NISO Plus: Artificial Intelligence and Machine Learning
1.
2. Brian Cody, CEO and co-founder -
@briancody
Modern academic journal publishing platform
used by 900+ journals with three product
options:
◎ Peer review
◎ OA publishing/hosting
◎ Typesetting
3. My lens on ML/AI
◎Developer background, not data scientist
◎Lowering costs through automation w/ ML
○ Example: creating JATS XML for full text + references
◎Vet other products as potential integrations
4. In traditional programming, you create
instructions for a machine to complete a
task. These instructions are often called
algorithms.
1. Rule-based – If this happens, then do
that…
1. Deterministic – Given inputs will yield
predictable outputs (i.e. no randomness
or guesswork).
What is traditional
programming?
5. While it still involves code, machine
learning is a completely different
paradigm.
Whereas traditional programming follows
a series of steps to complete a task,
machine learning uses data to make
predictions.
What is machine
learning?
6. How does machine
learning work?
Of course, this is a gross simplification: Gathering data takes time, choosing the
appropriate algorithm requires extensive testing, and predictions must be used ethically.
Data Algorithm Prediction
6
9. Examples of machine
learning in scholarly
publishing
▪ Binary classification – Will Jill accept my reviewer invitation?
▪ Sentiment analysis – Is this review favorable, neutral, disrespectful?
▪ Natural language processing – Parsing citations into structured data.
▪ Multi-label classification – Classifying articles by discipline.
▪ Network analysis – Recommending related articles to a researcher.
11. Concerns I hear from smaller publishers
◎“How much can we trust it?”
◎Loss of control over decision making
◎Loss of personal relationships
◎Unintended administrative burden
◎Time costs - new products all the time
12. Some of my concerns
◎Reproducing status quo, historical inequities
◎Increasing noise vs signal
◎Expertise needed to interpret results
◎How do we preserve human decision making -
and not defer to convenience?
13. "In all public safety and law enforcement scenarios,
technology like Amazon Rekognition should only
be used to narrow the field of potential matches…
Given the seriousness of public safety use cases,
human judgment is necessary to augment facial
recognition, and facial recognition software should
not be used autonomously."
From https://aws.amazon.com/rekognition/the-facts-on-facial-recognition-with-artificial-intelligence/
15. Questions for the community
◎What is the community role in verifying ethical
uses of ML/AI? (e.g. “ethical AI” badge?)
◎What sort of AI arms race should we expect
within our industry - and what can we do? (e.g.
predatory publishers, gaming the model, etc.)
◎Should the community have their own open
data sets and/or ML models?
Editor's Notes
Brian Cody, CEO and Co-Founder of Scholastica. If you’re not familiar with Scholastica, we are a web-based journal management platform with tools and services for peer review, publishing, hosting, and typesetting. Before starting Scholastica, Brian was doing doctoral work in sociology at the University of Chicago, and he is a self-taught Ruby on Rails programmer.
Ownership, copyright, privacy
Perturbed authors
Rules: deterministic vs predictions: probabilistic