Editor's note: This article is by Léa Deleris, manager of Risk Management at IBM Research-Ireland. Additional contributions made by Charles Jochim, research staff member at IBM Research-Ireland.
Alice: `Would you tell me, please, which way I ought to go from
Cheshire Cat: `That depends a good deal on where you want to get to.'
Alice: `I don't much care where—‘
Cheshire Cat: `Then it doesn't matter which way you go.’
Alice: ‘So long as I get somewhere.’
Cheshire Cat: `Oh, you're sure to do
that, if you only walk long enough.'
dilemma, and Stanford’s Decision Analysis Professor Dr. Rob Howard, who presented
the Alice in Wonderland
concept of indifference in a lecture, inspired me to study how mathematics can
apply to the risk, uncertainty, and personal preferences that influence the
decisions we make every day, about everything. What Alice and the Cheshire Cat
so eloquently illustrate is that preferences are not as obvious as they may
seem. I wanted to know, as a PhD student sitting among my fellow classmates,
could natural language processing and cognitive computing be applied to web
applications that could in turn, help us make more logical decisions?
forward to today, and from student to IBM research scientist, and I’m applying
artificial intelligence to our human intelligence – and how we can debate with
machines to help us make good decisions when facing uncertainty. Here’s how I
explained it to a TEDxParis
audience in February.
my team in Dublin is using machines to help medical professionals make more
rational decisions. The tool we developed, called MedicalRecap,
extracts information from PubMed’s 24
million online citations to create a risk model for doctors.
MedicalRecap’s semantic module allows doctors to
cluster the extracted terms (variables of the risk model) by
grouping similar or related terms into concepts. It also has an aggregation
module, which allows the user to combine the extracted dependence and probability
statements into a dependence graph, also known as a Bayesian network.
an instance of a doctor needing to understand the role of tea and coffee
consumption on the incidence of endometrial cancer. Currently, doctors would
address this task manually by searching for relevant papers, reading them,
taking notes (by hand or copy-pasting on a spreadsheet), and aggregating this
instead, presents extracted and aggregated data in an intuitive graphical
format, providing a way for the user to trace back through the summarised
information, to the original input. The tool also allows users to edit the
output of the algorithms if they encounter an error, which is fed back into the
system to improve its knowledge and performance over time.
also relies on the doctor's expertise, so ideally, errors are reduced by
combining the doctor's knowledge with the inferences the tool makes in finding
dependency relationships. The assumption is
that the doctor does not have all the input required, but is exploring the
space. The tool helps the practitioner look for answers, but does not provide
them. Ideally, it will reach the same conclusions that the doctor already has
made so that he or she will trust the system more.
We also want MedicalRecap to provide evidence for new conclusions to be
drawn. For example, if the doctor sees that coffee consumption is linked to
some cancers, which she already knew, the tool could show that in fact this is
primarily for certain populations, which she didn't know.
As similar as it
may sound, MedicalRecap is different to IBM Watson Health.
Our tool is a web-based GUI focused only on published medical literature and is
not designed for personalized medicine, but instead to make more global inferences
between diseases and related risk factors. But like Watson, MedicalRecap’s
Extractor, Clusterer, and Aggregator services are available on IBM’s SoftLayer cloud infrastructure
as a service.
Our risk information extraction models, like in MedicalRecap, can be applied
to other domains. In the future, oil and gas experts could use the tool to extract
information from academic papers about factors influencing reservoir capacity
and shape. As long as
we have the main ingredient of a large body of literature related to a profession
or domain, our decision support system tool might even be able to offer
Alice somewhere to go, no matter how unsure she is.