Player is loading...
Smartvote: the present and future of democracy-supporting tools
Dr. Daniel Schwarz, co-founder Smartvote and leader of Digital Democracy research group at IPST, Bern University of Applied Sciences (BFH)
Wednesday, Feb. 21, 2024 · 2:45 p.m. · 10m 21s
Online voting advice applications like Smartvote have become increasingly popular over the years. And the development will not stop. The advent of AI opens up new opportunities, but also creates pitfalls. Which aspects are particularly important for strengthening democracy and which paths will be problematic? The presentation will shed light on the topic using the example of the successful Swiss voting advice tool Smartvote.
Embed
Transcriptions
Note: this content has been automatically generated.
00:00:00
um thank you for the introduction and thank you for the invitation um i'm
00:00:05
here to speak about um democracy supporting
00:00:08
tools and uh my example is smart vote
00:00:13
and uh maybe the question by smart was still operates the same way as it
00:00:17
does what they'd twenty years ago and we are kind of a dinosaur in the field
00:00:23
um without the two thousand and three and many other apps of
00:00:27
course need yells of more famous apps and uh happened after us
00:00:33
uh and uh of course that uh that's the reason why a um we spill operate
00:00:38
with an quite a i free you know
00:00:40
way i'm i'm algorithms um the first um
00:00:47
just use the basic principle i'm smart what is um
00:00:52
voting advise application da as we call it uh
00:00:56
in our product designs line and v. a. a.s and
00:00:59
that's kind of a political arm off we commend the systems
00:01:03
we commend the seasons and you find it all over the internet uh from amazon to net
00:01:08
flicks and a google and so on so uh this this is them um is always the same
00:01:14
you've got the profile um of user um you've got
00:01:19
um some mm profiles there with market we've got to provide some candidates and parties
00:01:26
and uh then you try to match it and to a very cool
00:01:32
uh and you should voting for so that's and
00:01:35
the matching algorithm smugglers quite as successful um tool
00:01:44
both with candidates and with voters so we need both sides and
00:01:49
um and as you can see in the federal elections
00:01:54
um around eighty five percent of the candidates feel out our
00:02:00
questionnaire so we still have a traditional questionnaire that candidates have to fill out
00:02:06
and it's an eighty five percent and of these now six thousand candidates
00:02:12
and the this in two thousand twenty three on the vote just side
00:02:20
um if either quite successful um we had into dozens twenty three
00:02:26
and some thing about that two million boating advice is that too
00:02:33
a million times somebody cleats and to gets i'm voting
00:02:38
advice about recommendation was smart vote but the mm their
00:02:44
database to figures that we communicate the second one and that's actually the more important one that out that signal
00:02:50
uses and uh we know this from service that the sticks and it's
00:02:55
um as a from from representative uh studies um that
00:03:00
we have around five hundred thousand five hundred fifty thousand
00:03:04
um signal users bismarck about um which is roughly
00:03:10
twenty percent or big more than those who went to
00:03:14
the polls so where one fifth of the people
00:03:17
who really vote and used smart protein twenty twenty three
00:03:29
why is this the case um the success has a
00:03:33
reason and it's it's probably it has to do with
00:03:37
trust people trust the information and they get from smart
00:03:42
about and the candidates they trust the system that um
00:03:47
yeah that's the information um we get from them that we
00:03:52
use them um and reasonably so smart quotes um here's a um
00:04:01
a study uh from two dozen twenty three
00:04:04
um the average cost level um from voters
00:04:10
uh in and different institutions and information sources and
00:04:15
uh we see that mark wrote that the average trustees
00:04:18
higher down with t. v. and radio political parties and newspapers for example
00:04:24
and and that's quite striking with um really trust is in social media
00:04:31
and social media that's maybe most connected
00:04:34
to things like artificial intelligence and um
00:04:39
this is maybe also a a hint um what a i could
00:04:45
do with critical information sources the reason for the trust it gets
00:04:54
is as though i'm a declaration that
00:04:58
we follow in twenty thirty i guess if
00:05:03
and a couple of european researchers can to
00:05:06
last them and stay um formulated a declaration
00:05:12
um with the basic principles for b. a.s and there's a
00:05:19
general principle actually ace shockley open transparent impartial and likeable great sound and a
00:05:27
couple of um the client um principles
00:05:31
then like transparencies about operators funding calculation method
00:05:37
the third should be free to use for everybody that
00:05:41
there should be no m. exclusion of parties of ideological
00:05:45
reasons and so on so these are the principles that we follow um and
00:05:55
that's me you think that um the people they use model because they think
00:05:59
that's smarter this trustworthy and from both sides uh not only the boat just
00:06:04
but other candidates think this and of course they out the other reasons like
00:06:10
um user experience but it is free you of course to use
00:06:16
and said that it is quite purposeful so helpful
00:06:19
for voters to get information they need for their
00:06:23
boat and there's other downsides to its um when you want to use markets thing you need time
00:06:30
because you still have this questionnaire you have to fill out and you need some political
00:06:36
interesting political knowledge so without time and knowledge it's not possible
00:06:43
to use smart wrote now we are in discussions with uh the
00:06:54
researchers from other universities but also it internally in our group how
00:07:00
we could use artificial intelligence and in our case it's mostly um
00:07:06
mm natural language processing um how we can use it
00:07:12
for our system and and the standard proposal then is that
00:07:19
we could get rid of all these questionnaires and and detect the political profiles
00:07:27
by the information that's already there in the
00:07:29
internet from social media opposed to online news platforms
00:07:35
um maybe speeches in parliament other documents
00:07:38
in parliament um political parties and step programs
00:07:43
and so long so you have actually already everything there and you could just generates the
00:07:49
profile out of it and it sounds easy
00:07:56
but there's probably a um a problem that
00:08:03
um well i can start with the advantages of course it's less effort it's cost efficient
00:08:11
and it's also possible to get more
00:08:13
comprehensive profiles because the internet has more information
00:08:18
uh then just up questionnaires with fifty or
00:08:21
seventy questions and um there's also a potential that
00:08:26
caps can be filled so if i don't know uh something from a question at then maybe
00:08:33
somewhere in the internet's there's the information i need um but of course there is an egg
00:08:39
at this point which is quite huge um there is the perception does that a high algorithms
00:08:47
and that they are basically a black box
00:08:50
so it's not possible to um answer the question
00:08:56
why do i get this recommendation in the end
00:09:02
and and if you are not able to answer
00:09:06
this why question hands if um there's algorithm is
00:09:13
that does not explain is not possible to explain
00:09:16
why this um recommendation um you you get this
00:09:23
recommendation then you have a good reputation risk hands
00:09:27
potentially also this and loss of trust don't this is basically other the reason
00:09:34
why we still are in discussions with
00:09:38
um these groups researchers from uh universities
00:09:44
uh i'm not because we are not willing to implement anything but we have such
00:09:52
conditions and these conditions that are of
00:09:55
course um that's these and algorithms they must
00:10:01
be explainable the must be traceable um and we don't want to lose our reputation
