P2:
I can go next. 
I just started thinking of scenarios and just walking through a journey. 
So, I started with this one where where let's say someone reports a fraud and it goes through the system. 
And then of course the algorithm starts tweaking stuff because this has already been built. 
A key part for me here was explainability to a citizen of how these algorithms work. 
But also for the experts who make use of this information. 
Because a lot of the people who are also the controllers in a way of the decisions they also need to understand this in a sort of non-technical way. 
Can still be a little bit technical but not as much. 
So, what are those different parameters that we kind of consider for the algorithm but also how do we design a way for people to understand that. 
So, I thought about routing these rules towards the Amsterdam website where not only do you have all these parameters that go into the algorithm but you also start explaining how these get applied in a more human readable manner. 
For example, what is the implication of let's say the algorithm thinks about your source of income as one of the parameters that they look at people who don't have a consistent source of income and if they live in a good location how that kind of works together. 
And use that to kind of build personas as well and explain through that so this is how we kind of consider it. 
For example, anonymized persona of Joost, he lives at Herengracht he's not been working for five years has the house as an inheritance because they have that data has two rooms, has multiple garbage fines. 
And how all of that comes together multiple fines, no typical source of income, prime location as you mentioned as well. 
And how all of those go towards the algorithm kind of saying that this is a possibility. 
But also explaining to people that just because the algorithm says that it's a possibility doesn't mean we take it as that. 
But I think this part I think this is the biggest gap for me because nobody including people like you mentioned who are building these systems in their own organization understand how these rules work. 
So making that explainable that's step one for me. 
Then of course you already mentioned this as a part of human feedback loop to the machine learning algorithm. 
Now that I have all that I have that set of results that these are the possible list of frauds that you need to look into. 
There's a choice for the expert first so that's the first intervention of saying okay this looks fine doing a check mark or saying no something's off and then readjusting the results and also adding the feedback on why he thinks that this is incorrect or not. 
Because that's the qualitative aspect of the human being there. 
And that gets recorded into the algorithm as well. 
Who did it what was the decision making and that qualitative part goes into the algorithm as well. 
Let's say now they've gone to that person and they're like hey we got a report and our algorithms say we need to check that you're at whatever you're committing fraud. 
And they do decide that okay there is a chance of this person committing fraud and they levy a hefty fine. 
That this person has been renting for more than 30 days which is not allowed. 
This comes into your website I don't know the last thing these I'm not sure where it goes I have no idea. 
But once you get that charge explaining that charge itself is really important because right now in most websites when you are charged for something it's not, I can't understand what's the charge. 
What exactly is that charge how has it been levied. 
And then you should have the choice to contest it at that point itself before I start paying. 
Because usually I only get the chance of paying but not really contesting it and that's where you start...

P3:
Paying is made really easy. 
Contesting is made a lot harder.

P2:
So the idea is that you put this next to any action. 
This action is mostly for the system to put something into practice. 
But this one is for me as a citizen to practice my agency and my power to influence decisions. 
And then once you have that, I didn't finish this, but I was thinking about the contesting process itself and its ability to impact policy as well. 
But then I was first just thinking that if there is like a charge sheet can I contest different decisions. 
And some of them is so qualitative in nature for example let's say the neighbor itself is a nuisance. 
I don't think that's true here I don't know. 
But we used to have neighbors where they can also have a bias. 
So sometimes it's not the system but people who generally have it. 
So can I contest as in the charge sheet that okay this I want to contest because I think the neighbor has an issue this is the evidence I can provide. 
Another could be that yeah that your income source you've been putting a lot of income but you don't have like a job so how are you getting it. 
Then contesting that okay but my family has been supporting me I have evidence to show that things like that. 
So that what that contesting process could look like but then that also made me start it. 
I also started thinking that okay there is no process or policy around it. 
Like do you have a limited time for you to do that? 
And then I started thinking about this one. 
Then this is tricky because of...

R1:
Elections.

P2:
Yeah but like also in terms of... 
Yeah, if you want to implement this process as well, do you do you have a time limit to the contestation or not. 
Because some some governments do that that there's a time period within which if you can solve it great if not you pay. 
All of that becomes part of the policy and system development. 
So even if I as an organization as a public agency want to put this into process... 
Yeah I got stuck here which made me think that okay when you want to implement the system you need to start thinking about the implication of policy. 
And how it impacts this process as well. 
And then the bias part that you kind of built in like how to show it at an aggregated level I didn't get to that. 
But I really liked it. 
Because I think that's also important like how do you start showing that this is a repeated pattern happening over a period of time. 
Either because of the kind of profile the person has which is basically that they don't work I don't know full time or something yeah so then how do we remove that as a factor.