Суд ва ҳуқуқни муҳофаза қилувчи органлар фаолиятида рақамлаштириш
61
Марко Глобелник
Йожеф Стефан илмий-тадқиқот институтининг
сунъий интеллект бўлими раҳбари (Словения)
СУНЪИЙ ИНТЕЛЛЕКТДАН ФОЙДАЛАНИШ БИЛАН БОҒЛИҚ
МАСАЛАЛАРНИ ХАЛҚАРО ҲУҚУҚИЙ ТАРТИБГА СОЛИШ
Марко Глобелник
Руководитель отдела искусственного интеллекта
Исследовательского института Йожефа Стефана (Словения)
МЕЖДУНАРОДНО-ПРАВОВОЕ РЕГУЛИРОВАНИЕ ВОПРОСОВ,
СВЯЗАННЫХ С ПРИМЕНЕНИЕМ ИСКУССТВЕННОГО
ИНТЕЛЛЕКТА
Marko Grobelnik
Co-head of the Department for
Artificial Intelligence at Jožef Stefan Institute, PhD. (Slovenia)
INTERNATIONAL REGULATION OF ISSUES RELATED TO THE
APPLICATION OF ARTIFICIAL INTELLIGENCE
Uzbekistan could get involved a little bit more in international
organizations, in standardization of AI. AI has become popular last 10 years.
3-4 years ago international organizations started regulating the area of AI,
because there are not just positive sides of AI, but also negative sides. The issue is
to prevent the abuse of AI.
There is a set of 9 international organizations which are all active in
regulating standardization of AI. Thus, we are involved in this highlighted 5, by
we meaning our group. As I see Uzbekistan is present in UNESCO but not in others.
So this something Slovenia can help with, at least in terms of connecting you to
the right people so that you maybe get status of being observer.
Why is it important? Because it would give you international access to the
network and you could actively participate with your own views on AI regulation.
Thus, just to list some of these organizations, one is Council of Europe. Council of
Europe is an organization is Strasburg. It is not related to EU, it’s separate and a
little bit HR mostly. And yesterday we started new committee in AI which leads
towards to legally binding document treaty. At least countries which will sign this
treaty will have to obey. This is very important.
The second one is European commission, our continent EU is preparing the
most important document which will be released in one year from now roughly,
which is called EU AI Act. And this will be legally binding as well. This will be most
Суд ва ҳуқуқни муҳофаза қилувчи органлар фаолиятида рақамлаштириш
62
likely the first legally binding document on AI in the world. Now we already have
2 to 3-year discussion what it should be. Once we have legally binding document,
basically we need to obey it and it’s always on balance between the risks, on the
one side the risks of technology and on the other side we have risks not to prevent
innovations, products, companies and so on. So AI act is something you should be
careful about.
The third organization is FRA. This is connected to European Commission’s
Fundamental Rights Agency. And next organization is called GPAI, Global
Partnership on AI, which is at the moment has 18 countries. It started with France
and Canada but 1
st
founder countries are also Slovenia, the US, Germany and
couple of other countries. So this is maybe the place where you could apply and
get member. IDB is an American Development Bank maybe not relevant. The next
one is OECD. OECD is the most active organization, because OECD is technically a
think tank. It’s a big think tank with almost 1000 people which does economic
analysis from all sides. They establish this AI committee 4 years ago and I was a
member from the day one on and so OECD is leading the way in many ways. The
UN has also digital cooperation committee which has AI as a part of it but it’s a
little bit slow. Then UNESCO, you are a member of UNESCO. Last year UNESCO
adopted this AI ethics document which is also quite influential. And the last one is
the World Bank Group. Just to conclude here, we are happy to have Uzbekistan to
get involved at least as observer state with some of these organizations but you
need to be a little pre-active either as experts or policy makers. I think at
universities you have enough technical experts which could also participate here
or on the other side policy makers meaning people from government ministries
who could participate there as well. So this is quick overview of the organizations.
