Okay, so thank you very much for staying for the last talk of the day and last talk of
the year for the collaboration day room.
So thanks a lot.
I see people that stayed a lot during the day.
So thank you.
So the last presentation is about Matheus, which will be the challenges of creating
a fast fact-checking platform for the Brazilian community.
Thank you.
Thank you.
Yeah, can you hear me okay?
Yeah.
And thanks, appreciate everyone here.
I thought there would be less people, so I really appreciate that.
It's my first time at FOSDEM as well.
Let me introduce myself.
I am a senior product manager at the Wikimedia Foundation working on MediaWiki.
But this product is a different hat that I wear.
I'm a volunteer on an NGO in Brazil and we are trying to combat fake news and make something
that is open and in software, in data and in knowledge.
So what I want to talk about here is the mission of this project and I'm going to go over the
challenges of trying to do something against misinformation in Brazil.
Brazil is like a very fertile soil for misinformation, information, especially recently.
The reason for that to arise in my will was basically because seeing how my family was
sharing and spreading misinformation, it all started.
So I started imagining, so I'm a co-founder, I'm not the only one that volunteer on this
project, but we started imagining a society where everyone can freely access and engage
with true and reliable information about autonomy.
And I kind of stole a little bit of the Wikimedia Foundation mission here because we share
similar values.
So the mission for this project is to encourage ad-edu communication.
I'm not going to talk about it because the other co-founder, which is a journalist, is
the one that kind of coined or uses that term.
And basically it means that we can only achieve our mission if we actually educate people
and we need to communicate that.
So the platform and everything, all the product what entails, it all focuses on the specific
pilot.
But the idea is that for the values on accessibility, credibility and autonomy, we're creating individuals
or we're making autonomous individuals in Brazil being able to access information or
at least question information without losing the credibility.
And so when we started, so this is a study from Caspersky.
And so more than 70% of Brazilians with internet have believed in the fake news and 62% of Brazilians
failed to recognize false news.
So this was a study that at the time was what motivated us to keep going.
And challenges were immense because it forced us to tweak planning and change and pivot
a lot in the years of the foundational years, which is kind of the timeline, which I call
the foundational years.
So I'm even excluding here all of the technical exploration that I did since 2018.
And in 2020 when I thought that, okay, we have technology and maybe we're ready to proceed,
we signed up for the Mozilla Open Lab.
We participated in receiving product mentorship and preparing an ideation which came to be
Alete Effect.
And from that exploration with Mozilla it was very interesting because we decided, so
we came as dreamers like we want to do something multilingual for everyone.
We want to fix the world because the logo of that program that we participated was fixing
the internet.
And then we learned that it wouldn't be like that.
So we would have a focus group.
We would only focus in Brazil.
We would look into people that is already engaging with fact checking or at least a
reading about fact checking somehow.
So active readers or independent fact checkers, even professional fact checkers.
And then we would even look into a specific demographics, which would be from the age
of 18 to 29, would represent like 16% of Brazil.
And we set a goal that if we get 0.1% of that, 35,000 potential independent fact checkers,
which is an increase of 7,000% of professional fact checkers in Brazil.
And I'm going to talk about this a little bit because, so we did this exploration on
Mozilla Open Lab.
Then we started working on the infrastructure and launched the infrastructure and tried
to experiment more.
We participated on the trust, it's the TTL, it's the trust and truth online, yeah, conference
and so we introduced the concept of the democratization of fact checking in Brazil, which is something
that we are looking forward.
And in the same year, we started a residency at Projeto Comprova, which is a group of
news outlets and independent organizations that combat fake news in Brazil.
And by participating on that, we actually engaged with our personas or professional
fact checkers.
And from that point, we started exploring a platform focused on professional fact checkers
only.
So we wanted to do something to speed up their process, make sure that they were, the process
were optimized and they could actually chase fake news and combat it.
So with that in mind, we thought that we are ready to formalize and then we started understanding
what that would entail.
So with these learnings, we defined that with process transparency and the didactic
of representation, we could enable a fact checking manual or operational guideline and
then we could replicate that and create autonomy of the individuals that we wanted.
So the methodology should be accessible and understandable and that's the requirement
that forced us to be aligned with the Creative Commons license for our data and also everything
that we create from courses or workshops are all open and available.
And we also had some, in the last year, I'll also talk about that, but we had multiple
workshops and partnerships with universities in Brazil and all of that was free and based
on the knowledge sharing proposed.
And the platform or the product that I mainly worked on would be just a facilitator or a
place where people could use and engage with.
And our main goal was that, so we were here in 2021, we wanted to reach the Brazilian
elections with something that could be used for good.
