Player is loading...
AI in the federal administration and public trust: the role of the Competence Network for AI
Dr Kerstin Johansson Baker, Head of CNAI Unit, Swiss Federal Statistical Office
Wednesday, Feb. 21, 2024 · 9:30 a.m. · 15m 38s
Artificial intelligence (AI) and other new technologies have the potential to fundamentally change all areas of life including the Federal Administration itself. The Competence Network for Artificial Intelligence (CNAI) of the Swiss Federal administration facilitates the sharing of knowledge and access to the competence of numerous experts in the field of AI. The CNAI also fosters public trust by transparently communicating about ongoing AI projects within the Federal Administration.
Embed
Transcriptions
Note: this content has been automatically generated.
00:00:00
thank you i hope you can hear me it to just to be clear ideas subplot to federal administration
00:00:06
but i i'm not gonna speak today about the federal
00:00:10
administration as a regulator gonna speak by the federal administration
00:00:14
as an organisation that uses a i and the our some challenges did too and um
00:00:23
you you are in the fast moving world of set to go away i
00:00:28
um i i assume a day's suspect
00:00:33
things go let's quickly within the federal administration
00:00:38
but they have a they have been a few things going on and i just thought i'd give you these point is at least
00:00:45
you've heard about but it was a reporting twenty nineteen which seems really will now
00:00:50
when you think of it but uh that was produced within the federal administration about
00:00:54
the sort of the state of a. r. i. in switzerland not only within the
00:00:59
public sector but everywhere what needs to be done with several action points for different uh
00:01:05
uh domains et cetera so a few things were done
00:01:09
internally and within the administration one of them uh was
00:01:14
to create a competence centre for data signs that said
00:01:17
this reschedule statistical office to having truman competence put it aside
00:01:23
and a i as a tool as well of course uh
00:01:26
guidelines wait sure it's a maybe a city they are um
00:01:31
but not mandatory they um i capital but we'll pass your but they do
00:01:35
exist and it's for a human centred use of a i kind of the tools
00:01:42
and uh uh uh what's interesting more interesting directly interesting to me is the
00:01:46
creation of the i. c. n. a. i. in twenty twenty one and uh uh
00:01:53
maybe you know all will have heard about it in the media or not but didn't for november
00:01:59
last year so there is now an on an on got
00:02:02
this ongoing work to look at what kind of regulatory approaches would
00:02:08
be suitable for switzerland yeah when it comes to a i looking
00:02:11
at europe as well et cetera so the whole analysis is ongoing
00:02:15
nobody knows which directions gonna go and the results of this
00:02:21
this examination of possibilities uh will come at the end of the year
00:02:26
internally as well so does not everybody but there is a a working
00:02:30
group when they are within the administration it's trends departmental and they're looking
00:02:37
at more directly what needs to be done for the who does what's
00:02:40
the use of a i a government that that you're within the administration
00:02:45
and uh we actually use the web site of the sunni i to publish some uh um
00:02:50
something's produced by this group such as a fact sheet on the
00:02:55
use of jenny i wouldn't get ministrations or you might have seen
00:02:59
in the news or not but i ongoing work we're back to this you know yeah i saw that was with the um
00:03:06
there's federal statistical office got division mandate in twenty twenty one to create this network
00:03:13
uh you go running review more more in the first part of twenty two
00:03:19
it's in network it was a conscience choice can because uh
00:03:25
the report showed that there was some may but nobody wanted to
00:03:30
create a pray for a big central institution and they also agreed on the fact that
00:03:37
you couldn't do nothing so this was way in
00:03:40
the middle and network was a set up and that
00:03:45
so oh it's it is it has to be a touch somewhere
00:03:49
administratively so it belongs to the swiss pedals that it's got office
00:03:53
but it's meant as an enabler and that's a lot tighter and work to put
00:03:57
together compact competencies that exist across the administration
00:04:01
but we do provide some based services such
00:04:05
as a terminology so you can it's all on the c. n. a. i don't
00:04:10
swiss website which is updated before christmas so for in german the other languages are coming
00:04:16
um to try to have a common language um and we try to stay in line with o.
