story was drafted before we submitted the notes into ChatGPT.

The tool did what it does best: condensing information and providing a well-written article. However, it limited itself to a face-value interpretation of the answers at hand. When we read and interpreted the comments provided to us, we found a very different common thread that went beyond the threats of ChatGPT to education. We did what the tool could not do – take the human approach and engage with the content on a more critical level.

This tool could be revolutionary for our press statements and copywriting, but journalists can still provide some value by engaging critically with the world around them.