# Processing with AI
## Exploration : IA & Ethics

Name:
> Yaël VENTURA
>
Subject:
> Detect student presence in class using Face Recognition

>[TOC]

## Design brief

### Biais

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

>1. Some students might not be recognized and, therefore, might be considered as missing.
>2. If the student that is not recognised because he/she is from a minority, or has a special color of hair (ginger as example), etc. :
>2.1. Other student might make fun of him/her and he/she can experience this as racism, or at least as exclusion.
>2.2. The students that are detected by the program instead of the others might feel favoured and start believing that "this is normal" to be treated better according to the color of their skin, a common color of hair, etc.
>3. Adolescent might have acne which is common at their age. If the people from data base are older and the student's acne is strong, the program might not recognize him/her.

We will ensure that our model is not biaised by:

>1. Including every color of skin/hair/etc. in our data base.
>2. Adapting the data base considering the pourcentage of students of each color of skin/hair/etc. of the group.
>3. Include students with common skin problem (like acne) considering their age.
>4. Using the biggest data base as we can and train use different dataset like the example of L'Oréal, taking into account the biggest differences that appears between the students of a specific group.

### Overfitting

If we modify the data base based on the pourcentage of differences **specific to a group of students**, it won't be adapted to an other one.

But not doing this, and trying to train the model according to national figures, will lack of accuracy since it can change considerably from a school to an other one.

We believe that, since we deal with children, the ethics should be the top priority. Children can learn very fast something deeply false, we can't take the risk to create confusion and exclusion.

Therefore, we will **train the model differently for each students' group** taking into account the pourcentage of color of skin, color of hair, or other strong differences into the group.

### Misuse

We have to remind ourselves that our application could be misused by a public authority, using this to make **national ethic statistics** which is forbidden in France.

It can also be misused by private entities to do **target marketing and orient their production** based on statistics, which is also forbidden in France.

### Data leakage

We have decided that **our training dataset will be shared between every school of the region** as long as parents gave their **authorizations**. This will help us to get the dataset as precise as possible.

If the parents don't accept, we will keep their child's profil unshared.

### 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 **student pretend** to be at school when they are not.

Therefore, **professors should pay attention** to their classroom everyday and be aware of a student missing. Technology should help them, but can not replace their role of vigilance.