Transcriptions
Note: this content has been automatically generated.
00:00:01
now actually a i and privacy and we'll discuss that a beat
00:00:06
a short presentation about me maybe the only important aspect
00:00:10
is that i have a low degree love background the not
00:00:13
the technical one uh which would impact the bubble i'm walking
00:00:17
in which is not the same as most of you are
00:00:20
um i'm also a member of the she's data protection organisation i'm just mentioning it
00:00:25
like because the the name will come back again during the presentation does g.
00:00:29
q. and that is an important player in the the privacy the uh regulation
00:00:37
and i i decided to do a relate to the helicopter
00:00:42
a view of this issue and present it might might might resent
00:00:46
on the uh this topic so have three topics out i want to discuss and and
00:00:51
um i hope my presentation will be a sufficiently brief so that you uh here e.
00:00:57
i hear it into the n. um people jump right to the first one is that
00:01:02
privacy is not to call problem of uh artificial
00:01:06
intelligence and and maybe for you it's something entirely logical
00:01:10
and again that might be a question of bubble because in
00:01:14
the ecosystem or yeah i work when we see artificially intelligent the
00:01:19
first things that come to our mind is why how it impacts
00:01:23
privacy all the problems that i have with data protection but actually
00:01:28
he here we can seen these images that private information is just a a
00:01:32
small topic and the all the that i think are much more important that
00:01:36
a privacy so maybe you're gonna say that you know i'm i'm up i'm
00:01:40
forcing a an open door and that's all very logical that it probably is
00:01:45
but the question that it raises for me is
00:01:48
who should be responsible for supervising or managing artificial intelligence
00:01:53
and when i look at again my bubbled maybe different in your view
00:01:59
but from my perspective in companies the persons that are being in charge
00:02:04
of deciding wait uh you know artificial intelligence can be use or not
00:02:09
is the data protection officer or the legal officer
00:02:12
or some one that highest is regular battery back around
00:02:16
re actually there is a lot of all the questions that
00:02:19
are very important article once that has been discussed just before
00:02:23
and that maybe i d. p. o. is not the best a person to answer
00:02:27
and same goes with the technical issues maybe you would say well should be you know
00:02:32
uh and take up the text or detective about that should be responsible
00:02:37
but generally not the way it goes uh uh you don't ask you know trader
00:02:42
and to define the rules for a trading a commodities insecurities
00:02:47
but the question is who should be responsible in an organisation that the same
00:02:52
question and we have i think regarding government agency we
00:02:55
should be responsible to supervise the use of artificial intelligence
00:03:00
and again if we see at all it has been handled so far
00:03:04
and europe that has been clearly a task of the data protection supervisory authority
00:03:11
especially one in italian we come back to that i to be different in the
00:03:15
u. s. for instance is the activity at that has a a and say something about
00:03:22
what should a artificial intelligence do and how we
00:03:26
can market and our products that use artificial intelligence
00:03:32
and here is a news from uh the can hit the commute is the fringe
00:03:37
and that uh production supervisory authority have from twenty
00:03:42
a tool and they basically stating that the camille
00:03:46
would be yeah the authority that is responsible for um
00:03:51
managing the new regulation that you just
00:03:54
heard about in europe regarding an artificial intelligence
00:03:59
and for me it really raises a question is it's the good all authority to do that should we not create special
00:04:05
authority to do so uh like and uh spain has done
00:04:08
now and and of course the risk is that if you
00:04:12
choose someone whose primary view is privacy then they will only focus on that and maybe
00:04:18
uh not too much and they would not focus enough on the other aspect so if there are
00:04:23
and policymakers in this room i urge you to consider
00:04:27
this aspect because i think that it's an important one
00:04:31
so going aspect and privacy related risk of a
00:04:35
i i'm not new and that just more intense
00:04:38
and here's an image that came out a just a few weeks after a charger pity was really is
00:04:44
and i think it's the perfect view of you know what happened
00:04:47
to the re on my bubble again of privacy just after the events
00:04:52
you see macro seven open i a i just very happy the showing charge it it's a
00:04:58
such a pity trying away i. t. privacy and trade secrets
00:05:01
goal which uh didn't had released bother that time just looking
00:05:06
and a bit stupid lady i rage uh data protection authority turning its back on the problem
00:05:12
and very few authorities and specially the guarantee which is italian super easier to reach it
00:05:18
and uh being very afraid and and making and some moves directly
00:05:25
and and if we really want to see how a privacy the interacts with
00:05:30
and artificial intelligence we have to see the the big picture again
00:05:34
and i would have the a helicopter view or maybe yeah we over simplify technical problems and it's a
00:05:41
good thing does is why lawyers do the over
00:05:43
simplify take problems and they over complex vital ago once
00:05:48
and so with at least two the first job i will not do the second one
00:05:51
but if we look at how an artificial
00:05:55
intelligence models interacts with data i see three ways
00:05:59
we have here the training that that we have spoken quite a lot about the training that than i would
00:06:03
speak again about that and then we have the include that uh so it's the user of the mother that will
00:06:09
include the that uh to uh the uh to carry something to get efficient divisions and then
00:06:15
the the model we're regular generate some that's uh that's the out about that
00:06:20
so if we more directly to the first one it puts
00:06:23
that are and we try to consider whether the the privacy impact
00:06:29
um the first question that we have is are we actually processing any person but also it can be
00:06:34
it but often it won't be and in that case that one be a person no that's a protection usual
00:06:40
which also raise the question well should really did look new you know dealing with that
00:06:45
and open questions and if we do process person of that because we will often do
00:06:50
then we have very very traditional that the production issues i'm not saying that they're simple to answer
00:06:55
but actually very simple so if you use you know the tools
00:06:58
from the and uh the two people start speak uh spoke this morning
00:07:05
and then uh it's basically managing them so you want to have your band there that is trustworthy
00:07:11
you want to ensure that that ties not transfer over the border specifically to the u. s.