00:04:23
e. c. d. definitions of course and we are in touch with people when negotiating internationally so it
00:04:29
but we don't do a wrong thing you know corner that would be a bad idea and the most such
00:04:34
a prominent thing i think in terms of democracy transparency and what role will give a few examples is this
00:04:41
list of a. i. projects that we maintain that database it's not mandatory nobody has
00:04:47
to announce projects to us but it was really meant as a to showcase and
00:04:52
on the one hand show citizens look that's what we starting to do and also within the administration to
00:04:59
did not have redundancy is the key people can work together so that's the other aspect of course
00:05:06
it's knowledge sharing a specialist sharing and um
00:05:11
to have a sort of a network of expertise
00:05:14
and they're uh this takes more time but it's a network so setup competence happens to that
00:05:21
we are facilitators but when they are specific questions
00:05:24
for example relating to legal questions around the use of
00:05:28
a a high within the administration not really you drafting a new laws but the use of existing regulation
00:05:35
how do they apply to but they i to being used
00:05:38
for example in an interview does questions got to illegal hot
00:05:43
but that the responsibility for that is that for example that uh
00:05:47
in the department of foreign affairs and uh with some preview states are
00:05:52
collaborating day as well and then we have a competence of algorithm
00:05:56
this is mainly my colleagues from the data signs competence centre and um
00:06:02
we're hoping to expand with uh_huh your collaborations with economic
00:06:06
sector of course so we and that's why uh that
00:06:11
that's i think that is important message especially in today's context is work
00:06:17
it's key values and especially related to that database but less the project very low tech and it's a p.
00:06:23
d. f. we get quite a few messages oh very
00:06:27
angry people saying that it's not machine readable and uh
00:06:31
uh yes please be my guess vote for budgets we can do it differently but um
00:06:37
it was just that and i think in the contest
00:06:41
last year was extremely intense in terms of a parliamentary various
00:06:46
questions in the parking problem in about a guy what
00:06:49
is the pendulum instruction doing how we gonna do with regulation
00:06:52
so it's so turn out that it was actually quite good to be able to show okay it's
00:06:56
not complete but we we which when we train together who does what and what is being done uh
00:07:04
and yeah i think can visits and that works so it's really supposed to
00:07:09
to hell with little means to try to achieve as much knowledge exchange as possible
00:07:15
try at least uh i think high don't need to convince
00:07:21
this audience that a i can help him in different situations
00:07:25
and why would the the administration use a i i think that's very
00:07:29
similar to any business when the organisation um
00:07:34
now in an administration this is specific focus when have new
00:07:38
insights you might wanna use it for decision making holy see policymaking
00:07:44
but of course it is always a dual more with less less budget more chores more
00:07:50
more more more more communication or maybe saws
00:07:55
and everybody needs to increase efficiency so that's really
00:08:00
but i think that's not a very surprising and of course especially the g. p. t. now
00:08:06
chad but we're gonna solve all our problems and the people will be able to concentrate
00:08:10
on the true test if we have a male bought a chat but something can help citizens
00:08:15
but then responsibility comes in how i i think one main difference though is that this
00:08:21
these questions was too personal you see them everyday we're gonna solve everything with a i but
00:08:27
it's more than just rhesus assessment for a public administration it's
00:08:32
it's more than allah not if you lose the trust of the citizen with a few
00:08:37
citizens uh you is extremely hard to to get it back and in the implications of
00:08:43
responsibility for the chat but that tells you it's gonna rain with age
00:08:47
percent likelihood and uh it doesn't okay i took my brother for nothing
00:08:52
if you have a chat but setup that will answer you in which uh administrative
00:08:58
it fits you gotta follow if you need a permit of
00:09:01
something and it's actually wrong was responsible and this is not even
00:09:05
but it was k. so i think this makes it particularly sensitive and i'm
00:09:11
of course we have to think about that i think it's extremely important and i think
00:09:15
it's a two important that this if with power keeping up my own of a a a
00:09:22
just very quickly the the i mean it's not complete this list then will be updated probably my pro uh
00:09:28
i'm about seventy projects so most ongoing not not all
00:09:33
uh if you have a look at it you'll see that
00:09:37
then if you are actually in production most of the and then there is that on this list are indeed
00:09:44
peas proof of concepts people trying things out because also
00:09:48
the lingo frame with not always ready but at least
00:09:52
could we use it would it be useful would actually make it saves time
00:09:57
because that's a question to you wanna save money saved time but sometimes you end up
00:10:01
seeing that it's not that easy you have to take what will drift into account will actually