Now few things what we do. We have a list of important documents which have
been coming out of these organizations. Maybe one thing which could be
interesting for you, in OECD in 2018 we have already have started to work on
defining AI. Everydiv talks about AI, but actually in legal terms AI wasn’t very
clear. This is not an official definition of AI. What we see here is cat. This is just the
opposite of what AI is. Usually in AI machines, computers mimicking living beings
but here we see the cat which is mimicking the machine. This is just the opposite
of AI. The actual definition of AI is these 3 sentences which we were working with
8 months. At the end there was a vote with presences. 44 countries voted for this
definition. Later this definition was adopted by G20 and also with European
Commission with slight modifications. Every word here has meaning. Maybe
important element is human-defined objectives. We are not defining AI but an AI
system, which makes it more tangible. This is just the graphical representation of
this definition of which has lots of policy implications. On top of this definition we
can attach many other legal elements. This definition has lots of positive
consequences.
Then there is the OECD AI Policy Observatory. AI is very fast developing
area and exchanging every few month plenty of new things which one needs to be
aware of it. So the old fashion way of observing something like AI is just too slow.
That’s why we purpose to the OECD to make something which makes observing
Суд ва ҳуқуқни муҳофаза қилувчи органлар фаолиятида рақамлаштириш
63
AI in real time, as it happens. So this is how the OECD AI Policy Observatory
appeared. These are the addresses http://oecd.ai/ and http://oecd.ai/en/trends-
and-data.
This is visualizations which we are showing not to go too much into details.
Here we can see bubbles. Each bubble is one country. The more to the right, more
scientific production this country has in AI. If you start near to 2000, the lower
bubble China is still behind. But last 20 years China really made huge progress, as
you see it’s moving fast. Otherwise, it’s the US, the EU and China, these 3 are the
AI superpowers. We have plenty more of these visualizations inside.
AI is a global area. And as we see before it’s the US, China and EU. But if we
analyze who works with who, where the development actually happens, we see
in the world, globally two big bubbles, two big islands. On the left side we see the
island which is basically Chinese development, which is fairly separated from the
rest of the world. From the right side it’s European and the US and on the bottom
also Australian collaboration. And there are also a few connections really back to
the Chinese development. China is also really developing fast and developing
internally as well as you see from this picture.
It is another important view for policy makers. AI at the end it’s really battle
for the talent and battle for smart people. If you don’t have people, you can’t do
AI, physically this is the message. Now one of the things which we have calculated
is who is gaining AI talents and who is losing AI talents. Who is losing smart people
and who is getting, employing smart people. This is list of countries, the bottom
are the countries which are losing talent like Venezuela, Tunisia and couple
others. And the top is Luxemburg, Arab Emirates and some of the EU countries
and Singapore, Canada and so on. So this is important to be observed also
Uzbekistan. You should be careful not to lose people. You can’t recover from it. If
you start losing people, then you need some number of years to recover.
The last slide is about who is investing money into AI. At the end this is one
summary slide on venture capital. So it’s basically the US and China. Already EU
isn’t investing comparably well. Therefore, mostly the US and China. For China the
graph is too low. It’s because the service where we get the data we can’t cover
everything that is happening within China.
One of the messages here is 3 superpowers which has dynamics here. Next
superpower which is appearing is India. You can expect by all these hidden
indicators which are not hidden anymore for us, India most likely in 5-10 will
become one of these big bubbles on the right as well. They have such push inside
of science, technology, also seems like many recently that it’s inevitable they want
to become a big player.
Artificial Intelligence in Digital Governance: the “country digital twin”
paradigm.
Some new technologies are used in Slovenia and in not many other
places. One of these is digital twin of a country.
How do AI technology to digital government bring to itself? In many ways
AI is a set of tools which are solving sometimes complex but simple narrow
programs like face recognition or things like this. But governments or country
Суд ва ҳуқуқни муҳофаза қилувчи органлар фаолиятида рақамлаштириш
64
level things are way more complicated than just face recognition. And this area
how to use AI to monitor or regulate and observe or how to use AI for proper
decision making on the level of the country, this is the message. Thus, the
technology is digital twin technology. The main message here is to make digital
copy of the country from the data which is already available and let’s use this for
proper inform decision making. This slide is relevant for whoever looking for
answers. Unfortunately, too many decision makers go to the left way, for simple
answers. Not necessarily to right side. This is what digital twins will try to do:
going to the right side.