So then we formalized and then we participated on the Global Fact Nine conference, which
is organized by the IFCN, the International Fact Checking Network.
And when we got there, we went to validate some of the use cases that we have built
and it was like a bucket of ice on us because we understood that there is a different dynamics
happening on fact checking community.
So one of the things is they are very worried of being open, mostly because that can be
weaponized by bad actors and create more problems to them.
So every software that they write, it's not always open, sirs.
It's not always open.
Licenses, because they are mostly tied to news outlets, they don't always follow the
Creative Commons license.
So everything that we have built to create a shared space, like a public digital space
for people, was not going to work with professional fact checkers.
So we understood that, we pivoted, but we kept the same model, the same values and we
went forward and launched a platform with a few people, like 15 volunteers.
So in the platform, we created a process where we would listen to the debates during
the Brazilian elections and do live fact checking.
And it was good.
So this is just a screenshot of some of the parts of the functionality.
So here, if you have something highlighted, it means that it was fact-checked by someone
and the experiment was very good.
Very small as well, because if you look into our views, you're going to see that we only
have like a thousand views.
But the impressions from the people and what we were able to achieve was a good data to
proceed forward.
But because this is a very small project, it's a very small organization, we were dreaming
big for a presidential election, trying to do an impact and of course, the stretch goes
there, but what we learned is there is a use case, we can do that, but we need to begin
small.
So we took a step back and looked into what we can really do with our resourcing.
And from that, we decided that we would have three product pilers, the fact checker productivity,
reduce to credible information and reduce obstacles to product.
So talking about the last one first, because we were very small, it doesn't work like
our reference, which is Wikipedia, it doesn't work like that.
You cannot just go anonymously and do a fact-checking, you cannot go and create an account and start
doing it because we don't have a governance model or we didn't have at the time.
We didn't have a governance model, we had a lot of obstacles, so we put obstacles on
purpose so we could only test with a few people and understand what we need to do.
And then we need to now remove those obstacles because now we are becoming more confident
that now this can be used by everyone and we are going to have a procedure, a code of
conduct and a governance model that will help.
Access to credible information is the one that we are not focused too much right now
because we believe that from the model that we have plus the productivity, we are going
to create the credible information, but access to it is a little bit different.
So we need to be able to make sure that the audiences that we serve are actually being
able to access this and they can access the platform, we also provide goods SEO using
the claim review schema and to be able to be searchable without people having to go
to the platform. However, in Brazil we have a lot of people that, so there is an inequality
or disparity of resources in Brazil that includes internet access in a sense that people can
have data plans that can only work with WhatsApp but they cannot access internet. They can access
Instagram but they cannot access Wikipedia. So the access level is a total different
gain that we are still studying how we are going to approach, but because these are not
our focus we should not lose sight as well, so that's why it's a pilot. And the productivity
was because of an opportunity that we had because in 2023 we started a partnership with
the University of the State of São Paulo which I actually came from and did my bachelor
there and we had four interns and this is the other co-founder, our SEO, I like to call
her my president and these four students were our focus group, we had a group of students
in a sense that they are very fresh like first year journalists they know nothing about life
more or less, maybe, I don't know, maybe they can learn from Tik Tok now but the idea is
that they should have no knowledge at all and be just willing to work on. And we had
a very smooth process and it was very refreshing because they finished four months of internship
having the same level of productivity than news outlets when we were delivering fact-checking
material. So of course different levels of comparison but in any way as I mentioned the
platform was launched a while ago this 2022 is when we did the experiments with the debates
and then we see that at least from the platform we are increasing the engagement of the functionality
so we still have pretty much the same unique visitors but now people are using the platform
more because now we actually have productivity for the team. So these are just a few screenshots
of the platform, the code is in GitHub and you can access the link in the end of the
presentation. And so this is an example of a fact-checking report that is available after
the fact and as I mentioned so we focus a lot on the productivity of the fact-checkers
so we started putting in place tools that would be tied to a specific workflow which
is flexible. The way technically speaking we are using state machines here to control
everything and we adjusted the processes and adding different steps depending on how we
learned with the team. And the idea is to have a visibility of the productivity and
also be able to collect data and see is this actually improving because once we, if we
actually go, if we actually reach the goal that I mentioned before like 35,000 people
checking facts in Brazil it's going to be a very different thing to administrate and
make sure it runs smoothly. So yeah, this is I think that the learns. A few things that
I think I would like to mention from the experiment, it was in Brazil the open source community
is very spread, not so much well organized so the whole period of creating the software,
trying to test the software only captured six volunteers like actively working on it
which it looks like well this is actually pretty good but I forced most of my friends
to actually go there and hey come on you know how to QA things, can you help me QA this,
you are a DevOps engineer, can you create a pipeline to me so this was kind of a best