00:10:06
did you have to compute power it's a trip that's more later then yes it down projects
00:10:10
from almost all departments if you familiar with that
00:10:14
seven uh it departments plus the chance to read
00:10:18
ongoing and there's a lot of research going on it unless use some not sure if you collaborate demonstrates but uh
00:10:25
so that contributes to this very high columns of projects from their side they really
00:10:30
do a lot of research on a. i. n. various implications not only warfare or um
00:10:38
just quickly yeah why do we use it new insights increasing efficiency
00:10:43
it's actually more less fifty fifty because of the say this is a lot of research going
00:10:47
on uh i must use that skews the graph more or less a little bit and uh
00:10:53
was the target audience with very often it's more likely to
00:10:58
work right the output of the system will be used by um
00:11:02
but collaborate is for something or in another process sometimes it's external professionals from a
00:11:08
certain uh the main uh and for example maps or uh you know with the
00:11:14
or other authorities pretty seen this is probably supposed to it but but in the room
00:11:21
and the general public here meetings use for example i'm sure examples just uh i
00:11:28
think you're very familiar with what a i can do would not sometimes it's like that
00:11:32
detection of binge drinking cartels these uh uh uh
00:11:35
this is actually production come cool it's an extension
00:11:39
of work that's been done until now but with better tools and also international collaboration in this case
00:11:46
uh this is a local example from our uh of
00:11:49
face a statistic that used to take twelve he used to
00:11:53
complete because if it is the the to fake on the
00:11:56
usage of um of the grammar right if it's agricultural naught
00:12:02
done with aerial photographs taken over every cool
00:12:05
tool it take its use and the um
00:12:09
the evaluation of the pictures used to be a hundred percent
00:12:12
manual between people so now part of it is done with a
00:12:16
deep neural net with system and a device to harvest the
00:12:20
time it was also good use case to show and then following
00:12:24
what we do and actually stolen i think i don't have to
00:12:27
tell you you know it an ultimate decision of classification of holland particles
00:12:34
an i. r. system this is the new but this is our definition
00:12:38
on this in the eye but it's like the o. e. c. d. uh
00:12:42
definition so it is it has an l. model and it is it has an effect on the environment
00:12:48
and they are specific challenges and i wanna finish on that for
00:12:52
a public administration the model itself can be a very simple integration
00:12:56
can be a bit more complicated maybe random force that can be
00:12:59
a true black box right did neural networks we don't really understand so
00:13:04
why are there specific challenges related to that's i think really
00:13:10
data we can be scraped from the net uh it and mister deeds beta but
00:13:15
they they can be biased what about the quality the security privacy uh no surprise that
00:13:23
was projects are in production a. i. r. based on our personal data non personal like that sorry
00:13:30
and intellectual property of course let's see uh and of course and the computing that's a bit
00:13:37
what about sovereignty where do the models rhyme because if
00:13:40
the uh compute hungry and we usually within the administration we
00:13:45
don't really have the resources to make a by do we um does the powers imagery and do
00:13:52
we have internal competence as well i think the the human factors will is really important you can have
00:13:57
fantastic tools but if the uh i collaborate just don't now what they're working with uh don't know
00:14:03
if that actually a lot to challenge the result of the automated system that was so expensive to build
00:14:09
uh well they really do it and there are many under pressure to do more with less
00:14:14
do yeah yeah when you walk two miles you know that you have it you have a
00:14:18
a little you choose your last function or maybe there is an implicit objective but still she
00:14:24
the result choice somewhere what's acceptable in terms of a curiously might be some percentages might
00:14:30
be okay for the main population but benny for some good boarded tolerable people it's not
00:14:37
uh and really the black box comes with a lot of problems in terms of legal terms as well
00:14:44
you had as a public administration do g. t. to
00:14:47
let the big window to to justify your decision so
00:14:51
how do you choose complexity versus explained ability trace
00:14:55
ability interpret ability these um mean important points and it's
00:14:59
less sensitive probably with the non personal data
00:15:03
uh such as ah pollen particles don't really mind
00:15:07
but for other things the these are extremely important questions and at the
00:15:10
end of the day such a system has an effect on the environment
00:15:14
on rule of law we live in a democracy and on fundamental rights
00:15:18
because how do you what effect it had on your twenty of someone you've
00:15:23
if a decision made the decision is not even automatic but the
00:15:28
mere fact that it was suggested by a machine this automation by its
Conference Program
Opening and introduction
Prof. Lonneke van der Plas, Group Leader at Idiap, Computation, Cognition & Language
Feb. 21, 2024 · 9 a.m.
101 views
