AI can now produce texts almost no different from humans.
These algorithms made tremendous progress in 2021: they are called large language models (LLM). Meta (Facebook) just announced it created a massive new LLM, but is giving it away for free! (read below). But what if LLM started to write science articles? The very foundation of the centennial and global generation of knowledge would be shaked!
As Nature sums it up, LLM “can churn out astonishingly convincing prose, translate between languages, answer questions and even produce code”. But there are caveats: LLM “have deep flaws, parroting misinformation, prejudice, toxic language”, or problematic stereotypes in the zillions of documents they’re trained on. “And researchers worry that streams of apparently authoritative computer-generated language that’s indistinguishable from human writing could cause distrust and confusion.”
Shobita Parthasarathy, a specialist in the governance of emerging technologies at the University of Michigan in Ann Arbor, is one of them. She thinks, among others, that “some people will use LLM to generate fake or near-fake papers, if it is easy and they think that it will help their career”. More generally, “the algorithmic [texts] could make errors, include outdated information or remove nuance and uncertainty, without users appreciating this. If anyone can use LLM to make complex research comprehensible, but they risk getting a simplified, idealised view of science that’s at odds with the messy reality, that could threaten professionalism and authority. It might also exacerbate problems of public trust in science.”
Joelle Pineau, managing director at Meta AI, who created the newly announced LLM, answers to such critics, in a MIT Tech Review article (read below): “There were a lot of conversations about how to do that in a way that lets us sleep at night, knowing that there’s a non-zero risk in terms of reputation, a non-zero risk in terms of harm.” She dismisses the idea that you should not release a model because it’s too dangerous. “I understand the weaknesses of these models, but that’s not a research mindset,” she adds, concluding that “the only way to build trust is extreme transparency”.
For Parthasarathy, that is not enough. The scientist pleads along with her colleagues in a report for bodies to step in with general regulation of LLM: “It’s fascinating to me that hardly any AI tools have been put through systematic regulations or standard-maintaining mechanisms.”
The goal is simple, when talking about the possibility of writing science articles with such algorithms, and hence the production of new knowledge which might be hard to sort from fake content: to make sure that one of the most interesting and powerful achievements of science, LLM, does not become one of its most furtive enemies.
- Olivier Dessibourg, GESDA
|