Let me tell you on this slide a story which we, Slovenia but not only we but
other countries also had when COVID 19 started 2 years ago. The speed of
decision making increased, governments had to decide on majors fast on daily
basis or weekly basis not in a month. One month was huge amount of time. So the
speed of decision making increased but the systems were simply not prepared for
this speed up.
Next thing, we had to adopt to new situations. Nodiv knew COVID 19.
It was a new situation, adaptation became total necessity, but government
procedures of all countries were too slow and weren’t adopted. Therefore,
governments maybe good in solving isolated issues but COVID again brought
something which we had to understand consequences of our decision making. So
this is another observation. What is interesting is that data which basically we
would need to make proper decision making was available and each country
today has enough data to understand what’s happening in society. It is just the
problem that this data isn’t integrated properly or not accessible for legal reason or
technical reason. I know that data exists, but nodiv is using it. The last point is that
even more funny that technology is totally available but not used in critical spots.
These are 5 observations. Now what is the answer? How to respond to these
5 observations. Let’s make this digital twin technology, to solve these problems
the answer will be this digital twin of the country.
One important element when we talk about digital twin is to have proper
situational awareness. Think aware of situation today, right now, as it happens,
not in a month time or next year or at the end of the year. So this situational
awareness cycle. It has several elements: it has perception, comprehension and
projection. These are 3 key steps which we need make. Thus, perception meaning:
let’s get the data, comprehension: let’s to standard data and projection would
mean let’s use this data to form decisions or recommendations for decision
making. Perception usually can be today or tomorrow. This isn’t a big problem
anymore. Maybe it’s a legal problem but it’s not a technical problem. The problem
is a comprehension and projection. So this where AI can help. Let me show you
now approximately what we will try to do with digital twin. You can see small
metrics. On upper line we can see 7 types of questions. We need to answers to
these 7 types of questions: What? (What is happening?), When? (time), Where?
(geography), Who? (who is doing it?), Whose? (where is the origin? What
Суд ва ҳуқуқни муҳофаза қилувчи органлар фаолиятида рақамлаштириш
65
supports the data to get the trust?). The last 2 questions very important for
decision making: Why things happen? (Causality, how things are connected) and
What if? (simulate possible decisions/outcomes).
What we need to do in vertical side? There is history, present and future.
These 7 principals can be done for historical data, for current data and also with
AI we can do some forms of predictions as well.
The next slide is another type of the summary. What kind of questions
should we ask from digital twin.
1.
Understanding a history of last years or last 10 years
How were events in the past interconnected? How was the history
structured? Learn from the history since it’s very important to learn from the
history.
2.
If something happens we would like to understand the reasons for
this. So this is what we call Root-cause analysis. So where is the root-cause of some
events which happened.
3.
We see how things are happened and we would like also to predict
what will be the next. We cannot say 100% always but with high probability we
can often say what will be next. So given current situation what’s happening right
now, can we say what’s happening next days, next weeks or next month.
4.
As I mentioned before, we can simulate possible future scenarios by
what-is questions. So if we take the decision what are likely consequences in the
future.
5.
Sometimes society or country is just to complicate to us all possible
questions. And, the AI system also can suggest what are the questions could be
asked. Sometimes we don’t know all questions, and system can help detecting
what are relevant either anomalies or surprising phenomena such complex
system as running the country.
Couple of pictures from January 21, 2020. We were in European
Commission; this was the time when there were zero cases of COVID in Europe.
There was a meeting of state secondary and so I prepared this speech so Covid
was appearing. Covid was mostly in China far away from here. So I prepared it in
the afternoon few slides and it took me 15mins to show this situational
unawareness in that particular moment.