effort from a community but after all of this what happened was that a lot of entry-level
engineers were starting to looking for something to work on and they were, because we had
partnerships with universities we started having people just coming in and they were
trying to learn with the software which was a very good experience and something that
I would like to explore moving forward it requires more management on the technical
side being able to actually provide good feedback to them but it has, we now have like two or
three active volunteers that they are 20 years old and just learn how to code and now they
actually provide good development for the platform of course there is a skill to consider
but so yeah there was this challenge so the time frame when I look to this project like
five years is a lot for the stage that we are at but there were multiple factors there
like it's a product for a very specific area it's a problem that no one solves yet and
the only way that I see moving forward is doubling down the educational effort for the
actual goal which is being able to provide credible information and stop the spreading
of misinformation there is no other way around there is no other software there is no AI
there is nothing that can help other than having humans understanding what they are reading
and being able to have something that is accessible so I also put here multi-generational
because in Brazil the disparity on misinformation spread is based on age as well so the older
you are the bigger the chance to be a victim of misinformation and what we are also going
to be looking at the general TV AI so and I write here with all the truth in my heart
that I have no clue what we are going to do about it and the reason that I put it here
is because it's not that we need to use the general TV AI but we need to defend against
it which is I think very different perspective to look at of course maybe we need to put
it here and make pieces with the devil who is the same tools to combat at the same level
but the concerns are different because we are already seeing on multiple elections around
the world the usage of deep fakes and general TV AI to manipulate public speech and so this
is going to be a very difficult thing to do however because we are losing the battle
we need to consider and that's it the code is open source it works specifically for an
audience in Brazil it has been tailor suited for that but because since the beginning we
were concerned about having this for multiple audiences so it allows internationalization
the code is the reason that so the stack that we chose and one of the things so the stack
we chose was no JS and React because the ability to find more people to join the effort but
that of course we are now considering if it makes sense to keep the same stack or rewrite
some of the stuff because if we want to be lean and optimize for some use cases we might
also consider performance and other things that right now the platform doesn't provide
and I forgot to mention something very important everything that we do is integrated with Wicked
data right now and we have efforts to integrate with the whole public data infrastructure
the idea is that we only have what it needs to be done for fact checking but information
about personality more information on Wikipedia all of that should be included and integrated
with the ecosystem and in the end we encourage other people to build their own communities
and be part of the movement fork and change toss it off test the same things that we did
I think that because it's something very very important and it's going to change a lot
in the next few years I believe that we should double down the effort as much as we can that's
it it was supposed to be a gift but it's a PDF so thank you thank you very much for
the attention thank you Matheus any question?
Hello so I was looking at the website and I see the personalities and the declarations
and I see the reviewers but then how do you define or who puts the new declarations in
for the fact checkers to really check it?
Yeah so one of the things that we learned from the procedure of the fact checking is
that monitoring has a specific operational guideline so one example that a fact checker
might find to is that they receive this information but they look into how is this happening on
Twitter X how is this being spread into Facebook how is this being spread on WhatsApp is this
a big effort does it make sense to to actually check because there is one thing which is
called the strategic silence because if someone checks it makes more public and it spreads
more so the the decision on what to put in there is for the volunteers that have been
have the capability to operate the platform so right now in order to operate the platform
you need to go over the training understand how our code of conduct works sign that you
understand and you are going to vouch for it and once you understand the whole process
then you are able to operate the platform and you will be you will be responsible for monitoring
so these volunteers are the ones that also select what is putting there but we receive
we receive suggestions from community in Brazil from people that follow us and we use that
into consideration to decide on if we are going to put on the platform or not and because
it's a small group it's going to be small data now but the idea is to streamline this
process and grow and possibly monitoring will be we will need to evolve based on on on
that does that make sense thank you it's the last one from the UK we have a few fact checking
organizations there do you connect or are you intending to connect with other similar
organizations from the world yeah so we did when I talked about being part of the global
fact nine which is part of the the international fact checking network we learned a lot and
met a lot of the the I think full fact is one the biggest one and one that we get in
touch we got in touch and now we are in the process of being part in part of this network
there's a lot of criteria it's if we do it we are going to be the first open project
that actually enters so we are having some trouble to actually match on some of the criteria
that they ask but we can we have some connections with India we have some connections with the
Latin America network as well and recently we joined the the the network that is only
on Brazil so every out Nils out in Brazil are also part of this network and we are connecting
with them as well yeah we need that okay thank you other questions no
okay thank you very much and well that was the collaboration they've room for 224 so
thank you for staying until the end