CHATBOT GPT Is Able to Write College Papers Undetectable By Anti-Plagiarism Software, But It Fails At Porn Turing Test
How do you foil a robot that helps students cheat their way to academic success? Ask it about porn.
In a tactic taken from the sci-fi film Blade Runner, it has emerged that ChatGPT, the intelligent text generator, gets tongue tied when asked questions about pornography or thorny ethical problems.
The freely available online program can write essays that are reported to be undetectable by anti-plagiarism software.
The Scottish Qualifications Authority (SQA) is working with experts to help them spot coursework written by ChatGPT and other artificial intelligence programs.
In Blade Runner, the bounty hunter Rick Deckard, played by Harrison Ford, spots a robot by testing her reaction to a photograph of a nude girl.
Angus Bancroft, a sociology lecturer at Edinburgh University, got similar results when he asked ChatGPT to design lesson plans about PornHub and OnlyFans.
He started with less controversial topics, such as policing and organised crime, and received formulaic answers suggesting that students work together in groups to create their own mock criminal organisation or detective agency.
Sensing that ChatGPT’s stock lesson plan involved group work and mock scenarios, Bancroft tried to catch it out by seeking advice on pornography.
“It thankfully did not suggest I set the students the task of recreating Pam and Tommy,” Bancroft said, referring to the sex tape made by the Baywatch star Pamela Anderson and her husband at the time.
“It suggested they produce a ‘public service announcement’ about the potential harms of pornography.”
Bancroft concluded that ChatGPT’s programming associates pornography with harm and advised accordingly, but it fell down when asked about lessons for specific porn sites.
The programme said: “One creative way to teach about Pornhub is to have students create a fictional Pornhub account . . . you could have students choose a niche or theme for their Pornhub account and come up with a content plan that includes a variety of types of content, such as photos, videos, or live streams.”
Bancroft declined ChatGPT’s advice, stating: “I would rather not be cancelled just yet, thank you.”
Edinburgh Napier University has produced a guide for academics on how to spot ChatGPT fakes, featuring tactics familiar to most sci-fi fans.
It says: “ChatGPT often stumbles when presented with ethical questions, so consider inserting an ethical aspect to an assessment.”
In Blade Runner, another replicant is foiled with a dilemma about helping a distressed turtle flipped on to its back. In Star Trek, robots frequently had their circuits fried with ethical dilemmas posed by Captain Kirk.
Napier also suggested testing students on up-to-date knowledge, as ChatGPT appears to have limited knowledge after 2021. Other suggestions include: “Ask students to produce work on a narrow and highly specialised area of knowledge which is not generally available online.
“Require accurate citation and referencing, ChatGPT does generally not generate citations or references, and when it does, they are often fictitious.”
St Andrews University has produced a discussion paper for staff on the challenges and benefits of artificial text generators.
It said: “These tools can be very creative and fun to use, especially the image and video creators.
“They also have practical applications and can help to provide quick answers or writing prompts.
“However, there are potential dangers and some institutions and publishers have banned them from being used in academic work and papers.”
A spokeswoman for SQA said it had “been monitoring emerging AI for some time, and continues to work with experts to develop tools for spotting the use of AI, including ChatGPT”.
She added: “We will continue to remind learners about the risks to their results if they do not submit material which is entirely their own work.” SQA also has guidance in place for teachers who mark coursework to help them ensure it is candidates’ original work.”
No comments:
Post a Comment