This are the questions which I was showing before: What? Who? (who
means organizations, people are involved) Sentiment? (this is what sentiment of
people) Places which are relevant, when things were happening. In upper left
corner, in which languages it was reported, which media was covering Covid and
what topics were covered. This was an example. In January 21, 2 years ago we
already collected articles over 10000. But still you see, the EU didn’t basically
react and was waiting till it was too late as so many other counties
The next question was where things were happening. It was already seen
form our platform. This was Covid related events reported in media already, very
complete. Each dot which you see here was a set of either publication information
that something is happening mostly it was about China.
Суд ва ҳуқуқни муҳофаза қилувчи органлар фаолиятида рақамлаштириш
66
The next thing is which things are related to Covid and how they were
connected so this is one diagram. I can show this slide but it is a little bit
complicated.
And then what is the story like. This was big news back then “Deadly
coronavirus outbreak DID start from the animal market in Wuhan”. Everydiv
were reading this article and it was interesting
As an answer “Country Digital Twin” covered in last 3 slides.
How did we approach country digital twin in Slovenia at start of Covid. Now
I’ll be showing slides from end of January 2020. First the discussion in Slovenia
happened in March. Although, everything was happening already before. So, we
were called from the ministry and the question was: How can you help with AI?
And we purposed to use digital twin technology at least to establish situational
awareness across many areas which are shown in the next slide. Everything was
fine except the legal problems appeared like access to data. Also, fears, this crazy
AI will appear and will know more than we were supposed to know which was
not true. This fears from different government offices were indefinite. That why
the building of project was also a little bit slowed down. But within less than
2 weeks we built working protocol which was monitoring in real time couple of
aspects of society: economy, transports, media and well covered. But the problem
was with health the data which actually existed in real time. But for legal reasons
we had issues to get some of other data.
If we talk roughly about digital twin of a country what kind of data one will
need to have. They don’t to have everything but this would be an optimal case.
Every country has statistical office, and the data is available, it is not a big
problem. Health data, especially if we have health crisis it is necessary from
hospital, national health office or equivalent organizations. This something that
usually all countries have, we are tracking it. It is just the speed of collecting
information is a problem.
Then there is information from pharmacies and general stores, what people
started buying. If you monitor pharmacies, you can spot mental health, drugs or
substance are being bought or people start buying things which are unusual. This
like a signal before official signals can be spotted.
Economic data from transactions either from banks or depends on financial
administrations which I’m sure you have. You can see pretty much how
economies being shut down or how it’s going back in real time. So we were able
to do this in minutes not in hours. In minutes we saw how different parts of
economy were reacting to different measures.
It’s strategically very important to monitor supply-chain data. Supply-
chains are getting disrupted so you can expect problems. Transport data is one of
important one. We can track cars, trains and so on. So this is usually not a problem
to get. Energy data just exists you just maybe not interlinked. Media is not a big
problem with a proper system. Either social media or main stream media, this is
important indicator, especially social media for emotional state of the nation.
Despite low quality of this data, social media is pretty much real time indicator of
emotional state of the nation.
Суд ва ҳуқуқни муҳофаза қилувчи органлар фаолиятида рақамлаштириш
67
And informal societal information would also be organizations, NGOs and
so on which collect different data source. Connecting all these data sources into
one single system you can imagine that it helps with the questions that I stated
before. You can actually understand what is happening within the country
extremely efficiently. You don’t need to do any kind of privacy intrusion or so on.
So this is on the level of the country. So you see the trans in the society you are
not infringing any individual person.
In the conclusion, the message here was the real-time modelling of data
sources which most likely you already have. And using technology like digital twin
can help improving decision making in complex systems like governments. AI
today is mature enough. This digital twin technology is used for factories,
machines and this kind of systems but not for governments. So we were one of the
first ones on the level of the country.
Since Covid was a big global crisis which had bad sides. We didn’t ask for it,
but since it happened, we need to use it. The point is that don’t waste good crisis,
use crisis also to do something good. So this was the one of message we tried to
provide.