00:07:16
without some safeguards in place you want to know all the data that you would be
00:07:21
it would be we use beats by chad's b. t. o. by the tools that we heard
00:07:25
and was being processed days on consent for further training the model
00:07:30
and it was this question about collecting mater that that's information about
00:07:33
the use of the series and it's complex but it's known it's
00:07:38
what we have done as though you're in that uh protection for
00:07:40
years and it's not something that is very specific to a artificial intelligence
00:07:47
yeah but i wanted to introduce one maybe different topic we're getting into that that does anyone knows what it is
00:07:55
i i yeah exactly it's called the rewind pendant
00:08:02
and it's basically an a. i. device that will record all your conversations
00:08:09
and with the view of making you know a permanent record a record of anything that you've talked about
00:08:15
and the uh n. w. you know to go back and say okay uh i forgot what my wife that's example that you
00:08:21
uh ice media out for the dresser is so i can come back to that or
00:08:25
maybe i want to hear again the first well the from a new one baby yeah
00:08:30
yeah i've got it raises a lot of questions you
00:08:32
know there was this uh commands on the uh on x.
00:08:37
uh about very creepy device it really looks like um
00:08:42
blacken your eye it's actually there was one it is that whether it was the exactly the same thing
00:08:48
and so apparently it will be uh i it will exist there is
00:08:51
already a price tag attached to it so it's it's not yet there but
00:08:54
it will be here and of course here great many new questions regarding
00:08:59
h. put that up because you would interpret that that it is not yours
00:09:02
and that race extra questions regarding consent authorisation uh and
00:09:07
they are currently in studying son pervasive preserving their thoughts
00:09:11
uh with voice fingerprints where basically we have today i consent to being recorded
00:09:16
and they will register a a fingerprint of your voice and if you don't then it will not record you a voice
00:09:24
so that's it for the uh input that i now the training that that's something
00:09:27
we already talked about uh this morning and and here again that's nothing very new
00:09:34
so basically the only difference is that the new scale so i you know
00:09:38
that's available up that that's about with new not new but and the more questions
00:09:44
and there was this question that was actually just ask before can i
00:09:47
scrap that are from the internet and it is a very interesting question indeed
00:09:52
and the answer is quite expensive and because it depends whether or not
00:09:58
you nor your product or your company will be firing d. us they've and
00:10:03
for one case decided that the algorithm that where i
00:10:07
created based on data that was collected uh with all
00:10:10
the local bases and would need to be raised in
00:10:14
saudi that to wait which are a case in the us
00:10:17
very interesting because we don't have this sort of penalties in europe ah switzerland and regarding the legal bases well then
00:10:24
uh can i spread in that the internet the answer is i i it
00:10:28
depends where they're you know who you are what you do and who you ask
00:10:32
so if you are the google search engine there was actually a case in a
00:10:38
two dozen fought in from the highest um the court of justice in the new route
00:10:44
and they basically said that you can they can scrape the intended to
00:10:48
uh created a um a search engine based on their legitimate interested only the content that they do so
00:10:55
then there was another case regarding clear view you know the company that
00:11:00
two pictures from and every faces a available on the web and then created
00:11:05
a a fashion recognition software then uh the can you remember the clear fake late
00:11:10
in or so they find a clear view twenty million euros in a twenty twenty two
00:11:16
and then if you ask the get and there's so the other uh uh de
00:11:19
uh uh italian suit every zero to retire get h. i. g. p. t. a.
