UNSILO AI in academic publishing survey released

(1 Nov 2019)  Today, UNSILO has released the results of a survey on AI usage in academic publishing. How is it used? What factors impede its adoption? More than 3,000 new academic journal articles are published every day, and yet many of the tools for processing content through the
academic workflow are frequently very manually based. The UNSILO survey, with publishers
and other stakeholders in the scholarly workflow, was done between June and September
2019. The results were presented at a panel session at the Frankfurt Book Fair, chaired by Toby
Green, former head of publishing at OECD.

Take-up of AI

The findings show a steady take-up of AI tools by publishers. Two-thirds of respondents are
currently using at least one AI tool, and only 3% of respondents felt that AI could not benefit
their activities in some way. Forty-five percent of respondents not currently using AI plan to
introduce some AI tools within the coming twelve months.

As for the perennial ‘build or buy’ question, around one-third of publishers use only their own
in-house resources. The remainder use external suppliers or a mixture of the two. How the
publishers plan to expand their AI capability is interesting. The largest response was to
expand the publisher’s internal AI team, which suggests an emphasis on keeping skills
internal to the organisation, rather than buying in external tools.

For publishers currently using AI tools, the biggest justification provided for AI tools was
saving time (65%), suggesting that the current implementation of AI tools is based very much
around tools to improve specific pain points in the academic workflow. By far the biggest
application is to provide text analytics tools (over 40% of all implementations). The most
widely used AI tools are machine learning and NLP (natural-language processing), with
rule-based tools close behind. Linked data is used by less than 10% of respondents, and open
linked data by even fewer. The largest single use of AI tools is to add and to curate metadata.
Remarkably, over 40% of metadata is added by in-house staff, which suggests there is ample
scope for automation.

Trust, bias and accuracy

For all the coverage in the media about questions of bias and privacy being topics of major
concern, few of the respondents seemed to be taking action about these issues. Fewer than
10% of publishers check their AI tools for bias, and only 8% for privacy and compliance with
GDPR. There is a paradox here; the two biggest reasons cited for not using AI tools more were
“not enough time” and “uncertain quality of results”. But fewer than 20% of respondents
stated they were checking the AI tools they use for reliability and consistency.

Thomas Laursen, Chief Executive Officer at UNSILO, stated: “This survey confirms our
experience with several academic publishers, who are both keen to get involved with AI and
yet at the same time reluctant to relinquish human control over the academic workflow. We
UNSILO, Inge Lehmanns Gade 10, 8000 Aarhus C, Denmark www.unsilo.ai hope this survey will encourage more publishers to take the plunge with this transformational technology. It will be interesting to compare these results with the situation in a year’s time, as more publishers learn from their experience and provide feedback.”