ChatGPT and AI language tools banned by AI conference for writing articles

One of the world’s most prestigious machine learning conferences has banned authors from using AI tools like ChatGPT to write scientific papers, sparking a debate about the role of AI-generated text in the medium university.

The International Conference on Machine Learning (ICML) announced the policy earlier this week, saying, “Articles that include text generated from a large-scale language model (LLM) such as ChatGPT are prohibited from unless the text produced is presented as part of the experimental analysis of the article. The news triggered widespread discussion on social media, with academics and AI researchers defending and criticizing the policy. The conference organizers responded by publish a longer statement explaining their thinking. (The ICML responded to requests for The edge for comment by directing us to that same statement.)

According to the ICML, the rise of publicly available AI language models like ChatGPT – a general-purpose AI chatbot that launched on the web last November — represents an “exciting” development that nevertheless comes with “unforeseen consequences”. [and] unanswered questions.” The ICML says this includes questions about who owns the production of these systems (they are trained on public data, which is usually collected without consent and sometimes regurgitates this information verbatim) and whether AI-generated text and images should be “considered new or merely derivatives of existing work “.

Are AI writing tools just assistants or something more?

This last question ties into a delicate debate over authorship – i.e. who “writes” AI-generated text: the machine or its human controller? This is especially important given that the ICML only prohibits “wholly produced” AI text. Conference organizers say do not prohibiting the use of tools like ChatGPT “to edit or polish text written by the author” and note that many authors were already using “semi-automated editing tools” like Grammarly grammar-checking software for this purpose .

“I’m sure these questions, and many more, will be answered over time as these large-scale generative models become more widely adopted. However, we do not yet have clear answers to any of these questions,” the conference organizers write.

As a result, the ICML says its ban on AI-generated text will be reassessed next year.

However, the issues addressed by the ICML may not be easily resolved. The availability of AI tools like ChatGPT is confusing for many organizations, some of which have responded with their own bans. Last year, coding Q&A site Stack Overflow prohibits users from submitting replies created with ChatGPTwhile the New York City Department of Education blocked access to the tool for anyone on their network this week.

AI language models are auto-complete tools with no inherent sense of factuality

In each case, there are different fears about the adverse effects of AI-generated text. One of the most common is that the output from these systems is simply unreliable. These AI tools are vast auto-completion systems, trained to predict which word follows the next in a given sentence. As such, they don’t have a hard-coded database of “facts” to rely on – just the ability to write plausible-sounding statements. This means they tend to present false information as the truth since if a given sentence sounds plausible does not guarantee its factuality.

In the case of the ICML’s ban on AI-generated text, another potential challenge is to distinguish between writing that has only been “polished” or “edited” by the AI ​​and the one that was “entirely produced” by these tools. When do a number of small AI-guided fixes constitute a larger rewrite? What if a user asks an AI tool to summarize their article in a catchy summary? Does this count as freshly generated text (because the text is new) or just polish (because it’s a summary of the words the author wrote)?

Before the ICML clarified its policy remit, many researchers feared that a potential ban on AI-generated text would also harm those who do not speak or write English as their first language. Professor Yoav Goldberg of Bar-Ilan University in Israel said The edge that a general ban on the use of AI writing tools would be an act of control against these communities.

“There is a clear unconscious bias when judging articles in peer review to prefer more fluent ones, and this works in favor of native speakers,” says Goldberg. “By using tools like ChatGPT to help express their ideas, it seems many non-native speakers think they can ‘level the playing field’ around these issues.” Such tools can help researchers save time, Goldberg said, as well as better communicate with their peers.

But AI writing tools are also qualitatively different from simpler software like Grammarly. Deb Raji, an artificial intelligence researcher at the Mozilla Foundation, said The edge that it made sense for the ICML to introduce a policy specifically aimed at these systems. Like Goldberg, she said she’s heard from non-native English speakers that such tools can be “incredibly useful” for writing articles, and added that language models have the potential to make more drastic changes to text.

“I see LLMs as quite distinct from something like autocorrect or grammar, which are remedial and educational tools,” Raji said. “While they can be used for this purpose, LLMs are not explicitly designed to adjust the structure and language of already written text – they also have other more problematic capabilities, such as generating new text and spam.”

“At the end of the day, the authors sign the document and have a reputation to uphold.”

Goldberg said while he thinks it’s certainly possible for academics to generate papers entirely using AI, “there’s very little incentive for them to do so.”

“At the end of the day, the authors sign the document and they have a reputation to uphold,” he said. Even if the bogus article somehow passes through peer review, any incorrect statements will be associated with the author and will remain with him for his entire career.

This is especially important given that there is no completely reliable way to detect AI-generated text. Even the ICML notes that foolproof detection is “difficult” and the conference will not proactively enforce its ban by running submissions through detection software. Instead, it will only investigate submissions that have been flagged by other academics as suspicious.

In other words: in response to the rise of disruptive and new technologies, organizers are relying on traditional social mechanisms to enforce academic standards. AI can be used to polish, edit or write text, but it will always be up to humans to assess its value.