Science

Half of students are using ChatGPT to cheat, and it could rise to 90%


Half of college students are likely already using ChatGPT to cheat, experts have estimated.

They warn the revolutionary AI has created a cheating epidemic that poses a huge threat to the integrity of academia.

Rehan Haque, of artificial intelligence company Metatalent.ai, said: ‘We’re already at the point where AI can write entire projects, and then a different AI tool can reword it to make AI undetectable.

‘At present, well over half of students are likely using AI tools to cheat the education system in exams or essays, but it wouldn’t surprise me if that number were already higher.’

Could educators resort to written tests to deal with AI cheating?

He added: ‘If educators make the mistake of ignoring the threat of AI-based cheating, I can honestly see more than 90 percent of students cheating in this way [in future].’

OpenAI’s new GPT-4 update (GPT-3 and GPT-4 are the models which underlie ChatGPT) is able to get 90 percent on a huge number of exams, including the American bar exam.

The AI bot is also capable of writing human-like essays on any subject in seconds, in response to simple text prompts.

Creator OpenAI is working on a tool to detect AI-written content, but warns it’s not 100 percent accurate.  

A survey by Study.com of 203 teachers found that 26 percent of K-12 teachers had already caught at least one student cheating using the software.

South Texas College of Law Houston law professor Josh Blackman wrote: ‘This technology should strike fear in all academics.’

The Los Angeles Unified School District, Seattle Public Schools, the New York Department of Education and Oakland Unified are among the American school boards which have banned or blocked the use of ChatGPT.

At university level, New York’s Yeshiva College updated its cheating policy to include ‘something/someone else’s language’ so that cheating via ChatGPT was banned.

Haque says that educators may resort to ‘technological regression’ as a temporary measure to battle AI cheating.

He said: ‘This might even mean returning to the old-fashioned way of writing everything down. But, even then — what stops a student from copying an AI-produced essay off a screen?’

Rehan Haque of Metatalent.ai believes the problem is widespread - and will get worse

Narmeen Makhani, executive director of the ETS AI Labs

Rehan Haque of Metatalent.ai believes the problem is widespread – and will get worse. Narmeen Makhani, executive director of the ETS AI Labs

Haque believes that AI detection software will evolve, but that educators need to rise to using AI in the classroom — rather than banning it. 

Research by GoStudent found that 59 percent of young people want to see advanced tech such as AI on the curriculum.

Haque said: ‘To be fair to the students, if education systems fail to adapt to a technology Bill Gates recently described as “the most revolutionary technology in decades”, then the onus can’t sit exclusively with them.

‘Instead, teachers, examiners and educators need to recognize that technology is changing, and they need to move with it.’

AI has positive uses within the classroom, agrees Narmeen Makhani, executive director of the ETS AI Labs, which makes tools which can detect AI cheating.

She said: ‘Generative AI is also used for good purposes like advancing learning. Generative AI tools like ChatGPT help learners more efficiently generate text and summarize information that can help them with schoolwork.’

Makhani said that scientists in her lab have been working on strategies to detect AI – including keystroke analysis to spot people editing essays generated by AI.

She said that there are giveaway signs that essays have been generated by AI rather than a human being.

She said: ‘Although, it’s very hard for humans to detect AI-generated text, Generative AI detection tools look for minute patterns such as repetitive simple words or the lack of any inconsistencies to detect patterns typical of AI generated text.

‘In addition, AI models tend to write consistent sentences as opposed to humans who typically write in bursts. AI detection models also look for whether a piece of text ‘appears’ random or chaotic compared to text that appears non-chaotic or less ‘perplexing’ which is most likely AI generated.’

Makhani says that no tool can accurately detect AI-generated text 100 percent of the time, but she and her team will continue to investigate how to spot cheaters.

She says that the technology offers educators a way to free the education system from focusing on memorizing facts and regurgitating large amounts of text.

Makhani says, ‘Educators have an opportunity to use AI tools to augment learning and improve assessments to enable them to rely less on rote memorization and text generation but instead on asking learners to demonstrate their understanding of concepts and their ability to analyze information.

She also believes that AI tools such as ChatGPT will be useful in the classroom – and in teaching students skills which will be useful in the workplace of tomorrow.

Makhani said: ‘They can also empower students to leverage AI tools to evaluate AI outputs to both support and demonstrate higher-order skills. Generative AI tools can help educators quickly create personalized content that is socio-culturally relevant for their students and can help to motivate and assess learners in new ways.’



READ SOURCE

This website uses cookies. By continuing to use this site, you accept our use of cookies.