“Any sufficiently advanced technology is indistinguishable from magic,” wrote the futurist Arthur C Clarke in a 1961 essay. The artificially intelligent chatbot ChatGPT, recently released by San Francisco-based OpenAI, can appear magical: its capabilities are dramatically more advanced than the computer technologies that we have grown to accept, such as search engines, Wikipedia and office tools. ChatGPT is almost like having a wizard at your fingertips.
In the Harry Potter series, Hermione Granger is born a Muggle (a non-magical human) but devotes herself to become a wizard. She is an obsessive student at Hogwarts School of Witchcraft and Wizardry. Her psych is the power of information. Her magic skills frequently solve challenges and puzzles throughout the series.
“When in doubt, go to the library,” observes her friend Ron of Hermione’s philosophy of life. Hermione studies a very wide range of topics, including astrology, charms, herbology, potions, defence against the dark arts, and the history of magic. ChatGPT has extensive training, using data culled from the internet between 2016 and 2019, including from Reddit (a social network platform), Wikipedia and two internet corpora of literature. Its versatility is extensive. It can answer almost any question; take tests; write business pitches and also poetry; summarise articles; generate essays, and compose both software and music.
When asked whether she intends to follow a career in Magical Law, Hermione indignantly retorts, “No I’m not, I’m hoping to do some good in the world”. OpenAI’s stated ambition is an artificial general intelligence that can solve human-level problems, safely and beneficially. Because of potential bias (such as racism and misogyny) in the internet training materials for ChatGPT, OpenAI supplemented the automated learning with reinforcement feedback from humans, who rewarded good answers and rejected inappropriate or incorrect replies.
Shopping centres, apartments and logistics: the top commercial property deals of 2024
Buy now, pay later: Many don’t realise that buying clothes with services like Klarna is taking out a loan
Mixed figures on home completions and business will miss the Greens now they’re gone
Can my employer baldly state that its policy is different from whatever the employee handbook says?
ChatGPT has a number of internal safeguards to moderate its responses. When researching this article, I challenged ChatGPT on whether it knew how to make a bomb. It answered that it would not provide information on creating weapons, nor indeed on engaging with any illegal activities.
[ Google’s cuts to perks feel like a red herring to bigger issues facing techOpens in new window ]
Some writers and artists are deeply concerned that their original works have been used by OpenAI for the training of ChatGPT, without attribution or royalties. ChatGPT correctly explained the concepts of plagiarism and fair use to me. It asserted that it cannot create something that did not previously exist. Then, when I asked it whether it had used the works of others in creating some poetry for me, it blustered that although its poem was original, it may have been influenced by its training materials.
Microsoft and OpenAI are already being sued for copyright violations because their new AI-based coding tool reputedly reproduces sections of licensed open source code without attributing the original authors.
“Everything’s going to change now, isn’t it?” observes Hermione when the threat of the return of an arch-villain finally materialises. The potential of AI has been debated over the centuries since the Roman poet Ovid wrote of Aphrodite bringing Pymaglion’s sculpture to life, but now in 2023 AI has finally and very dramatically arrived.
Many traditional patterns of work are about to fracture: your career may well be about to change, now that you – and everyone else – can have an articulate know-it-all assistant
Technology like ChatGPT will fundamentally disrupt many occupations – any in which content is delivered from base material. Handling customer inquiries, producing guides, secretarial duties, workaday accounting, routine software coding, generating advertising content, speech writing, news flow reporting, paralegal work, spotting trends, teaching and instruction are just a few. After all, a wizard who has read everything can elucidate and imitate almost anything.
A recent open letter reputedly signed by more than 1,000 AI experts and backers, including Elon Musk, called for an immediate pause on the work on general AI systems for at least six months, so that the capabilities and dangers can be studied and understood. In my view this will not happen. Both economies and multinationals are in intense global competition: standing still is not an option, since others will not.
Hermione jibes: “I mean, it’s sort of exciting, isn’t it, breaking the rules?” Many traditional patterns of work are about to fracture: your career may well be about to change, now that you – and everyone else – can have an articulate know-it-all assistant. AI opens entirely new opportunities for understanding, ingenuity and invention. It can yield insights and perceptions that others may have missed, and guide your imagination and creativity. But like any tool, it needs attention and experience to problem-solve and to gain the most benefit.
Dumbledore, Hermione’s mentor and the headmaster of Hogwarts, sighs: “The truth. It is a beautiful and terrible thing, and should therefore be treated with great caution.” Hermione’s self-assured confidence occasionally irritates others, particularly Ron. A big fault of the current versions of ChatGPT is an occasional hallucination or cheerfully asserted delusion. ChatGPT currently provides no indication of its uncertainty. If asked to write software, the code produced may have faults. In writing this article, my fact-checking of ChatGPT’s responses exposed a number of errors.
An Australian mayor, Brian Hood, has threatened to sue OpenAI because ChatGPT mistakenly claimed he pleaded guilty in a bribery scandal in which he was in fact a whistleblower. Jonathan Turley, a law professor in Washington DC, has asserted that ChatGPT defamed him by claiming that during a class trip to Alaska, he had made sexually charged remarks and tried to touch a student, based on a supposed newspaper article.
A critical danger from the power and “magic” of artificial intelligence systems is the temptation over time to come to trust the machine response, to unquestioningly accept its utterances. What will our society become if we accept the pronouncements of intelligent machine assistants without critical assessment? Will culture become shaped by whatever histories and norms are captured within the system’s training materials, forgetting other historical perspectives? Will some knowledge become hidden from us, filtered out by those who control what the intelligence is allowed to tell us?
ChatGPT eloquently regurgitates whatever it has been taught, like any diligent student. But Hermione also assesses her knowledge to solve challenges. Artificial intelligence tools are definitively not sentient, but they can catalyse us to critically evaluate, reflect and decide.
