‘OpenAI’s GPT-3 has powerful implications for the education sector’

The sector must shape the direction of GPT-3’s development so that it becomes a powerful tool to enhance learning – and not just a way for students to cheat, says Roshan Gandhi, director of strategy at City Montessori School

The world has been talking about how the COVID-19 pandemic will revolutionise education.

This may be true, but a recent breakthrough in artificial intelligence could prove just as disruptive and revolutionary to the sector – if not more so.

Introducing GPT-3

GPT-3, an artificially intelligent, natural language-processing application launched recently by the Elon Musk-backed OpenAI, has been received with a flurry of excitement and astonishment across the global tech community. The model builds on its already-impressive predecessor GPT-2, which contained 1.5 billion parameters (connections between nodes in the AI’s ‘neural network’); GPT-3 has 175 billion parameters. It was ‘trained’ on an unfathomable quantity of text data from the internet – to illustrate, all of Wikipedia’s six million English articles comprise just 0.6% of GPT-3’s 45TB training data.

The power of this tool has already proved to be immense. Released in Beta mode as an API available to interested researchers, thousands of individuals across the world rapidly sought access to the platform and started to experiment. The results boggle the mind.

After ‘training’ the software with just a handful of questions or statements, GPT-3 was able to compose creative fiction and Shakespeare-style sonnets of a quality almost indistinguishable from human writing; automatically generate programming code based on short natural-language explanations of app requirements; produce web designs from brief natural-language descriptions of the contents; and manipulate spreadsheets (along with in-sheet AI-driven autocompletion of data) through natural-language commands. Users can enter an instruction such as “Write a poem about [topic X] in the style of [author name]”, and the system spits out a remarkably believable result. One blog presented an informative article about GPT-3, at the end of which the author surprised readers by revealing that the article had been written entirely by the AI. More impressive uses of GPT-3 are being discovered almost daily.

There has been much speculation about GPT-3’s likely impact on jobs, as it appears that the bulk of routine work in text-driven fields like software development and creative writing is now much closer to being automated. But just as far-reaching are GPT-3’s implications for education, and educators ought to make themselves aware of this development. Just as writers, coders, designers and others have begun extensive ideation and experimentation on potential GPT-3 applications, so too should educators.

“Just as writers, coders, designers and others have begun extensive ideation and experimentation on potential GPT-3 applications, so too should educators”

‘Learn philosophy from Aristotle’

Already, a few ideas for educational uses of GPT-3 have emerged. A GPT-3-based language-learning bot corrects your grammatical errors during conversational text-chatting. A GPT-3 quiz generator formulates relevant questions on any topic of choice and accurately marks answers. A GPT-3-driven app called Learn From Anyone prompts the learner to enter the name of any well-known figure to be the ‘teacher’, and to answer queries. Students can thus ‘learn philosophy from Aristotle’ or ‘learn about rockets from Elon Musk’: the system churns out detailed, clear, accurate answers to questions in the style of the chosen teacher, and the student can ask follow-up questions or request alternatives if they’re unsatisfied with the first answer.

Advertisement

Far from trivial, this has been a development of deep consequence for education, almost overnight. A large part of one-to-one tuition is about the learner having the opportunity to ask specific questions privately to a tutor.

If the AI can answer the questions instead, tuition becomes free rather than costly; universally accessible rather than reserved for the privileged; and time-flexible rather than scheduled in hour-long sittings. Homework will also be impacted, as GPT-3 opens up new ways for students to learn through research and discovery. The education community must put its minds together to explore the depth of possibilities offered by GPT-3 and its potential for integration into curriculum and practice.

Preserving academic integrity

Teachers have reason to be concerned that pupils may already be turning in AI-generated essays, passed off as their own work and undetected by plagiarism-checkers: an essay-writing app built on GPT-3 responds convincingly to essay prompts. Julian Togelius, AI researcher and associate professor of computer science and engineering at New York University, has criticised the quality of the nascent GPT-3’s general text output with the suggestion that it “often performs like a clever student who hasn’t done their reading trying to bullshit their way through an exam. Some well-known facts, some half-truths, and some straight lies, strung together in what first looks like a smooth narrative.” This criticism may hold true for GPT-3’s wider applications, but for that clever, lazy student, the same critical comment confirms that GPT-3 would be more than adequate to earn them a passing essay grade.

Again, this has significant implications for education. This is not a prediction of some future AI-driven world – GPT-3 is making this happen right here and now, before our eyes, and educators must adapt urgently. Teachers now need to rethink how they assign writing tasks, expanding their scope, length, and complexity so that they remain ahead of the GPT-3 AI’s capabilities.

Critics are right to point out that GPT-3 is still far from perfect: it sometimes produces incoherent writing, misunderstands instructions, answers questions incorrectly, and makes simple errors that humans would not. It also reflects the bias inherent in the textual data with which it was trained, occasionally spouting racist and sexist statements.

We must remember, however, that it’s still only a provisional product in a beta-testing phase, demonstrating capabilities that its creators did not themselves anticipate. Thousands of testers distributed across the planet are generating data and insights that OpenAI will use to implement major refinements to GPT-3 by the end of 2020. And before we know it, GPT-4 will also be released – with even more billions of parameters, based on even more billions of text inputs.

It’s only a matter of time before GPT-3’s flaws are ironed out and it’s fully ready to revolutionise education and other sectors. Teachers, school leaders, university faculty, education technology firms, and policymakers must come together to brainstorm and shape the direction of GPT-3’s development for education so that it becomes a powerful tool to enhance learning outcomes and not just a facility that enables students to cheat.


You might also like: Unethical use of AI means ‘progress will be stifled by inevitable backlashes’


 

Leave a Reply

Advertisement

Independent Education Live

Five hours of live discussions with influential heads addressing lockdown plans, diversity, nutrition education strategies and cross-sector partnerships.

(LAST FREE TICKETS REMAINING)