178
Austrian Research and Technology Report 2020
Transport, Innovation and Technology (BMVIT),
183
the
experts questioned were only able to give a handful
of examples of AI being used in the public sector.
With regard to the narrower realm of government ad-
ministration, security applications such as pattern
recognition in fraud cases,
image recognition for
criminological analyses or video analyses for security
applications were mentioned relatively frequently.
Some very important AI applications are being antic-
ipated in the medical/healthcare industry at present.
In Austria, too, there are various developments, com-
panies and real-life applications that are relevant
here. Current examples of AI being used in adminis-
tration also include AI in the electronic file (ELAK). As
part of efforts to further develop and ultimately re-
place the ELAK, AI methods are to be used in future
to help users make decisions and choose courses of
action,
save time, and speed up workflows. In partic-
ular, the inbuilt smart search function will use AI to
increase accuracy by making semantic suggestions.
AI is also used for a number of electronic communi-
cation tasks such as automatically identifying send-
ers, automatic keywording, logging and assigning in-
formation. In addition, AI forms part of the official
services provided digitally via the oesterreich.gv.at
platform: its chatbot “Mona” is on hand to provide
administrative
assistance, currently for passport re-
minders and the mobile signature service, and is be-
ing expanded on an on-going basis. The chatbot was
also deployed to the USP company service portal
during the coronavirus crisis, where it served as a
hub for all company-related information throughout
the crisis. The SourcePIN Register Authority has also
already embraced automation solutions (
robotic pro-
cess automation
) and AI elements to improve its ser-
vices by speeding up searches and preparing results/
data for subject specialists. Currently at the planning
183 See Prem and Ruhland (2019).
184 See Prem and Ruhland (2019).
185
Privacy-preserving machine learning.
186 For example, methods like these allow operations to be run and output on database contents without disclosing those contents.
stage, a pilot project run by the Federal Ministry for
Digital and Economic Affairs (BMDW) aims to use AI
to enable companies to receive automatic recom-
mendations for suitable funding. As the prerequisites
for funding can be expressed as logical rules (“if x,
then y”), the goal of this initiative is to convert the
prerequisites for funding into a machine-readable
format.
The analysis commissioned
by the Federal Minis-
try for Transport, Innovation and Technology (BMVIT)
on “AI Potenzial in Österreich” (The Potential for AI in
Austria)
184
concludes that there are currently a num-
ber of major barriers preventing AI from being used
in public administration in Austria. One major obsta-
cle is the fact that, in principle, public authorities are
only allowed to use data for the purpose for which
they were collected. The
public sector thus often
employs rule-based systems that generally do not
learn from personal data. The lack of legal clarity
over the use of AI systems in the public sector also
makes those responsible extremely cautious. A cor-
responding debate on data protection or a broader
debate on data use may be needed in order to create
greater clarity. Another way would be to set up a
public-sector or public-sector-dominated centre of
excellence for AI and data that covers the whole
spectrum of administration-related
AI activities, from
research to implementation and so-called regulatory
sandboxes. As well as legal and technical aspects
and standardisation, this would also, and in particu-
lar, have to deal with topical research issues such as
questions about anonymisation, privacy-preserving
machine learning
185
or
homomorphic encryption
methods.
186
The current government programme ad-
dresses a number of these points.
The reticence being shown towards AI applica-
tions in the core areas of public administration is also
