Academia needs to set standards for the use of artificial intelligence in research, open science expert says
Standards for the use of artificial intelligence in research should be set by the academic community, and there should be more scrutiny of how scientific papers are “feeding the machines” of big AI companies, according to one open science researcher.
Tony Ross Hellauer told a European Universities Association webinar this week that the academic community should work together to create shared resources on the role of AI in research, and that the academic community should set standards for the use of these technologies.
While a centralized high-level body may set general principles for use, “for something as diverse as science, centralization is not the way to go,” said Ross Hellauer, leader of the Open and Reproducible Research Group and a senior researcher at the Know Center, an Austria-based AI and data science research center.
Instead, he added that some kind of academic “pluralism” needs to set the standard, arguing that academic values, especially those of open science, are needed to foster trustworthy AI-driven research.
Exploitation by AI companies
Ross Hellauer also raised concerns that some in academia are questioning the relationship between open science and AI, given that AI models can scrape academic papers and exploit knowledge and writing styles.
“If you think about what they do, [AI firms] “If we're giving back compared to what they're getting, I think we as a research community have a right to be a little…perplexed.” “It makes some people wonder, whose benefit is it that we share everything, is it always for the public good?” he continued.
Ross Hellauer added that a key value of open science is fairness, and that in this case, open science “really just feeds the machine, and the big companies get bigger.” It is therefore “in the interests of open science” to think carefully about the field, and open science has reached a “reflexive, perhaps critical point.”
Collaboration between open science and AI
Ross-Hellauer highlighted the 2019 European Commission report that defined seven principles for trustworthy AI. He then compared these to UNESCO's recommendations on open science, which emphasize many of the same values, including transparency, quality, co-benefits, equity and inclusiveness.
“So I really think trustworthy AI and open science are a perfect match,” Ross Hellauer said. “So the question is, how do we make scientific AI trustworthy?”
Next, Ross Hellauer outlined his vision for integrating AI into a shared academic commons. He recommended prioritizing open source AI tools and models so that researchers know exactly which generative AI models produced each set of results, which is important for reproducibility.
Ross Hellauer also emphasized the need for open guidance on the strengths and limitations of AI models, as well as safeguards such as testing tools to reduce risks such as bias and error, monitoring the impact of AI on open science, and critically evaluating the resource strength of AI systems.
“I think these tools have great potential to help document and automate research,” he said. “Open science relies heavily on good documentation, but there is clearly a risk of administrative overload. [AI has] There is a possibility here. ”
