# Processing with AI
## Exploration : IA & Ethics

Name:
> RABAH Laïla
>
Subject:
> Detect student presence in class using Face Recognition

>[TOC]

## Design brief

### Bias

If we don't source our dataset with enough rigor, the following bias might appear:

>1. a The machine will confuse students who may look alike (people from the same family for instance such as brothers, sisters or cousins.)
>
>2. b Some school students from different skin colors will may not be detected in class by the Face recognition
>
>3. c The machine will not recognize some students when they have different haircuts, accessories such as glasses, piercings, bandannas...)

We will ensure that our model is not biased by:

>1. Sourcing our data from a large library of pictures. This library will be composed of every pictures taken by the students when they apply to the school. These pictures needs to be updated when there is a major change (haircut, surgery. It will be interesting to update students pictures (every 3 months).
>
>2. Making sure our datas will be double-cheked by independent organization.
>2. Making sure our data take into account minorities and ethnic diversity. We need to train our model to better recognize people from minorities, from various skin colors to better perform.
>
>3. Using a diverse team to create our model (with womens, mens, people with various ages, ethnies, religions, sexual orientation...)

### Overfitting

We will make sure our model does not overfit by
> Checking the accuracy of our model on various types of schools, various cities and countries.

### Misuse

>We have to remind ourselves that our application could be misused by police or the defense authorities to track the behaviours of some students. Indeed, some students that already have a criminal record could be then tracked by these institutions in schools.

>We have to remind ourselves that our application could be misused by government in some autoritharian countries to strictly control student behaviours. The government would use this model to listen to students opinions (political opinions for instance)and instaure an authoritarian regim.

>We have to remind that this model could be used to predict the probability of a student to become a criminal. Indeed, some researchers or politicians could try to use these statistics to state that if a student don't go to school, he will be bound to become a criminal.
>
>The model could be used to determin if a school is attented by large percentage of minorities or by few minorities and some people will chose to go or not to go on a school based on theses datas.

### Data leakage
*Choose the most relevant proposition:*

>In a catastrophic scenario, where all of our training dataset were stolen or recovered from our model, the risk would be to have companies that will uses this data to promote some products whithout the consent of these students or adapt their marketing strategy by spying on students and listening to their tastes and opinions.

**OR**

>We have decided that our training dataset will be fully open-sourced, but before we made sure that we have find solutions for the biais and overfitting problems.

### Hacking

> If someone found a way to "cheat" our model and make it make any prediction that it want instead of the real one, the risk would be that some students could make the machine consider them as present in class whereas they were absent and it could be problematic if they have sholarships or if their parents think there are at school whereas they are not.
>
> Also, we can imagine that a student will try to take revenge on of his classmate for a personal mater and make the machine not detect its presence in class. The student will be dismissed by his school because of attendance datas extracted from the model.
> We can also imagine that some students have family problems and that some of their mother or father are a threat to them. If the datas are hacked, the person that represent a threat to this child could now where he is stuydig and try to kidnapp him or worse harm him.