00:11:23
yeah we have an interesting case 'cause i don't know if you remember
00:11:27
but when judge judy was released the italian guy and they say that now
00:11:33
you cannot uh i'm i'm not sure that you can comply with the g. d. p. or so i order you to stop
00:11:39
i'm using the services in italy and if you would go to if you
00:11:43
would be any today you could not accept access for something like a month
00:11:48
h. i. g. p. t. at all and one of the question that was raised was that there was a lack of legal
00:11:53
basis and it was a lot of discussion in my small bubble about that get because there was a really a question what will
00:12:00
and the guarantee do after the thirty days because this
00:12:03
was really clear that's tragic beauty would not changing thirty days
00:12:09
and after that said today is it well basically the uh
00:12:12
the guarantees that okay all think liaise again really open doors
00:12:16
uh you have not provided you know uh uh make an is for you to complain
00:12:21
and the uh charge would be to uh say or twenty i have said
00:12:25
that they would make a fourth correcting nectar it data although they cannot do that
00:12:30
uh and that's sufficient but no well uh or there are you know view
00:12:33
on whether or not you can describe the internet and then it based on you
00:12:37
uh illegitimate interest it's more i would say oh confirmation that you can do how much
00:12:43
time we have left okay and mark and you mentioned that you can do if you
00:12:49
ask me then there is the question about the secondary use of personal data so it's
00:12:54
links to scrapping that up the internet but it's also if you have your private we got
00:13:00
and you want to uh uh and reuse the data for uh
00:13:04
training y'all i'm a i'm a does then the question
00:13:07
is do i need to get consents and and here in
00:13:11
switzerland at least we have these general authorisation to process
00:13:14
press another data or what we call non person other purposes
00:13:18
so as long as your processing that up for me
00:13:21
and that is not to identify handling to specific individual
00:13:25
then you can reuse it without the need for consents it is the data protection view their specific
00:13:32
wills i stated here in the human research act i guess we're gonna have a uh some more information
00:13:38
on that just a afterwards or or will also return the uh
00:13:42
not to go on that topic that you can access a on
00:13:45
the um and then very quickly on us but that ah so
00:13:51
uh what ever you know come out comes out of the yeah mothers
00:13:56
and he i think we really have reached a we have privacy risks quite a lot
00:14:01
and a lot of other issues that you know your already have discussed about now but it's
00:14:07
where we can have i think the the mall rates from a personal uh point of view and
00:14:13
uh of course the the the the entry ponds for privacy laws is that
00:14:17
the the needs to be personal that that in all in the output data
00:14:21
and then you can have inaccurate result about the problem which i g. t. t. and here um the problem is that
00:14:28
it's one of the key principle of data protection laws that you have to ensure the accuracy of you press no that that
00:14:34
and it's often not do able but apparently no one cares and then you
00:14:38
can have two accurate results about individuals on the it's just questions about the capacities
00:14:44
of fashion recognition or the type of uh what cognition capabilities um and it's
00:14:51
a real real program and then there is all these you know uh did fakes
00:14:55
uh issues and it's i think a very interesting question is it your personal that down
00:15:00
uh if you want to go go there is a um um the case with that with uh get that you know the digit player
00:15:07
that actually recorded the song go with the voice of i
00:15:11
mean an but not with i mean i mean self uh um
00:15:15
and the uh created that song so it it is the voice of i mean and it is the style of i mean them
00:15:21
but it's not the song for many minimum and the question is well is it's the privacy right the
00:15:26
feminine or not because it's it's not really him that's polk and i think this was very very interesting questions
00:15:32
and the last topic that's relevant is all the topics about automated decisions
00:15:37
so the risks that you know computers will take decisions that impact those without a human
00:15:43
in the middle um i'm not sure it's actually a privacy issue but it's an issue that
00:15:48
has been button privacy laws so we have to do a without and and clearly it's a
00:15:53
it's an open a problem uh last binds
00:15:58
a solution it's exist but flexibility is needed and
00:16:02
when i say flexibility is more than actually you should be really not
00:16:07
to be compliant and i've had a few time you know deeper compliance
00:16:11
this morning i don't think that's something that can be achieved generally and
00:16:14
i really don't think that it can be achieved in the a. i.
00:16:18
walls and so there are steps that you can do to reduce your
00:16:21
risks have first you could avoid working with personal that whenever it's possible
00:16:27
there are techniques twenty mice the data and it do what
00:16:30
what it was but it's still a a good thing to do
00:16:34
then you can add up to all these privacy by design and by the full
00:16:37
techniques it's a requirement of the new figure that the production go in six in hand
00:16:43
and so it all the techniques that we have spoke about this morning to
00:16:46
reduce the identity ability et cetera et cetera and then there is does neither
00:16:52
and it's important in the new data protection act to conduct data protection
00:16:56
impact assessments so it's a systematic review of the privacy and security risk
00:17:01
of your processing activity and it will be required in most a. i.
00:17:05
project so you definitely need to do that and that will help you already
00:17:09
and take certain actions uh and mitigation so and last
00:17:13
but not least you will have to accept that whatever you
00:17:16
will do will not come from entirely with the regulations and