On the 9th of November, 2017, we received the following press release: A startup is releasing software that it claims will create articles using data from researchers.
According to the business that invented it, sciNote LLC, the programme, named “Manuscript Writer,” employs artificial intelligence (AI) to compose documents. According to a spokeswoman, the programme provides a first draught that the scientist should review, but it does not write the Discussion, which is “the most creative and original part of the scientific article.” Can it, however, generate any coherent text?
According to a sciNote press release, Manuscript Writer (a supplement to the company’s Electronic Lab Notebook, or ELN):
…by utilising technical developments in machine learning and artificial intelligence, has the potential to considerably ease the process of preparing scientific articles. Recognising the significance of rapid publication of scientific findings by the global scientific community, the add-on seeks to drastically minimise the time required to create initial content. Using data from the ELN and references from open access publications to create a structured draught for the author to modify and improve further.
According to a corporate spokeswoman, the company’s ELN is the first to generate scientific publications.
What about preventing the problems that paper mills occasionally face, such as plagiarism? We turned to Charles Seife of New York University, who has investigated paper mills in the past. Seife claimed he couldn’t comment on Manuscript Writer because he doesn’t have it on his PC, but it appears “dodgy:”
I could easily envision a useful system that would take lab notes and try to fit data, protocols, and notes into various templates; format references; and even construct an outline of what information goes where in a report. However, this method promises more, implying that the software will deliver a “first draught.” This seems to me that, unless the scientist has already entered a significant amount of writing, the programme will get it from someplace else… which is troublesome, to say the least.
According to the terms of service, the draught will be generated not just from the data submitted by the user, but also from “relevant keywords and open access references.” Obviously, an AI isn’t capable of interpreting and digesting language in the same way that humans can, so I’m not sure how it’ll be able to develop any form of derivative work based on open-access references that isn’t plagiaristic or incoherent (or, more likely, both.)
Seife went on to say:
So, I can’t say for sure because I haven’t tried the programme, but I’d be prepared to bet money that it’s grabbing prose from the introductions of references, jumbling it up in some way, and plunking it down for the researcher to use in his own introduction. For obvious reasons, this is not a good idea.
So, yeah, I am concerned about this business model. It is a negative thing if it helps automate the process of taking other people’s prose, lightly modifying it, and passing it off as one’s own.
We expressed Seife’s concerns to the representative, who responded:
Manuscript Writer will create the manuscript’s materials and methods section based on the scientist’s project and experiment data, protocols, and notes in sciNote…Furthermore, Manuscript Writer will produce an introduction based on relevant keywords and DOI numbers given by the scientist. Manuscript Writer will extract material from selected references and, based on the appropriate keywords, will hunt for other relevant open access references to include in the draught. The scientists will receive an introduction with citations for each phrase or paragraph, and all references will be added to the list of references (another section of the paper prepared by paper Writer).
The programme checks for plagiarism, according to the spokesperson:
Following each paragraph in the introduction, the scientist sees the number of the reference and a percentage (e.g., 100%), indicating that a certain paragraph is cited from the designated reference and is 100% the same text. This information cannot be disregarded because it is part of the text and also informs the scientist that she/he should alter it…It is then their obligation to modify and proofread the writing. As it would be in any other scenario when composing manuscripts.
When the scientist receives the draught, we also alert them to amend the received text. The key advantage is that Manuscript Writer can incorporate engaging paragraphs on the issue at hand in the opening, giving the scientist a head start while writing.
Read full article here